0% found this document useful (0 votes)

104 views202 pages

SAR Processing Training

Uploaded by

Dimitry BEKALE

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

104 views202 pages

SAR Processing Training

Uploaded by

Dimitry BEKALE

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 202

Processing SAR data with

Geomatica Banff
Version 2.7
Geomatica Banff
© 2019 PCI Geomatics Enterprises, Inc ®. All rights reserved.
COPYRIGHT NOTICE
Software copyrighted © by:
PCI Geomatics Enterprises Inc., 90 Allstate Parkway, Suite 501, Markham, Ontario, L3R 6H3, CANADA
Telephone number: (905) 764-0614
RESTRICTED RIGHTS
Canadian Government
Use, duplication, or disclosure by the Government is subject to restrictions as set forth in DSS9400-18 “General Conditions —
Short Form — Licensed Software”.
U.S. Government
Use, duplication, or disclosure by the Government is subject to restrictions set forth in subparagraph (b)(3) of the Rights in
Technical Data and Computer Software clause of DFARS 252.227-7013 or subparagraph (c)(1) and (2) of the Commercial
Computer Software-Restricted Rights clause at 48 CFR 52.227-19 as amended, or any successor regulations thereto.
PCI, PCI Geomatics, PCI and design (logo), Geomatica, Committed to Image-Centric Excellence, GeoGateway, FLY!,
OrthoEngine, RADARSOFT, EASI/PACE, ImageWorks, GCPWorks, PCI Author, PCI Visual Modeler, and SPANS are
registered trademarks of PCI Geomatics Enterprises, Inc.
All other trademarks and registered trademarks are the property of their respective owners.

PCI Geomatics Page iii

Publication history

Version Released Revisions Revised by

2.3.1 2015 08 13 Updated for 2015 G. Gosselin
2.4 2016 04 04 Updated for 2016 G. Gosselin
2.5 2018-01-23 Updated for 2017-SP4 G. Gosselin
J. Flatt
2.6 2018-06-29 Updated for 2018 R. Downey
2.7 N/A Updated for 2019 (Banff) G. Williams
G. Gosselin

PCI Geomatics Page v

Contents
Introduction 9
Course overview 9
About this guide 9
Radar-processing workflows 10

Module 1: SAR data structure and support in Geomatica 16

Lesson 1.1: Supported SAR data format in Geomatica 16
Lesson 1.2: Data structure and matrix type 19
Lesson 1.3: Complex data support in Geomatica 25
Lesson 1.4: Ingesting and extracting a calibrated backscatter image 29
Lesson 1.5: Conversion utilities 40

Module 2: Applying radiometric enhancements 45

About this module 45
Lesson 2.1: Applying SAR-speckle filters to detected data 49
Lesson 2.2: Applying the PSBOXCAR filter to fully polarimetric data 55
Lesson 2.3: Applying the PSPOLFIL filter to fully polarimetric data 59

Module 3: Polarimetric decompositions 63

About this module 63
Lesson 3.1: Introduction to radar polarimetry 63
Lesson 3.2: Freeman-Durden decomposition 77
Lesson 3.3: Cloude-Pottier decomposition 83
Lesson 3.4: Touzi decomposition 88

Module 4: Using the SAR Polarimetry Target Analysis tool 99

About this module 99
Lesson 4.1: Introduction to SPTA and the target-selection mode 99
Lesson 4.2: Coherent target analysis in SPTA 106
Lesson 4.3: Incoherent target analysis in SPTA 113

Module 5: Classifying SAR data 121

About this module 121
Lesson 5.1: Classification of polarimetric data, possible workflows 121
Lesson 5.2: Unsupervised Wishart classifications 123
Lesson 5.3: Supervised Wishart classification (with targets defined in SPTA) 128

Module 6: Change detection using polarimetric data 135

About this module 135
Lesson 6.1: Examples of possible workflows for change detection 135
Lesson 6.2: Performing an intensity-change detection 141
Lesson 6.3: Wishart change detection 145
Lesson 6.4: Performing a phase-change detection 149
Lesson 6.5: Incoherent change-detection workflow 153

PCI Geomatics Page vii

Contents Processing SAR data with Geomatica Banff

Appendix A 167
Processing with single-pol or dual-pol detected data 167
Applying ratio or difference to detected channels using EASI Modeling 167
Applying an intensity-change detection on detected data 171
Computing SAR texture measures 175

Appendix B 183
Polarimetric discriminators 183
Generating polarimetric discriminators from coherency matrix eigenvalues 184
Generating polarimetric discriminators from analysis of the Poincaré Sphere 189
Synthesizing a backscatter image for arbitrary transmit and receive
polarizations 194
Maximizing the contrast between two targets 199

Page viii PCI Geomatics

Introduction

Course overview
Welcome to the SAR processing with Geomatica training course.
The course is designed for experienced users of geospatial software and introduces
you to the radar analysis tools available in Geomatica.
This guide contains six modules. The lessons in each module are designed for tasks
you are likely to perform in your analysis of radar imagery. They provide instruction
for using the software to carry out key processes while sampling key Geomatica
applications and features.

About this guide

Each module contains a series of hands-on lessons in which you work with the
software and sample data set. Lessons have brief introductions followed by tasks
and procedures of numbered steps. The lessons provide you with an opportunity to
work with several Geomatica programs, such as Focus, OrthoEngine, and SPTA
(SAR Polarimetry Target Analysis), to carry out various tasks. The goal is to
become familiar with the tools available for analyzing synthetic-aperture radar
(SAR) data, and to develop your expertise with Geomatica.
The scope of this guide is confined to the core Geomatica programs. Some radar
remote-sensing concepts are reviewed in the modules and lessons.
The modules in this course provide instruction on activities associated with
processing SAR data. This course includes the following modules:
• Module 1: SAR data structure and support in Geomatica
• Module 2: Applying radiometric enhancements
• Module 3: Polarimetric decompositions
• Module 4: Using the SAR Polarimetry Target Analysis tool
• Module 5: Classifying SAR data
• Module 6: Change detection using polarimetric data
The companion data you will use in this course can be found in the radar data sets
you downloaded from the PCI website.

PCI Geomatics Page 9

Introduction Processing SAR data with Geomatica Banff

Radar-processing workflows
Geomatica includes several programs you can use to work with radar data.

Table 1. Key Geomatica programs for working with SAR data

Toolbar
button Name Description
Focus You use Focus to display radar data. If information about the
geolocation of the data is available, it is projected and
resampled on the fly (Module 1). Metadata is also imported,
which you can view (Module 1). You can resample, reproject,
and crop your data, and create histograms for entire channels
or for a specific area of interest (AOI).
Note: Topics not specific to radar data are covered in the
Geomatica I and Geomatica II training courses.
Focus features Algorithm Librarian, a collection of
algorithms you use to apply custom processing of radar data,
such as importing raw data, speckle filtering, polarimetric
decompositions and classification, change detection, and
more (Table 2). Each algorithm is presented in an easy-to-
use graphical user interface (GUI).
Focus includes a Python Scripting environment in which you
can create, edit, and run Python scripts.
SPTA With SPTA, which stands for SAR Polarimetry Target Analysis,
you can explore your SAR data. You can select targets in a
polarimetric SAR scene, draw a target (or load an existing
target), extract polarimetric parameters (from the image),
and display the results numerically and graphically. While you
can use SPTA to analyze all types of SAR data products, your
data must be fully polarimetric and in complex format to
exploit the full functionality of the program.
You can use SPTA with Focus to develop applications. For
example, in Module 4 and Module 5 an example of a land-
cover classification is presented.
EASI EASI is a command-line interface that provides you with
access to all the algorithms available in the Focus Algorithm
Librarian. With the EASI programming language, you can
automate processes and run processes in batches.
OrthoEngine You can use OrthoEngine to correct geometric distortions and
geocode SAR data based on radar-satellite modeling for
orthorectification or on other math models, such as the
polynomial or the thin-plate spline models. OrthoEngine
features powerful capabilities to collect ground control points
(GCP) either automatically or manually.
OrthoEngine also features support for complex data. This
means you can either geocode your data first, and then
process it, or vice versa.

Page 10 PCI Geomatics

Processing SAR data with Geomatica Banff Introduction

Depending on needs and objectives, the programs that comprise Geomatica can be
used alone or in combination. For example, in Figure 1, a SAR scene characterized
by its header file and image data is first loaded and displayed in SPTA for data
exploration.
Based on selected areas of interest (AOI), several statistics related to the scattering
intensities or polarimetric parameters can be extracted simultaneously. After
finding the most relevant algorithms or parameters, the entire scene can be
analyzed in Focus by selecting the corresponding algorithm from Algorithm
Librarian or in EASI (Table 2).
Finally, the image data and the results extracted from it can be georeferenced in
OrthoEngine before distribution. Examples of workflows using the various
components of Geomatica are demonstrated in Module 4.3, Module 5.1, Module
5.3, Module 6.1, and Module 6.5.

Figure 1. Workflow diagram

PCI Geomatics Page 11

Introduction Processing SAR data with Geomatica Banff

SAR-processing algorithms
The following table identifies the most relevant algorithms available in Geomatica to
process SAR data.

Table 2. Algorithms for processing SAR data

Process Algorithm Description

SAR data SARINGEST Imports a SAR data set to a PCIDSK file
ingestion and (Module 1)
calibration CDSARA Reads SAR imagery from CD (LGSOWG format)
CDSATA Reads satellite imagery from CD
CDASARA Reads ASAR imagery from CD
CDALOSA Reads ALOS imagery from CD
SARINCDA Generates calibrated radar backscatter
SARBETAA Generates radar brightness
SARSIGMA Generates calibrated radar backscatter
SARSIM2 SAR image simulation
A - Algorithms superseded by SARINGEST in most situations

Speckle filtering PSBOXCAR POLSAR Boxcar filter (Module 2)

add noise PSPOLFIL POLSAR Lee adaptive filter (Module 2)
reduction
FEFROST Enhanced Frost adaptive filter
FELEE Enhanced Lee adaptive filter
FFROST Frost adaptive filter (Module 2)
FGAMMA Gamma filter
FKUAN Kuan filter
FLAP Laplacian filter
FSPEC SAR-speckle filters
FSTDDEV Standard deviation filter
FTOUZI Touzi filter
FME Median filter
FMO Mode filter
Polarimetric PSEABA Calculate entropy, alpha, beta, and anisotropy
decompositions (Module 3)
and scattering PSFREDUR Freeman-Durden classification (Module 3)
characterization
PSG4U2 General four-component decomposition
PSPHDW POLSAR PHDW decomposition
PSKROG Krogager decomposition
PSTOUZIDEC POLSAR Touzi decomposition (Module 3)
PSTOUZIDIS POLSAR Touzi discriminator
PSPOLDIS POLSAR discriminators (Appendix B)
PSSSCM Symmetric scattering characterization
(Module 4)
PSVANZYL Unsupervised van Zyl classification

Page 12 PCI Geomatics

Processing SAR data with Geomatica Banff Introduction

Process Algorithm Description

PSWHITE Whitening filter target detection (Module 4)
Change CCDINTEN Intensity change detection (Module 6)
detection CCDPHASE Phase change detection (Module 6)
CCDWISH Wishart change detection (Module 6)
CHDETB SAR image change detection (Appendix A)
B - Superseded by CCDINTEN

Texture analysis SARTEX SAR texture analysis (Appendix A)

HISTEX Histogram-based texture measure
TEX Texture analysis (Appendix A)
SAR-specific PSSWIS Supervised Wishart classification (Module 5)
classification PSUSWIS Unsupervised Wishart classification (Module 5)
SARCLASS SAR classification using regions
SARSEG SAR segmentation into regions
Conversions and PSIQINTERP Convert complex data to other interpretations
utilities (Module 1)
PSCONV Convert matrix representation (Module 1)
PSINANG Incidence angle map
PSFARA POLSAR Faraday rotation angle
Compact PSCOMPACT POLSAR compact synthesis
polarimetry PSCC Correlation coefficient
PSCONF POLSAR conformity coefficient image
PSRECONS Create a pseudo fully polarimetric covariance
matrix
PSCOMDIS Calculate compact polarimetric discriminators
PSCOMDEC Apply decompositions to compact polarimetric
SAR imagery
Interferometry DEMADJUST DEM adjustment to elevation points
INSADJUST Adjust interferometric phase
INSCALDEFO Adjust unwrapped deformation
INSCOREG Automated coregistration
INSINFO Generate report of interferometric parameters
INSRAW Generate a raw interferogram
INSTOPO Flat earth and topographic correction
INSUNWRAP Unwrap interferometric phase
Channel PSPOLSYN POLSAR synthesis (Appendix B)
synthesis PSPOLSYNC POLSAR synthesis using maximum contrast
(Appendix B)
PSPOLSYNR POLSAR synthesis using targets
PSTOTPOW POLSAR total power
PSPEDHT POLSAR pedestal height

PCI Geomatics Page 13

Introduction Processing SAR data with Geomatica Banff

Process Algorithm Description

PSINTRAT Intensity ratio
PSS2C POLSAR synthesis of user-defined pair of
coherent channels
PSPHDIFF Phase difference (Module 3)
Geometric AUTOSHIFT Shift SAR image position
correction ORTHOC,D Generate orthorectified image
AUTOGCPC,D Enhanced automatic GCP collection
AUTOTIEC,D Automatic tie-point collection
FFTMVECC SAR image simulation
FFTMPOLYC Extracts GCPs with image-to-polygon matching
C- Also available in Geomatica OrthoEngine
D - Supports complex-valued data

Page 14 PCI Geomatics

Module 1: SAR data structure and
support in Geomatica

Lesson 1.1: Supported SAR sensors and data formats

Several SAR image formats are supported by the PCI generic-database technology,
known as GDB. With GDB, programs can directly access data using the vendor’s
format, in a uniform manner, without having to reformat it before using the data in
one or more programs. GDB includes access to imagery, vectors, attributes, ground
control points (GCP), and other auxiliary information of interest to users of
geomatics software. The Geomatica radar suite supports many SAR sensors1,
including the following:
• Radarsat-1, Radarsat-2, Radarsat Constellation Mission (RCM)
• Sentinel-1A, Sentinel-1B
• COSMO-SkyMed
• PAZ
• TerraSAR-X/TanDEM-X
• Kompsat-5
• UAVSAR
• ALOS-1 PALSAR, ALOS-2 PALSAR
• ERS-1, ERS-2
• Envisat-ASAR
• Risat
Data acquired by these sensors is diverse and often available in different acquisition
modes, data formats, and processing levels, excluding different resolution and
beam modes. This variety can categorize the data in different ways, and among
them, the Geomatica radar suite uses two basic categories:
• Channel type
• Transmit/receive configuration
These categories define which algorithms you can use to process your SAR data.

1 To verify whether a sensor is supported by GDB, browse the GDB file formats in the
Technical Reference section of the Geomatica Help.

PCI Geomatics Page 16

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

The channel type is divided in two subcategories:

• Complex data
• Detected data
Complex data, also known as single-look complex (SLC) data, corresponds to the
first processing level after the signal transformation into an image. Each pixel of a
given channel form corresponds to a complex number composed of an imaginary
(Q) and a real (I) part. Complex (coherent) arithmetic is therefore possible
between channels of a single image or between channels of various images, which
leads to applications like radar polarimetry, compact polarimetry, Interferometry
(InSAR), and Polarimetric Interferometry (POLinSAR). Module 3 provides more
information about radar polarimetry.
Detected data corresponds generally to the next processing level in which the real
and imaginary parts are combined. Detected data form layers (images), which are
easier to interpret for the human eye, and where the only available information for
any given pixel is the backscattered power. Therefore, a surface or object that
reflects most of the incident signal toward the sensors will appear bright on the
image while a surface that reflects most of the incident signal away from the sensor
will appear dark.

Table 3. Channel types

Channel type Subtype Example

Complex channel Single-look For a complex number a + bi, where a
complex (SLC) corresponds to the real part (I) and b to the
or imaginary part (Q).
Multilook complex ex.1: I= 0.2068 and Q = 0.0338, the
(MLC) complex number corresponds to
0.2068 - 0.0338i.
Detected channel Intensity, Intensity corresponds to (I^2)+(Q^2).
amplitude, decibel ex.1: (0.2068^2)+(–0.0338^2)
= 0.0439
Note that intensity is also known as power.
Amplitude, corresponds to the square root of
the intensity (I^2 + Q^2)^0.5:
ex.1: (0.0439)^0.5 = 0.2095
Decibel: corresponds to 10* log10 (intensity)
or, equivalently, to 20*log10 (amplitude)
ex.1: 10*log10 (0.0439)= -13.576 dB
ex.2: 20*log10 (0.2095)= -13.576 dB
Decibels can be converted to intensity or
amplitude:
ex.3: 10^(-13.576/10) = 0.0439
ex.4: 10^(-13.576/20) = 0.2095

The transmit/receive configuration refers to the polarization or polarizations

used for transmitting ( Tx ) and receiving ( Rx ) the signal and to the number or
channels forming a single image.

PCI Geomatics Page 17

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Table 4. Configuration types

Configuration Channel type Details

Single polarization Complex or ⃗ 𝑥 : one polarization, 𝑅⃗𝑥: one polarization
𝑇
detected
ex.1: ⃗𝑇𝑥 (H), 𝑅⃗𝑥(H) → HH

ex.2: ⃗𝑇𝑥 (V), 𝑅⃗𝑥(V) → VV

Dual-polarization Detected Tx: one polarization, Rx: two polarizations
⃗ 𝑥 (H), 𝑅⃗𝑥(H)&(V) → HH + HV
ex.1: 𝑇
⃗ 𝑥 (V), 𝑅⃗𝑥(H)&(V) → VH + VV
ex.2: 𝑇
Compact polarization Complex ⃗ 𝑥 : one polarization, 𝑅⃗𝑥: two polarizations +
𝑇
phase (Φ) preservation between channels
ex.1 CTLR (circular transmit with linear
receive): 𝑇 ⃗ 𝑥 (R), 𝑅⃗𝑥 (H)&(V)
→ RH + RV + ΦRH-RV
ex.2 DCP (Dual Circular Pol):
𝑇⃗ 𝑥 (R), 𝑅⃗𝑥 (R)&(L)
→ RR + RL + ΦRR-RL
ex.3 PI/4 (linear transmitter at 45 degrees
with linear horizontal and vertical receivers):
⃗ 𝑥 (45°), 𝑅⃗𝑥 (H)&(V)
𝑇
→ 45°H + 45°V + Φ45°H-45°V
Fully polarimetric Complex ⃗ 𝑥 : two polarizations, 𝑅⃗𝑥: two polarizations
𝑇
(quad-pol) + phase (Φ) preservation between channels.

ex. 𝑇
⃗ 𝑥 (H)&(V), 𝑅⃗𝑥 (H)&(V)
→ HH, HV, VH, VV + ΦHH-HV, ΦHH-VH,
ΦHH-VV, ΦHV-VH, ΦVV-HV, ΦVV-VH

Module 3 and Appendix B provide more information on the concept of polarization.

Page 18 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Lesson 1.2: Data structure and matrix type

In this lesson, you will:
• Display a RADAR (SAR) image in Geomatica Focus
• Explore the SAR data structure and georeferencing
• Learn about channel and matrix types

Data structure
With Focus, you can work with data in a variety of formats by using GDB and the
PCIDSK file format. This means that most RADAR data is supported in its original
distribution format and can be opened in Focus or SPTA using the key file name.
• Sensors [key file name]
• Radarsat-2, TerraSAR-X [*.xml]
• UAVSAR, [*.ann]
• Cosmos-SkyMed, Kompsat-5 [*.h5]
• ALOS-1 PALSAR, ALOS-2 PALSAR [LED-*, summary.txt]

To open a file in its original distribution format using its key file name
1. In Focus, click the File menu, and then click Open.
2. In the File Selector window, open the folder that contains the file you
want, select the key file, and then click Open.
3. If data calibration is supported for the sensor used, at the prompt, select
a calibration type, and then click OK.
4. If supported, select a projection, and then click OK.

For each sensor, depending on the product and acquisition type, different image
layers and auxiliary segments are available and can be imported.

PCI Geomatics Page 19

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Figure 2. Radarsat-2, Single Look Complex (SLC), Fine quad-polarization, beam 29

(FQ29), May 7, 2010.

Figure 3. Sentinel-1A, Ground-range detected (GRD), Stripmap, (SM) dual-

polarization, January 1, 2015

Page 20 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Figure 4. UAVSAR, Calibrated multi-look complex (MLC), Polarimetric mode,

August 5, 2009.
Today, most sensors provide information describing the geolocation of the acquired
SAR data obtained by GPS receivers on board the satellite. In its simplest form, the
geolocation information corresponds to the geographic coordinates of each corner.
Some sensors also provide rational polynomial coefficients (RPC) based on
hundreds of GCPs also obtained from the satellite GPS receivers for greater
accuracy.

Figure 5. Viewing vendor-provided GCPs in Focus

When available, Focus and SPTA automatically read the geolocation from the file
and display the image according to the map projection from the file projection and
bounds, rational functions model segment, or from the orbit segment.

PCI Geomatics Page 21

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Figure 6. Selecting georeferencing source

Figure 7. Select georeferencing layer

Note If an image is already open in Focus, but with a different

projection, a warning message appears, prompting you to
choose whether to add and reproject the file.

Page 22 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Finally, you choose whether to display the file North Up or Raster Up.

Figure 8. North Up or Raster Up selection

Note Focus and SPTA read geolocation information "as is", due
to the variation in positional accuracy of sensors and
acquisition modes. SAR data is also resampled and
projected "on the fly", meaning the data always remains
in its original format, unprojected, and at full resolution.
Any processing will be applied on the original data format.
If you want to increase the positional accuracy by
collecting external GCPs and permanently apply map
projection to the data of a given spatial resolution, you
must do so in OrthoEngine.

To take advantage of all the features offered in Geomatica, convert your data files
to PCIDSK format (.pix). By doing so, you can, in particular, create overviews
faster, and store auxiliary layers, such as lookup tables (LUT), pseudocolor tables
(PCT), bitmaps, and vectors. More information on the PCIDSK format is provided in
Lesson 1.5: Conversion utilities.

Matrix type
Geomatica characterizes SAR data according to a matrix type, which is determined
from the file metadata, the channel type, and the transmit/receive configuration.
Lesson 1.1: Supported SAR sensors and data formats discussed that channel type
is either complex (indicated by "c") or detected (indicated by "r"). A SAR image can
contain a complex or detected channel or channels only or a mix of both. The
number of channels of each type determines the matrix type. The matrix type is
important because many SAR algorithms require the input to be of a particular
matrix type.
The following table describes the matrix types for complex data.

PCI Geomatics Page 23

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Table 5. Matrix types: complex data

The following table describes the matrix types for detected data.

Table 6. Matrix types: detected data

Symbol Full name Total no. of Channels

Incomplete nonsymmetrized covariance
c4r matrix (fully polarimetric detected, single- 4
look, or multilook)
Incomplete symmetrized covariance matrix
c3r (fully polarimetric detected, single-look, or 3
multilook)
Incomplete covariance matrix (dual-pol,
c2r 2
detected, single-look, or multilook)
Incomplete covariance matrix (single-pol,
c1r 1
detected, single-look, or multilook)

Page 24 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Lesson 1.3: Complex data support in Geomatica

Native support for complex data was introduced in Geomatica 2014, in which the
real and imaginary parts of each channel are no longer stored in separate channels
and detected channels are no longer created and added to the file.

Figure 9. Radarsat-2, Fine quad-polarization, beam 29 (FQ29), May 7, 2010

opened in Geomatica 2013

Figure 10. Radarsat-2, Fine quad-polarization, beam 29 (FQ29), May 7, 2010

opened in Geomatica 2017
This native support of complex data offers many advantages, such as a reduced file
size on disk. The data interpretation of any complex channels can be changed
dynamically for faster data-content exploration.

PCI Geomatics Page 25

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

To change the data interpretation of complex images

1. In Focus, click the Maps tab.
2. Right-click a complex image, point to Data Interpretation, and then
click an interpretation type.
The interpretation type is applied automatically and the status bar displays the
numeric values displayed corresponding to the selected interpretation.

Figure 11. Interpretation options for complex data

Numeric Values window

You can view the numeric values of the selected applications and the original
complex values in the Numeric Values window.

To open the Numeric Values window

• Click the Layer menu, and then click Numeric Values.
The Numeric Values window appears.

Page 26 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Figure 12. Numeric Values window: Raw (Intensity)

Figure 13. Numeric Values window: Complex data

PCI Geomatics Page 27

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Other benefits of complex data support in Geomatica include the following:

• In OrthoEngine, the ability to geocode, orthorectify complex data, or both,
and keep the orthorectified files in complex format. For example, this is
necessary for coherent change detection when a greater positional
accuracy is needed while complex data is required.
• The ability to generate overviews, reproject, clip, or subset complex data.

Page 28 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Lesson 1.4: Ingesting and extracting a calibrated

backscatter image
In this lesson, you will:
• Ingest a SAR image
• Generate a calibrated backscatter image using various LUTs
• Explore the image metadata

Ingesting SAR data

There are two techniques you can use to ingest (import) SAR data into a PCIDSK
file.
The first is to open an image in Focus by its key file name, and then import it as a
PCIDSK file. The second technique is to use the SARINGEST algorithm.

To ingest a file using the Import to PCIDSK option

1. In Focus, open the image using its key file name.
2. If prompted, select a calibration type and a projection.
3. Click the Maps tab, and then select the file to import.
4. On the File menu, point to Utility, and then click Import to PCIDSK.
The source file is set automatically with the selected file in Focus.
5. In the Destination file box, enter a path and file name.

6. Click Import.
The File Selector window appears.
7. Enter a path and file name.
8. Click Save.
The image is imported (ingested) into a PCIDSK file.

Ingesting a file with SARINGEST

Alternatively, you can use SARINGEST to import (ingest) SAR data into a PCIDSK
file. SARINGEST produces the same output file as the Import to PCIDSK option.
However, an advantage of using SARINGEST is that you can automate the ingestion
process to import multiple files.

PCI Geomatics Page 29

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

To ingest SAR data using SARINGEST

1. In Focus, click the Tools menu, and then click Algorithm Librarian.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find box, type SARINGEST, and then click Find Next.
4. Double-click SARINGEST.
The SARINGEST Module Control Panel window appears.
5. Under Output Ports, below Output: SAR Dataset, select Untitled.pix,
and then click Browse.
6. In the File Selector window, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\ folder.
7. In the File name box, type FQ29_20100507sig.pix, and then click
Save.
The new path and file name appear in the SARINGEST Module Control
Panel window.
8. Click the Input Params 1 tab.
9. Click Browse, and then select the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\pro
duct.xml file.
10. For Data Calibration, click Sigma Nought, and then accept the default
settings for the remaining input parameters.
11. Select the Viewer check box, and then click Run.
The image automatically opens in Focus, along with the output GCPs. A
pop-up window appears, prompting you to choose whether to display the
image North Up or Raster Up.
12. Click North Up, and then click OK.
The RADARSAT-2 FQ-29 image is displayed in the view area with north at
the top of the viewer.

Page 30 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

Note When a file is opened in its original distribution format in

Focus, the calibration is applied only "on the fly" and not
permanently to the original data. However, after
importing (ingesting) a file to a PCIDSK file with a
calibration type, it will be applied permanently and can
neither revert to uncalibrated data nor to another
calibration type. To apply another calibration type, you
must reimport (reingest) the file.

Exploring the metadata

During ingestion, metadata is also imported. The metadata is retrieved directly
from the file header (usually an .xml or .txt file accompanying the data) or created
by Geomatica. The metadata is found at two levels, at the channel level and at the
file level.
At the channel level, metadata is limited to a set of standardized descriptors,
including the polarization type (HH, HV…), the data scale (linear, decibels), and the
matrix-element position (1_1, 2_1, 4_4…) of the channel. Metadata at the channel
level is read and used by many SAR processing algorithms.
At the file level, the metadata varies according to the sensor. While most of the
metadata is only for information, some is used by certain algorithms, such as the
matrix type, calibration type, and the number of looks.

To view the metadata at the file level

1. In Focus, click the Files tab.
2. Select the image for which you want to view the metadata.
3. Right-click the file, and then click Properties.
The File Properties window appears.

PCI Geomatics Page 31

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

4. Click the MetaData tab.

The file-level metadata is displayed.

Note It is not mandatory to import a file in PCIDSK format to

view its metadata.

To view the metadata at the channel level

1. In Focus, click the Files tab.
2. Select the image for which you want to view the metadata, and then
expand the Rasters list.
3. Right-click a raster channel, and then click Properties.
The Raster Properties window appears.

Page 32 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

4. Click the MetaData tab.

The channel-level metadata is displayed.

SAR geometry and radiometric calibration

For most geoscience applications—quantitative analysis (biomass, soil wetness, for
example) and time-series analysis in particular—SAR data must be calibrated. You
also may need to compensate for distortions in the imagery due to the geometry of
acquisition.

SAR geometry
Most synthetic-aperture radars (SAR) used in geoscience application are usually
side-looking. A series of waves, or pulses, are transmitted by the antenna toward
the ground. Between each transmission, the same antenna is used to receive the
transmitted signal.

PCI Geomatics Page 33

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Each transmitted pulse is carefully controlled; that is, the frequency, polarization,
and phase of the signal are known. After the signal is scattered back to the sensor,
the travel time, backscatter power, and phase is compared to the original pulse.
Typically, over a thousand pulses are coherently averaged to form a single pixel.
With a SAR image, the location of each pixel and its resolution is a combination of
the time it took to be reflected back to the sensor (x, range resolution) and the
time between two pulses relative to the platform velocity (y, azimuth resolution).
Because it takes more time for a pulse to reach the far range of the radar swath,
the slant-range resolution is coarser at the far range than the near range (Figure
14 and Figure 15).

Figure 14. Slant-range image versus ground-range image2

The SAR-image geometry can be determined by the following parameters3:
• H: Height of SAR-imaging system moving at a velocity VSAR along the
azimuth direction (y).
• Lx and Ly: Antenna physical dimensions in height (Lx) and width (Ly),
typically in meters, aiming perpendicular to the flight direction (y).
Directed slant-wise toward the ground with an incidence angle  0 .

• x: Ground-range direction
• y: Azimuth direction
• r: Antenna radial axis (radar line of sight [LOS])

2 Source: European Space Agency (ESA),2010-2018, Radar Course 2 – 14. Slant

range/ground range.
3 Lee and Pottier, 2009

Page 34 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

• Radar swath: Area scanned by the antenna beam

• Antenna footprint: Defined from the antenna apertures (  X , Y ) given

by  X   LX and  Y   LY , where  is the wavelength corresponding
to the carrier frequency of the transmitted signal

Figure 15. SAR Geometry4

Slant-range geometry is the natural result of radar-range measurements.
Sometimes, however, it is preferable to convert the data to ground-range
geometry, which is often characterized by square pixels. Typically, this is done
using a resampling of the data when it is geocoded.
Transformation to ground range also requires a radiometric correction at each data
point to compensate for incidence angle and local terrain slope and elevation. This
is especially true for wide-swath-imagined modes with the swath spanning more
than hundreds of kilometers. This process is called radiometric calibration.

4 Source: Lee and Pottier (2009). Fig,1.1, p.6.

PCI Geomatics Page 35

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Extracting a calibrated backscatter image from SAR data: Radiometric

calibration
Calibration is the process whereby the digital values of a received remotely sensed
image can be related to physical quantities of the scene, such as brightness
(reflectivity), phase, and location.
"Radiometric calibration of the data is required so that information on the
magnitude of the radar backscatter coefficient of the imaged terrain can be
extracted from the processed image data. There are two fundamental requirements
for radiometric calibration (RSI, 2004)":
• Relative radiometric calibration stability (evaluation of the electrical
stability of the radar sensor hardware, antenna elevation beam pointing
estimation, and pattern correction to reduce errors due to variations in the
angle of satellite roll)
• Absolute radiometric calibration knowledge (using precision active
transponders, set to a known radar cross-section (RCS), and passive
corner reflectors or areas of known distributed target backscatter
coefficient previously measured independently by other SAR sensors or
scatterometers)
The goal of radiometric calibration is to adjust for all of the contributions in the
radiometric values not due to target characteristics, so that the backscatter value of
targets can be compared to one another or to a reference. A proper radiometric
calibration is essential for any quantitative and temporal analysis of SAR data.

Output scaling of the processor

Most data products are calibrated radiometrically by the data supplier prior to being
distributed. The last step in processing is the application of lookup tables (LUT),
during calibration from floating-point data to digital numbers (DN) to ensure the
best use of the 8 or 16 bits in storage media. These LUTs apply a fixed offset and a
range-dependent gain function to the data prior to generating the final image
output. The scaling used can vary by scene, making it difficult to directly relate
information between each.
The LUTs provided with the data are used by Geomatica to extract the calibrated
image. To ensure proper SAR processing, pixel values (DNs) should first be
converted to sigmanought (  ), and used subsequently for any further analysis.
0

Geomatica also has tools for converting from DNs to both betanought (  ) and
0

gammanought (  ).
0

The radar backscatter coefficient (σ0 ) is used commonly and it is expressed per
unit area in ground range (Figure 16).
"A problem arises if there is a nonzero slope at the local terrain site. In this case,
the projected area is determined by the local incidence angle. It follows that the
correct values for σ° cannot be obtained unless one has at hand a reliable estimate
of the local slope (Raney, 1998)".
The radar brightness (β0 ), is the most natural and observable radar measurement
(Raney et al. 1994) and corresponds to the backscatter-per-unit area in slant
range, and requires no knowledge of local incidence angle (Figure 16, Figure 17 and
Figure 18). For the detected product, the radar brightness corresponds to:

Page 36 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

 0j = (DN 2j + A3)/ A2 j
th
where DNj is the digital number that represents the magnitude of the j pixel from
the start of a range line in the detected image data, and A2 j is the scaling-gain
th
value for the j pixel, and A3 is the fixed offset. Radar brightness in decibels (dB)
is given by:

 0j (dB) = 10 * log 10 (DN 2j + A3)/ A2 j 

For complex (SLC) single-beam products, the pixel number j is related to the LUT
index i, using the same procedure as for detected products. The radar brightness
th
for the j range pixel is then given by:


 0j = (DN I j / A2 j )2 + (DN Q j / A2 j )2 
th
where DNIj and DNQj are the digital values of the I and Q components of the j
pixel from the start of the range line, and A2 j is the corresponding range-
dependent gain. The offset is not used in SLC-product generation. For complex
data, radar brightness in decibels (dB) is given by:


 0j (dB) = 10 * log 10 (DN I j / A2 j )2 + (DNQ j / A2 j )2 
 0 can be converted into  0 using:

 0j =  0j * (sin  j )

 0j (dB) =  0j (dB) + 10 * log 10 (sin  j )

 0 can be converted into  0 using:

 0j =  0j * (tan  j )

 0j (dB) =  0j (dB) + 10 * log 10 (tan  j )

where j is the incidence angle at the j

th
range pixel.

In Figure 16, dR is the slant-range distance entering into the definition of 0.
0 represents the average reflectivity of a horizontal material sample, normalized
with respect to a unit area AL on the horizontal ground plane.

PCI Geomatics Page 37

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

 0 is defined with respect to incident area Ai , orthogonal to the incident ray from
the radar.

( 0 ) ( 0 )

( 0 )

Figure 16. Definition of surface area and incident area used to determine  0,  0
and 05
With newer SAR sensors, such as TerraSAR-X or RADARSAT-2, for example, the
calibrated data is read on the fly, or calibration is performed during ingest with the
SARINGEST algorithm.

5 Source: Raney (1998), fig.2-8, p.25 and CCRS (2007)

Page 38 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

j j
Figure 17. Depression, elevation, and incidence angle: =

j j
Figure 18. Depression, elevation, and incidence angle: ≠

PCI Geomatics Page 39

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

Lesson 1.5: Conversion utilities

In this lesson, you will:
• Convert complex data to other interpretations using the PSIQINTERP
algorithm
• Convert one matrix type to another using the PSCONV algorithm
Whether to meet the specific needs of a workflow or simply for data exploration, it
is sometimes necessary to convert data from one format to another.
Geomatica includes the following two conversion algorithms to provide you with
such flexibility:
• PSIQINTERP
• PSCONV

Converting complex data to other interpretations

Some SAR-processing algorithms require detected data in the input; such as most
SAR-speckle filtering algorithms (FEFROST, FELEE, FGAMMA, FTOUZI, et al).
The PSIQINTERP algorithm converts SAR images in single-look (SLC) or multilook
(MLC) complex format to other interpretations, such as detected data or phase
product. PSIQINTERP can also extract the real or imaginary part of a complex
number, or store each part in a separate channel.

To convert a complex channel to other interpretations

1. In Focus, click the Tools menu, and then click Algorithm Librarian.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find box, type PSIQINTERP, and then click Find Next.
4. Double-click PSIQINTERP.
The PSIQINTERP Module Control Panel window appears.
5. Under Input Ports, click Browse, and then select the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
\FQ29_20100507sig.pix file.
6. Select the first channel (HH).
7. Under Output Ports, click the Output: SAR Dataset entry, click
Untitled.pix, and then click Browse.
8. In the File Selector window, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
9. In the File Name box, type FQ29_20100507sig_HH_dB.pix, and then
click Save.
The new path and file name appear in the PSIQINTERP Module Control
Panel window.
10. Click the Input Params 1 tab.
The output data-interpretation options are displayed.

Page 40 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

11. In the Output data interpretation list, click Decibel.

12. Click Run.

Exercise 1: Using the PSIQINTERP algorithm, extract from

FQ29_20100507sig.pix different complex channel interpretations for one or all
channels.

Converting a matrix representation to another matrix type

PSCONV converts a polarimetric SAR (POLSAR) data set between different matrix
formats, required by different polarimetric SAR algorithms. The available output-
matrix formats depend on the input-matrix type, as described in the following
procedure. Here, the entire input data set is converted.

To convert a matrix representation to another matrix type

PCI Geomatics Page 41

Module 1: SAR data structure and support in Geomatica Processing SAR data with Geomatica Banff

5. Under Input Ports, click Browse and select the

~\ SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
\FQ29_20100507sig.pix file.
6. Select all channels (HH, HV, VH, and VV).
7. Under Output Ports, click Untitled.pix, and then click Browse.
8. In the File Selector window, open the
~\ SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
9. In the File Name box, type FQ29_20100507sig_HH_c4r6c.pix, and
then click Save.
The new path and file name appear in the PSCONV Module Control Panel
window.
10. Click the Input Params 1 tab.
11. In the Output matrix representation list, click c4r6c.
This corresponds to a non-symmetrized covariance matrix.
12. Click Run.
The original complex-scattering matrix (s4c) is converted to a
nonsymmetrized covariance matrix (c4r6c).

Page 42 PCI Geomatics

Processing SAR data with Geomatica Banff Module 1: SAR data structure and support in Geomatica

You can use the PSCONV algorithm when a specific matrix format is required for a
polarimetric algorithm that does not automatically perform the conversion. Some
decomposition algorithms require a specific matrix representation as input. For
example, the Cloude-Pottier decomposition (PSEBA) requires a filtered-coherency
matrix while output from polarimetric filters corresponds to a covariance matrix.
Because the matrix type is read from the metadata, Geomatica will automatically
convert the covariance matrix to the coherency matrix (matrix type) before
applying the decomposition.

Table 7. Possible matrix conversions

Conversion s4c S3c s2c s1c c4r6c C3r3c t4r6c T3r3c k16r K9r
to ↓↓↓
s4c -- √
S3c √ --
s2c --
s1c --
c4r6c √ -- √ √ √
C3r3c √ √ √ -- √ √ √ √
c2r1c √
t4r6c √ √ --
T3r3c √ √ √ -- √ √
k16r √ √ --
K9r √ √ --
c2r √
c1r √

In conversions between a symmetrized and a nonsymmetrized form of the same

matrix type, the cross-polarization terms are duplicated. In conversions between a
nonsymmetrized and a symmetrized form of the same matrix type, the cross-
polarization terms are averaged to produce a single common value.

PCI Geomatics Page 43

Module 2: Applying radiometric
enhancements

About this module

Module 2 has three lessons:
• Lesson 2.1: Applying SAR-speckle filters to detected data
• Lesson 2.2: Applying the PSBOXCAR filter to fully polarimetric data
• Lesson 2.3: Applying the PSPOLFIL filter to fully polarimetric data

Radiometric enhancement
The goal of radiometric enhancement is to improve the interpretation of the
radiometric information in an image using speckle and spatial filters. These filters
can reduce speckle, detect edges, analyze texture, and visually enhance the image.
Image variance, or speckle, is a granular noise that is inherent to SAR imagery.
Speckle gives a grainy, salt-and-pepper appearance and tends to be a dominating
factor in radar imagery. Speckle filters are used primarily with radar data to
remove high-frequency noise (speckle), while preserving high-frequency features
(edges).

SAR-speckle filters
Coherent signal-scattering in SAR data often causes image speckle or a salt-and-
pepper effect. Speckle is inherent to most SAR images, and can inhibit accurate
image interpretation. There are several types of speckle filters and they generally
occur in either of two general categories:
• Nonadaptive, or template
• Adaptive
Nonadaptive filters apply to the parameters of the whole image. They do not take
into account the local properties of terrain backscatter or the nature of the sensor.
Examples of nonadaptive filters are Mean, Median, Edge Detection, and Sieve.

PCI Geomatics Page 45

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

Adaptive filters accommodate changes in local properties of the terrain backscatter.

They modify the image based on statistics extracted from the local environment of
each pixel. The main advantage of adaptive filters is that they maintain an accurate
estimate of the backscattering coefficient inside homogeneous (stationary) areas,
while preserving edge and texture structure in nonstationary scenes. Examples of
adaptive filters are Frost, Lee, Gamma, and Touzi. Adaptive filters are designed
specifically for SAR images.
The ideal speckle-reduction filter reduces the speckle, but has minimal loss of
information. The filter should preserve radiometric information in homogeneous
areas, and preserve radiometric information and spatial-signal variability (textural
information) in textured areas.

Table 8. SAR-speckle filters available in Geomatica

Group Code Filter Channel Description

(input)
Single-channel [FLE] Lee adaptive Detected Adaptive, preserves
filters and high-frequency features,
complex such as edges
[FELEE] Enhanced Detected Adaptive, preserves
Lee and high-frequency features,
complex such as edges, preserves
isolated point target
[FKUAN] Kuan Detected Adaptive, preserves
and high-frequency features,
complex such as edges
[FFROST] Frost Detected Adaptive, preserves
adaptive and high-frequency features,
complex such as edges
[FEFROST] Enhanced Detected Adaptive, retains edges
Frost and and sharp features,
Adaptive complex preserves isolated point
target
[FGAMMA] Gamma MAP Detected Adaptive, preserves
high-frequency features,
such as edges
[FTOUZI] Touzi Detected Multiresolution adaptive
and filter, preserves high-
complex frequency features, such
as point targets,
curvilinear features, and
edges
[FLAP] Laplacian Detected Filters high-frequency
and noise (speckle), but
complex enhances high-frequency
features, such as edges
Polarimetric [PSPOLFIL] Lee adaptive Detected Adaptive, preserves
SAR filters and edges
complex

Page 46 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Running PSPOLFIL or PSBOXCAR is a prerequisite for many polarimetric

decompositions (Module 4). Contrary to others filters, they can modify the matrix
representation and channel type for the following inputs:
• Input s4c → output c4r6c
• Input S3c → output C3r3c
• Input s2C → output c2r1c
For input data in all other matrix formats, the output matrix type and the number
of channels in the output file are the same as in the input file.

Common scenarios for filtering SAR speckle

SAR-speckle filtering occurs at different stages depending on the application and
data type of the input.
Most SAR sensors acquire data in single-pol and dual-pol acquisition modes. At the
lowest processing level, the data is delivered as single-look complex (SLC). This
data must first be converted to detected data with PSIQINTERP before applying a
radiometric SAR filter to remove the speckle (scenario A). SAR data can be
delivered (ordered) at higher processing levels in which they have been already
converted to detected and, sometimes, multilooked. In such cases, you can directly
apply a radiometric SAR filter to the data (scenario B).
Some SAR sensors also have a fully polarimetric imaging mode. This type of data is
used typically in polarimetric analysis, which includes the polarimetric
decomposition (Module 3). In this situation, you must use PSPOLFIL or PSBOXCAR
to filter the data (scenario C). Single-look complex (SLC) single-pol or dual-pol
imagery can also be filtered with PSPOLFIL and PSBOXCAR, but contrary to
scenario A, the output will remain complex (scenario C).
Fully polarimetric images (s4c, S3c) can be converted to fully polarimetric detected
data (c4r, C3r) before being filtered with the radiometric SAR filters (scenario D).
Note that, in this scenario, phase information is lost.

PCI Geomatics Page 47

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

Scenarios E and F depict situations in which the emphasis is on the characterization

of particular targets or land-use classes. Instead of relying solely on HH, HV, VH,
and VV channels, you can synthesize new channels (PSPOLSYN) for an arbitrary set
of transmitted and received polarizations. It also possible to maximize the contrast
of a target (PSPOLSYNR) or between two targets (PSPOLSYNC) by finding the set of
transmitted and received polarizations that maximize this contrast in power
(Appendix B).
The following are typical situations were scenarios A, B, and C occur most
commonly.

A s2c → Conversion to detected → Radiometric SAR filters → Analysis

s1c data [PSIQINTERP]

c1r
B → Radiometric SAR filters → Analysis
c2r

Polarimetric
s4c Polarimetric SAR filters decomposition
S3c
C → [PSPOLFIL] or → (possible only with s4c → Analysis
s2c
[PSBOXCAR] and S3c data, see
s1c
Module 3).

s4c Conversion to detected

D → → Radiometric SAR filters → Analysis
S3c data [PSIQINTERP]

[PSPOLSYN]
s4c [PSPOLSYNR]
E → → Radiometric SAR filters → Analysis
S3c [PSPOLSYNC]
(Appendix B)

Polarimetric SAR filters [PSPOLSYN]

s4c [PSPOLSYNR]
F → [PSPOLFIL] or → → Analysis
S3c [PSPOLSYNC]
[PSBOXCAR] (Appendix B)

Figure 19. Common scenarios with SAR-speckle filtering

Page 48 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Lesson 2.1: Applying SAR-speckle filters to detected

data
In this lesson, you will:
• Filter an image using different SAR-adaptive filters
• Compare different filtering results from SAR-adaptive filters

Applying an enhanced Frost filter

With the Frost filter, the unspeckled pixel value is estimated for each pixel by
computing the weighted sum of the center pixel value, the mean value, and the
variance calculated in a circular kernel surrounding the pixel. The enhanced Frost
filter minimizes the loss of radiometric and textural information.

Data preprocessing
Using the concepts learned in Lesson 1.4: Ingesting and extracting a calibrated
backscatter image and Lesson 1.5: Conversion utilities and the Vancouver scene
located in ~\SAR_Training\Radar\Vancouver_RS2_FQ2_SLC, do the following:
1. Ingest the product.xml file into a PCIDSK file using sigma as the
calibration type.
Name the file Van_RS2_FQ02sig.pix.
2. Use the PSIQINTERP algorithm to convert the complex data to detected
data.
Convert the HH channel to Intensity. Name the file
Van_RS2_FQ02sig_HH.pix.
This creates a (simulated) detected single-pol image in HH.

To open the FSPEC algorithm

To set up the parameters for FSPEC

1. Under Input Ports, expand Input: Unfiltered Layers.
The tree expands to show the array segments.
2. Click Browse, open the
~\SAR_Training\Radar\Vancouver_RS2_FQ2_SLC folder, select
Van_RS2_FQ02sig_HH.pix, and then select the HH channel for this file.
3. Under Output Ports, click the Output: Filter Layer(s) entry, click
Untitled.pix, and then click Browse.
The File Selector window opens.
4. Open the Radar Data folder.

PCI Geomatics Page 49

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

5. In the File Name box, type Van_RS2_FQ02sig_HH_FFrost7.pix.

6. Click Save.
The path and file name are updated in the FSPEC Module Control Panel
window.
7. Click the Input Params 1 tab.
8. In the Filter Type list, select Enhanced Frost.
9. In the Filter X list, select 7.
10. In the Filter Y list, select 7.
The filter kernel size will be 7 x 7 pixels.
11. In the Number of Looks box, enter 1.
12. In the Damping Factor box, you need not change the default value of 1.
You must enter correct parameters for the image files you are working
with. The number of looks and the image-type information is available in
the format definition included with your data or in the file metadata.
13. In the Image Units list, select Power.
Image units specify the image format of the radar image.
Note: Intensity = Power, Amplitude = Magnitude
14. Click Run.
The Enhanced Frost filter runs and the result displays in Focus.

Page 50 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Applying a Touzi filter

Unlike other filters, the Touzi filter adapts the size and shape of the filter-
processing window based on signal nonstationarity. It can also detect
multiresolution ratio edges for better filtering of contours and edges. This is
effective when applied to fine structures, such as roads and trails, which other
filters tend to smooth out.

To apply a Touzi Filter

1. In the FSPEC Module Control Panel window, click the Files tab.
2. Under Input Ports, if Van_RS2_FQ02sig_HH.pix is not already listed,
click Browse, and then select the file in the
~\SAR_Training\Radar\Vancouver_RS2_FQ2_SLC folder.
3. Select the HH channel for this file.
4. Under Output Ports, specify Van_RS2_FQ02sig_HH_FTouzi11.pix as
the file name.
Save the file the in the
~\SAR_Training\Radar\Vancouver_RS2_FQ2_SLC folder. The path
and file name are updated in the FSPEC Module Control Panel window.
5. Click the Input Params 1 tab, and then do the following:
▪ In the Image Units list, select Power.
▪ In the Filter Type list, select Touzi.
▪ In the Filter X list, select 11.
The filter kernel size will be 11 x 11 pixels.
▪ In the Number of Looks box, enter 1.
▪ In the Contour Threshold box, type 0.5.
▪ In the Edge Threshold box, type 0.5.
▪ In the Gradient Threshold box, type 0.1.
6. Click Run.
The Touzi filter runs and the results display in Focus.
The two SAR speckle filtered images can now be compared in Focus.

Exercise 1: With FSPEC, run the enhanced Frost filter again on the same file
(Van_RS2_FQ02sig_HH.pix). Specify a window size of 11. Specify the following
name in the Output: File Layer (s) port:
Van_RS2_FQ02sig_HH_FFrost11.pix.
Run the Average filter to get another comparison basis to evaluate the results
obtained from the SAR adaptive filters.
Compare the results to the original HH SLC channel (Van_RS2_FQ02sig_HH.pix),
both visually and numerically.

PCI Geomatics Page 51

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

Stats: HH SLC
• Water: 0.343±0.35
(med: 0.226)
• Forest: 0.199±0.24
(med: 0.12)
•H-D. Urban:
2.159±5.56
(med: 0.50)
• L-D. urban:
0.212±0.32
(med: 0.11)

HH Enh. Frost 7 x 7 (HH SLC) – (Enh. Frost 7 x 7)

Stats: HH Enh. Frost
11 x 11
• Water: 0.340±0.12
(med:0.314)
• Forest:
0.198±0.12
(med:0.175)
• H-D. Urban:
2.150±5.34
(med:0.559)
• L-D. Urban:
0.214±2.26
(med:0.160)
HH Enh. Frost 11 x 11 (HH SLC) – (Enh. Frost 11 x 11)

Page 52 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Stats: Touzi 11 x 11
• Water: 0.263±0.102
(med:0.258)
• Forest:
0.152±0.077
(med:0.148)
• H-D. Urban:
1.515±3.03
(med:0.592)
• L-D. urban:
0.163±0.117
(med:0.133)

Touzi 11 x 11 (HH SLC) – (Touzi 11 x 11)

Stats: Average 7 x 7
• Water: 0.338±0.65
(med:0.329)
• Forest: 0.197±0.05
(med:0.19)
• H-D. Urban:
2.126±2.06
(med: 1.38)
• L-D. urban:
0.227±0.32
(med:0.08)

Average 7 x 7 (HH SLC) – (Average 7 x 7)

Stats: Average
11 x 11
• Water: 0.331±0.03
(med:0.330)
• Forest:
0.196±0.03
(med:0.190)
• H-D. Urban:
2.184±5.56
(med:1.39)
• L-D. urban:
0.228±0.32
(med:0.05)
Average 11 x 11 (HH SLC) – (Average 11 x 11)

Figure 20. Result of different filters and windows size

PCI Geomatics Page 53

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Filtered an image using enhanced Frost, Gamma, and Touzi adaptive
filters
• Compared visually and numerically the results of the different filter
operations

Page 54 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Lesson 2.2: Applying the PSBOXCAR filter to fully

polarimetric data
In this lesson, you will:
• Apply a polarimetric filter to a RADARSAT-2 image
• Compare the effect of different processing window sizes
The PSBOXCAR algorithm applies a boxcar filter to an input polarimetric SAR
(POLSAR) data set. The boxcar filter replaces every image pixel by an average of
pixel values in a window centered on the pixel. All of the raster channels in the
input image are filtered independently.
Boxcar filtering is usually used to increase the effective number of looks (ENL) in
the data. This step is required if an algorithm requires a minimum ENL that is
higher than that of the input data. The input image must represent a
nonsymmetrized or symmetrized fully polarimetric data set, which can be in
scattering (s4c, S3c or s2c), covariance (c4r6c or C3r3c), coherency (t4r6c or
T3r3c), or Kennaugh (k16r or K9r) matrix format.
Input data sets in scattering-matrix format are converted to the covariance-matrix
format before filtering. With such data, the output-matrix type and the number of
channels in the output file depend on the input-matrix type, as follows:
• Input s4c: output c4r6c, 10 channels
• Input S3c: output C3r3c, six channels
• Input s2c output c2r1c, three channels
For input data in all other matrix formats, the output-matrix type and the number
of channels in the output are the same as those of the input. The physical quantity
of output pixel values (sigmanought, betanought, or gammanought) is always the
same as in the input data set.

Applying the PSBOXCAR filter

The filter will be applied to the fully polarimetric Radarsat-2 FQ data.

To open the PSBOXCAR algorithm

1. In Focus, on the Tools menu, click Algorithm Librarian.
The Algorithm Librarian window appears.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find what box, type PSBOXCAR, and then click Find Next.
4. Double-click PSBOXCAR.
The PSBOXCAR Module Control Panel window appears.

PCI Geomatics Page 55

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

To set up the parameters for PSBOXCAR

1. Under Input Ports, expand Input: Input polarimetric SAR image.
2. Click Browse, open the
~\SAR_Training\Radar\Vancouver_RS2_FQ2_SLC folder, and then
select the Van_RS2_FQ02sig.pix file generated in Lesson 2.1.
3. Select each of the channels in the list.
4. Under Output Ports, expand Output: Output polarimetric SAR
filtered image, select Untitled.pix, and then click Browse.
The File Selector window appears.
5. In the File name box, type Van_RS2_FQ02sig_PSBOXCAR_7.pix, and
then click Save.
6. Click the Input Params 1 tab.
7. In the Horizontal filter size list and Vertical filter size list,
respectively, select 7.
The choice of a particular window size will have a strong impact on the
results produced by a polarimetric decomposition algorithm.
8. Click Run.
Repeat steps 3 to 8, specifying a different window size, and then compare the
results.

Page 56 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

SLC 3x3 5x5

7x7 9x9 11 x 11

15 x 15 21 x 21 31 x 31

Figure 21. Various window sizes with the PSBOXCAR algorithm

PCI Geomatics Page 57

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Applied a polarimetric filter to the RADARSAT-2 image
• Compared the effect of various sizes of processing window

Page 58 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Lesson 2.3: Applying the PSPOLFIL filter to fully

polarimetric data
In this lesson, you will:
• Apply a polarimetric filter to the RADARSAT-2 image
• Compare the effect of different processing window size
The PSPOLFIL algorithm applies a speckle filter to the fully polarimetric SAR data
set. Unlike the boxcar filter, this filter preserves the edges. Edge-detection and
homogeneity parameters are estimated from the local power image. The input
image must represent a nonsymmetrized or symmetrized fully polarimetric (quad-
polarization, complex) data set in the scattering (s4c or S3c), covariance (c4r6c or
C3r3c), coherency (t4r6c or T3r3c), or Kennaugh (k16r or K9r) matrix format.
Input data sets in the scattering-matrix format are converted to the covariance-
matrix format before filtering. With such data, the output-matrix type and the
number of channels in the output file depend on the input-matrix type, as follows:
• Input s4c: output c4r6c, 10 channels
• Input S3c: output C3r3c, six channels
• Input s2c output c2r1c, three channels
For input data in all other matrix formats, the output-matrix type and the number
of channels in the output file are the same as in the input file.

Applying a speckle filter

The PSPOLFIL algorithm applies a speckle filter.

To open the PSPOLFIL algorithm

To set up the parameters for PSPOLFIL

1. Under Input Ports, expand Input: Input polarimetric SAR image,
click Browse, and then open the Van_RS2_FQ02sig.pix file generated
in Lesson 2.1.
2. Select each of the channels in the list.
3. Under Output Ports, expand Output: Output polarimetric SAR
filtered image, select Untitled.pix, and then click Browse.
The File Selector window appears.
4. In the File Name box, type Van_RS2_FQ02sig_PSPOLFIL_7.pix, and
then click Save.
5. Click the Input Params 1 tab.

PCI Geomatics Page 59

Module 2: Applying radiometric enhancements Processing SAR data with Geomatica Banff

6. In the Filter size list, select 7.

Note: The choice of a particular window size will have a strong impact on
the results produced by a polarimetric decomposition algorithm.
7. Accept the default Number of Looks (1).
Note: You can estimate the ENL by calculating the mean squared-to-
variance ratio (of the intensity data) in a homogenous area.
8. Click Run.

Exercise 3: When the module runs to completion, open the file in Focus, and then
compare the results of this filter to those of the boxcar filter.

Not applicable

SLC 3x3 5x5

7x7 9x9 11 x 11

15 x 15 21 x 21 31 x 31

Figure 22. Various window sizes with the PSPOLFIL algorithm

Page 60 PCI Geomatics

Processing SAR data with Geomatica Banff Module 2: Applying radiometric enhancements

Lesson summary
In this lesson, you:
• Filtered a quad-polarized image using the PSPOLFIL algorithm
• Compared the PSPOLFIL filtering with PSBOXCAR and the original
unfiltered data

PCI Geomatics Page 61

Module 3: Polarimetric
decompositions

About this module

Module 3 has four lessons:
• Lesson 3.1: Introduction to radar polarimetry
• Lesson 3.2: Freeman-Durden decomposition
• Lesson 3.3: Cloude-Pottier decomposition
• Lesson 3.4: Touzi decomposition

Lesson 3.1: Introduction to radar polarimetry

The study of polarimetry refers to the science behind, and the techniques relating
to, the analysis of the polarimetric properties of electromagnetic waves, and the
scattering of these waves.
In single-polarized images, the return is proportional to the target cross-section.
The amount of backscattering the radar system measures is dependent on the
target, the relative orientation of the target and radar system, and the polarization.
When orthogonal polarization combinations are used, the complete scattering
properties of a radar target can be determined for a specific frequency, incidence
angle, and azimuth direction. The extraction of information, such as terrain
classification, is an important component of polarimetry.
With a monostatic (that is, the 𝑇
⃗ 𝑥 and 𝑅⃗𝑥 antennas are colocated) polarimetric
radar, the transmitting antenna (𝑇 ⃗ 𝑥 ) and the receiving antenna (𝑅⃗𝑥 ) can be modeled
as a vector containing two polarization states that form an orthogonal basis.
Because the polarization state of the transmitted signal is known, and the state of
the received signal, the polarimetric properties of the target can be represented.
This matrix is called the complex-scattering matrix ([𝑆]).

⃗ 𝑥 = [𝐻 ], 𝑅⃗𝑥 = [𝐻 ]
𝑇
𝑉 𝑉

𝑇 𝑆 𝑆𝐻𝑉
⃗ 𝑥 • 𝑅⃗𝑥
𝑇 = [𝑆], [𝑆] = [ 𝐻𝐻 ]
𝑆𝑉𝐻 𝑆𝑉𝑉

PCI Geomatics Page 63

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

It is important to stress that the retrieval of the polarimetric properties of a target

is possible only if the phase information of the H and V polarization is carefully
controlled in the SAR hardware, and carefully measured and preserved at the
signal-reception and focusing (processing) stages.

Figure 23. Diversity of polarization modes6

There are several advantages to using polarimetric data in image analysis.
Agricultural applications include crop-type identification, crop-condition monitoring,
and soil-moisture measurement. Forestry applications include biomass estimation
and species identification. Other applications of polarimetric data include geology,
hydrography, oceanography, and wetlands mapping.

Canonical targets
Canonical targets correspond to simple geometric structures whose interpretation
of diffusion is facilitated by the presence of symmetry planes in the matrices used
to represent them. Interpretation of polarimetric responses, like that of the
parameters from a polarimetric decomposition, is often based on a comparison with
the canonical targets.

Scattering matrix Co-pol polarimetric

Target
(H-V basis) response

1 0 
Sphere = S = 
0 1 

6 Source: JAXA (2012), adapted from eoPORTAL

Page 64 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Scattering matrix Co-pol polarimetric

Target
(H-V basis) response

1 0 
Trihedral = S= 
0 1 

1 0 
Dihedral = S = 
0 −1

For an oriented dihedral

cos 2 sin 2 
S= 
 sin 2 − cos 2 

Horizontal dipole
1 0 
S= 
0 0 

Vertical dipole

0 0 
S= 
0 1 

PCI Geomatics Page 65

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Scattering matrix Co-pol polarimetric

Target
(H-V basis) response

Oriented dipole
 1 2 
 cos  sin  
2

S= 2

 1 sin 2  sin  
2
 2 

Left helix

1 j 
S= 
 j −1

Right helix

 1 − j
S= 
 − j −1

Figure 24. Canonical targets representation (in H-V basis), (Source: van Zyl et
Ulaby (1990), p.33-45.)
The sphere and the trihedral are both characterized by an odd number of bounces,
which results in a phase difference of zero degrees in the backscatter alignment
(BSA) convention. Each produces a uniform scattering for all linear polarizations
(χ=0°), which results in HH = VV. For an even number of bounces, (dihedral, for
example) the target introduces a phase difference of 180 degrees between the HH
and VV polarization and HH = -VV.

Page 66 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

However, the HH and VV channels are still equal in intensity. If the dihedral in a
plane perpendicular to the radar line of sight (LOS) is rotated, it introduces a
depolarization of the signal, and the HV and VH channels are no longer equal to
zero. For a pure canonical dihedral, the signal is repolarized more than depolarized.
If the value of the angle ϕ is found, it is possible to cancel out its effect and retrieve
a pure dihedral scattering.
A dipole will produce a strong scattering in only one polarization channel according
to its orientation. Like the dihedral, it is possible to cancel out the effect of the
orientation angle ϕ in a case where it is not equal to zero (HH) or 90 degrees (VV).
Finally, the helix is an abstract construction, because it does not correspond to a
real physical target. This kind of scattering can occur in an urban environment
where multiple scattering is common. One way to produce a pure-helix scattering is
to place two dihedrals, oriented at 45 degrees from each other. The phase
difference between the HH and VV channel will be 180 degrees.
If a pixel, or a small group of pixels, corresponds to a canonical target, it will
generally produce a highly polarized and strong scattering. These pixels are called
point targets or coherent-point targets.

Backscattering mechanisms
The reality, however, is often more complex. Only a small fraction of an image pixel
corresponds to a coherent-point target. The scattering mechanisms tend to be
horizontally and vertically superposed. In such a case, it is necessary to average
(multilook) several pixels, to estimate the dominant scattering mechanisms, if any,
and the degree of polarization.
For example, a "forest stand can be broken down into individual components that
drive radar backscatter, these components are (Lo, 1998):
• (a) Direct backscattering from the soil surface
• (b) Volume scattering from foliage, shrub canopy, and leaf litter, if present
• (c) Direct backscattering from big branches and trunks if they are rough
or at normal incidence
• (d) Interaction components due to corner reflection from tree trunks
• (e) Other interaction components due to multiple scattering between
foliage and ground surface, big branches and surface, foliage and shrub
canopy, and so forth"
With the same area, depending on the characteristics of the sensor (incidence and
orientation angle, spatial resolution, and wavelength), various scattering
mechanisms might dominate the scattered signal. Several polarimetric
decompositions have been proposed to facilitate the interpretation of the scattered
mechanism of a fully polarimetric image.

PCI Geomatics Page 67

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Figure 25. Horizontal and vertical superposition of scattering mechanisms7

Polarimetric decomposition
There are two families of polarimetric decomposition: the coherent and the
incoherent decomposition.
The coherent target decompositions are applied only on SLC images and
generally pixel-by-pixel for the characterization of a point target. Although a
dominant scattering mechanism can be found for each pixel of an image, a
coherence test is applied generally on each before the decomposition to ensure a
meaningful result. This topic will be covered in Lesson 4.
As mentioned previously, most natural targets, called extended targets, are
incoherent; that is, they cover more than one pixel, are partially polarized, and in
most cases have more than one scattering mechanism for any given pixel. The
complex-scattering matrix ([S]) is no longer appropriate to represent an incoherent
target, and a second-order representation is needed. Several matrices can be used
to represent an incoherent target: the Mueller ([M]), Kennaugh ([K]), the
coherency ([T]) and the covariance ([C]) matrices are used commonly:

Symmetrized-covariance matrix

7 Source: Lo (1998)

Page 68 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Symmetrized-coherency matrix

Where ... indicates a temporal or spatial ensemble averaging. Symmetrized cases

are presented by (HV=VH).
The "objective of the incoherent target decomposition theory" is to express the
average-scattering mechanism as the sum of independent elements to associate a
physical mechanism with each component. Target-scattering decomposition permits
the extraction of target-characteristic information, provided that the decomposition
satisfies the general requirement of being robust under a change of wave
polarization basis (that is, roll invariant) (Touzi, 2007)". The angle ϕ in Figure 24
can be calculated and its effect compensated (Touzi, 2007).
There are several types of incoherent decomposition, but each can be distinguished
by two broad classes: the physical and the mathematical incoherent
decompositions.

Physical incoherent decomposition

The Freeman-Durden decomposition (PSFREDUR) corresponds to a model-based
decomposition of the coherency matrix developed for forested areas. The author
hypothesized that the total power—that is, the span—could be decomposed as the
sum of three scattering mechanisms:
• Odd: corresponding to a slightly rough surface modeled as a first-order
Bragg surface scattering. This component contains the contribution of all
single-bounce or odd-bounce scatterers. For example, a trihedral will
produce a strong contribution to the slightly rough surface component.
• Double-bounce: modeled as a dihedral corner reflector and representing
the ground-tree-trunk scattering (enhanced when the forest is inundated).
• Volume: modeled as the contribution from a cloud of randomly oriented
cylinder-like (dipole) scatterers that represent the volume scattering from
a forest canopy.
Today, this decomposition is used to decompose (interpret) the scattering of many
environments, such as urban areas, wetlands, and agricultural lands.

Mathematical-incoherent decomposition
The mathematical decompositions are not based on a physical model; however,
they can be applied to the analysis of all kinds of land-use and land-cover classes.
The current mathematical decompositions are based on an eingenvector
decomposition of the coherency matrix, which is analogous to a principal-
component analysis (PCA).

PCI Geomatics Page 69

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

The eigenvector decomposition, while mathematically unique, is not necessarily a

straightforward interpretation (when compared to a physical decomposition). This is
because there is no guarantee that the eigenvectors will represent any known
physical scattering mechanism directly (van Zyl and Kim, 2011).
In this module, the results of two mathematical incoherent decompositions will be
examined: Cloude-Pottier and Touzi.
The Cloude-Pottier eigen decomposition proposes an unsupervised classification
of the coherency matrix. Many parameters representing the scattering
characteristics of a distributed target are determined from the resulting
eigenvectors and eigenvalues. Among the parameters, the average alpha (  ) and
beta (  ) angles are the most used. While ̅ corresponds to the average scattering
α
mechanism of an extended target, β̅ correspond to its orientation around the radar
LOS.
Determined from the eigenvalues, the entropy (H), and the anisotropy (A) complete
the description of the scattering type. The entropy (H) is used to characterize the
scattering randomness. A low entropy value (H ≈ <0.3) indicates the presence of a
dominant and deterministic scattering mechanism (dominant eigenvalue), which
means that only one scattering mechanism is contributing significantly to  .
As H increases, the randomness of the scattering increases, which indicates a
greater mix among various scattering mechanisms. At H=1, the scattering is
random (completely depolarized) corresponding to pure white noise. The
polarimetric anisotropy (A) measures the relative importance of the second and
third eigenvalues. A high anisotropy value indicates the presence of a second
significant scattering mechanism, along the dominant scattering mechanism.
The average alpha angle and the entropy are often combined to form the H /α ̅
plane, which leads to an unsupervised classification of a polarimetric image into the
following nine zone classes:
• Zone 9: Low-entropy surface scatter. Low-entropy scattering
processes in which α
̅ values less than 42.5 degrees occur. These include
GO (geometrical optics) and PO (physical optics) surface scattering—
Bragg surface scattering and specular scattering phenomena which do not
involve 180-degree phase inversions between HH and VV. Physical
surfaces, such as water at L and P-bands, sea-ice at L-band, and very
smooth land surfaces, all fall into this category.
• Zone 8: Low-entropy dipole scattering. Strongly correlated
mechanisms that have large imbalances between HH and VV in amplitude.
An isolated dipole scatterer would appear here, as would scattering from
vegetation with strongly correlated orientation of anisotropic scattering
elements. The width of this zone is determined by the ability of the radar
to measure the HH/VV ratio; that is, on the quality of the calibration.
• Zone 7: Low-entropy, multiple-scattering events. Corresponds to
low-entropy double, or even-bounce scattering events, such as those
provided by isolated-dielectric and metallic-dihedral scatterers. These are
characterized by α
̅> 47.5 degrees.

Page 70 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

• Zone 6: Medium-entropy surface scatter. Reflects the increase in

entropy H due to changes in surface roughness and canopy-propagation
effects. In surface-scattering theory, the entropy H of low-frequency
Bragg scattering is zero. As the roughness and correlation length of a
surface increases, its entropy H also increases. Furthermore, a surface
cover comprising oblate-ellipsoidal scatterers (leafs or discs for example)
generates an entropy 0.6 < H < 0.7.
• Zone 5: Medium-entropy vegetation scattering. Moderate entropy H,
but with a dominant dipole-type scattering mechanism. The increased
entropy H is due to a central statistical distribution of orientation angles.
Such a zone includes scattering from vegetated surfaces with anisotropic
scatterers and moderate correlation of scatterer orientations.
• Zone 4: Medium-entropy multiple scattering. Accounts for dihedral
scattering with moderate entropy H. This occurs, for example, in forestry
applications, in which double-bounce mechanisms happen at P and L
bands following propagation through a canopy. The effect of the canopy is
to increase the entropy H of the scattering process. A second important
process in this category is urban areas, in which dense packing of
localized scattering centers can generate moderate entropy H with low-
order multiple scattering dominant. The boundary between zones 4, 5, 6,
and 1, 2, 3 is set as 0.9. This is chosen on the basis of the upper limit for
surface, volume, and dihedral scattering before random distributions
apply.
• Zone 3: High-entropy surface scatter. Not part of the feasible region
in H-  space (surface scattering with entropy H>0.9 cannot be
distinguished). This is a direct consequence of the increasing inability to
classify scattering types with increasing entropy. It is included to reinforce
the idea that increasing entropy really does limit the ability to use
polarimetric behavior to classify targets. Radar polarimetry will then be
applied most successfully to low-entropy problems.
• Zone 2: High-entropy vegetation scattering. High-entropy volume
scattering arises when α̅ = 45° and H > 0.9. This can occur for single
scattering from a cloud of anisotropic needle-like particles, or for multiple
scattering from a cloud of low-loss symmetric particles. In both cases,
however, the entropy H lies above 0.9, where the feasible region of H/ α ̅
plane is shrinking rapidly. Scattering from forest canopies lies in this
region, as does the scattering from some types of vegetated surfaces with
random highly anisotropic scattering elements. The extreme behavior in
this class is random noise; that is, no polarization dependence, a point
which lies to the extreme right in Figure 26.
• Zone 1: High-entropy multiple scattering. ln the H> 0.9 region, it is
still possible to distinguish double-bounce mechanisms in a high-entropy
environment. Again, such mechanisms can be observed in forestry
applications or in scattering from vegetation with a well-developed branch
and crown structure.8

8 Source: Cloude, S.R., Pottier, E. (1997). An Entropy based Classification scheme for Land
Applications of Polarimetric SARs. IEEE Transactions on Geoscience and Remote Sensing.
Vol.35, no.2, p.68–78.

PCI Geomatics Page 71

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

̅ classification plane9
Figure 26. H-α
The Touzi decomposition (PSTOUZIDEC) is based on the characteristic
decomposition of the coherency matrix. With reciprocal targets, the characteristic
decomposition leads to the representation of the coherency matrix as the
incoherent sum of three single scatterers, each weighted by its normalized and
positive eigenvalues (λi  i , i =1, 2, 3).

The Touzi decomposition uses the Touzi scattering-vector model to represent each
coherency eigenvector with unique target characteristics. Each coherency
eigenvector is characterized uniquely by five independent parameters. Scattering
type is described with a complex entity, whose magnitude ( αsi ) and phase (Φαsi )
characterize the magnitude and phase of target scattering. The helicity (  i )
characterizes the symmetric-asymmetric nature of target scattering.
The Touzi decomposition is similar to the Cloude-Pottier decomposition except that:
• It does not proceed to the weighted sum of each eigenvector’s parameters
by their respective eigenvalues
• It takes into account the polarimetric phase (Φαsi )

• It compensates for symmetric-asymmetric nature of target scattering (  i )

The orientation angle ( i ) is the conventional Huynen tilt angle.

9 Source: Lee et Pottier, 2009

Page 72 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Target scattering can be characterized by a deep analysis of the parameters of the

three eigenvectors. Touzi and others have shown that the analysis of dominant
scattering parameters can lead to efficient wetland classification.
The psi angle compensates for the rotation of an object in the plane perpendicular
to the radar LOS to provide an unbiased estimation of the scattering type (alpha
angle). The Touzi decomposition also compensates for a lack of symmetry (of an
object), in a plane parallel to the radar LOS based on the value stored in the
helicity parameter.
A symmetric object will correspond to a helicity value close to zero, while an
asymmetric object will correspond to higher positive and negative helicity values.
Target asymmetry will introduce a helix-like type of scattering.

Polarimetric decompositions and other algorithms for

scattering characterization
The following table shows the available decomposition algorithms and other
algorithms for characterizing scatter in Geomatica. Additional information and
references can be found in the Help topic of each algorithm.

Table 9. Polarimetric decompositions and other algorithms for characterizing

scatter in Geomatica

Code Input/matrix type Description

[PSEABA] ▪ Covariance (c4r6c or Seven parameters:
Entropy, alpha, C3r3c), Coherency
▪ Mean scattering type (  ), the alpha
beta, and (t4r6c or T3r3c), or
anisotropy Kennaugh (k16r or angle

Cloude-Pottier K9r) ▪ Target orientation (  ), the beta

(1997) See note A. angle

▪ Anisotropy ( A)
▪ Entropy ( H )

▪ Eigenvalues ( 1 ,  2 ,  3 )

[PSFREDUR] Recommended: Three parameters: Power contributions

Freeman- ▪ Covariance (c4r6c or due to:
Durden C3r3c), Coherency ▪ Double-bounce scattering
decomposition (t4r6c or T3r3c), ▪ Volume scattering, which includes
Freeman and Kennaugh (k16r or multiple scattering
Durden (1998) K9r)
▪ Rough-surface scattering (odd number
Other: of bounces)
▪ Scattering (s4c or S3c)

PCI Geomatics Page 73

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Code Input/matrix type Description

[PSG4U2] Recommended: Four parameters: Power contributions
General four- ▪ Covariance (c4r6c or due to:
component C3r3c), Coherency ▪ Double-bounce scattering
decomposition (t4r6c or T3r3c), ▪ Volume scattering, which includes
Sing and al, or Kennaugh (k16r or multiple scattering
(2013) K9r)
▪ Rough-surface scattering (odd number
Other: of bounces)
▪ Scattering (s4c or S3c) ▪ Helix scattering
Note: Helix scattering is not calculated
with s4c or S3c.
[PSPHDW] Highly recommended: Four parameters: Power contributions
POLSAR PHDW ▪ Covariance (c4r6c or due to:
decomposition C3r3c), Coherency ▪ Plate scattering
Yamaguchi and (t4r6c or T3r3c), or ▪ Diplane scattering
al. (2008) Kennaugh (k16r or
K9r) ▪ Wire scattering
Other ▪ Helix scattering
▪ Scattering (s4c or S3c)
[PSKROG] Recommended: Three parameters: Power contributions
Krogager ▪ Covariance (c4r6c or due to:
decomposition C3r3c), Coherency ▪ Spherical scattering
Krogager (t4r6c or T3r3c), or ▪ Diplane scattering
(1998) Kennaugh (k16r or
K9r) ▪ Helical scattering
▪ Scattering (s4c or
S3c).
[PSTOUZIDEC] ▪ Covariance (c4r6c or 15 parameters: The dominant,
POLSAR Touzi C3r3c), Coherency secondary and tertiary:
decomposition (t4r6c or T3r3c), or ▪ Symmetric scattering types
Kennaugh (k16r or
Touzi (2007)
K9r)
( s1 ,  s 2 ,  s 3 ) that is, the alpha
angles
See note A.
▪ Symmetric scattering types phases
( as1 ,  as2 ,  as3 )
▪ Target orientation angle
( 1 ,  2 ,  3 ) , that is, the psy angles
▪ Target helicity ( 1 , 2 , 3) angles,
that is, the tau angles
▪ Eigenvalues (1 , 2 , 3 )

Page 74 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Code Input/matrix type Description

[PSTOUZIDIS] ▪ Covariance (c4r6c or Four parameters:
POLSAR Touzi C3r3c), Coherency ▪ Maximum polarization response
discriminators (t4r6c or T3r3c), or
Kennaugh (k16r or ▪ Minimum polarization response
(Touzi and al,
1992) K9r) ▪ Touzi anisotropy
▪ Difference between minimum and
maximum polarization response
[PSPOLDIS] ▪ Covariance (c4r6c or 16 parameters:
POLSAR C3r3c), Coherency ▪ Maximum and minimum degree of
discriminators (t4r6c or T3r3c), or polarization
Kennaugh (k16r or
(Touzi and al, K9r) ▪ Maximum intensity of the completely
1992) polarized component
▪ Orientation (psi) and ellipticity (chi)
angles at the maximum of the
completely polarized component
▪ Minimum intensity of the completely
polarized component
▪ Orientation (psi) and ellipticity (chi)
angles at the minimum of the
completely polarized component
▪ Maximum and minimum intensity of
the completely unpolarized component
▪ Maximum and minimum of the
received power
▪ Maximum and minimum of the
scattered intensity
▪ Coefficient of variation
▪ Fractional polarization
[PSSSCM] ▪ Scattering (s4c or S3c) Six parameters:
Symmetric ▪ Degree of coherence (Psym)
scattering
characterization ▪ Scattering vector direction (η)
Touzi and ▪ Phase difference (φ) on the trihedral-
Charbonneau dihedral basis
(2002) ▪ Target coordinates psy (ψc) and chi
(Χc) on the Poincaré sphere. Used to
identify the scattering type.
▪ Rotation angle (ψ) of the maximum
symmetric scattering component
[PSVANZYL] ▪ Covariance (c4r6c or One parameter:
Unsupervised C3r3c), Coherency ▪ A classified image in which each pixel
van Zyl (t4r6c or T3r3c), or is classified according to a single
classification Kennaugh (k16r or dominant scattering type:
van Zyl (1989) K9r) - Odd number of reflections, or
- Even number of reflections, or
- Diffuse (volume), or
- Nonclassifiable

PCI Geomatics Page 75

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Code Input/matrix type Description

A: When required, the input data set is converted by this algorithm to the symmetrised coherency matrix,
T3r3c, to perform the calculations.

Page 76 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Lesson 3.2: Freeman-Durden decomposition

In this lesson, you will:
• Ingest a RADARSAT-2 fully polarimetric image using the SARINGEST
algorithm
• Apply a boxcar filter on the ingested SAR data with the PSBOXCAR
algorithm
• Use the PSFREDUR algorithm to perform the Freeman-Durden
decomposition
• Use the EASI Modeling feature in Focus to create a normalized version of
the Freeman-Durden parameters

Ingesting a fully polarimetric image

To ingest a fully polarimetric image, you use the SARINGEST algorithm.

To open the SARINGEST algorithm

1. In Focus, click the Tools menu, and then click Algorithm Librarian.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find what box, type SARINGEST, and then click Find Next.
4. Double-click SARINGEST.
The SARINGEST Module Control Panel window appears.
5. Under Output Ports, in the Output: SAR Dataset entry, select
Untitled.pix, click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
6. Enter FQ29_20100507sig.pix as the output file name.
Note: If you already ingested this file, jump to the next procedure,
Applying the boxcar filter.
7. Click the Input Params 1 tab, beside the File box, click Browse, and
then select the product.xml file.
8. When prompted to select a calibration type, click Sigma Nought, and
then click OK.
9. Run the SARINGEST algorithm.

Applying the boxcar filter

You apply a boxcar filter using the PSBOXCAR algorithm.

To set up the parameters for PSBOXCAR

1. Under Input Ports, expand Input: Input polarimetric SAR image.
2. Click Browse, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder, and then select FQ29_20100507sig.pix.

PCI Geomatics Page 77

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

3. Select each of the channels in the list.

4. Under Output Ports, in the Output: Output polarimetric SAR filtered
image entry, select Untitled.pix, and then click Browse.
The File Selector window appears.
5. In the File name box, type FQ29_20100507sig_PSBOXCAR_7.pix,
and then click Save.
6. Click the Input Params 1 tab.
7. In the Horizontal filter size list and the Vertical filter size list,
respectively, select 7.
8. Click Run.

Performing the Freeman-Durden decomposition

You apply a Freeman-Durden decomposition using the PSFREDUR algorithm.

To open the PSFREDUR algorithm

To set up the PSFREDUR algorithm

1. Under Input Ports, click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
2. Select the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Select each raster of the imported file.
4. Under Output Ports, click Browse, and then enter
FQ29_20100507sig_PSBOXCAR_7 _PSFREDUR.pix as the name of
the output file.
5. Click Run.

To view the result of PSFREDUR

1. In Focus, click the File menu, and then click Open.
2. Open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
3. Select the FQ29_20100507sig_PSBOXCAR_7 _PSFREDUR.pix file,
and then click Open.

Page 78 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Exercise 1: Examine an RGB composite made of the Freeman-Durden parameters,

and each parameter in grayscale. You can compare the Freeman-Durden
decomposition to the Landsat-5 mosaic provided with this course:
~\SAR_Training\Landsat\Flevoland_L5_20100906_p198_r23r24.pix

Exercise 2: Locate and examine some features that produce a strong scattering in
only one of the components, in two components, or in all components.
For example, locate the bright area centered on 1789P, 3073L (671 036E, 5 823
936N). This industrial area produces a strong double-bounce scattering, as
predicted by the theory, but many buildings also seem characterized by a stronger
volume scattering compared to the surrounding forest, which is less intuitive. This
is because the Freeman-Durden decomposition directly uses the HV channel to
calculate the volume contribution of the total backscattered signal.
In urban areas, many scattering processes can create a strong return in HV, which
are not volumic by nature. Among these are multiple scattering (the addition of
many scattering mechanisms in one cell of resolution), a building not aligned with
the radar LOS (orientation effects), or a nonsymmetrical object.
Interpreting the Freeman-Durden decomposition can be made easier by normalizing
the radiometry to calculate the fraction of the total scattered power associated with
each component.

Producing a normalized version of the Freeman-Durden

decomposition
You can produce a normalized version of the Freeman-Durden decomposition in
Focus with the EASI Modeling feature.

To add four new 32-bit real channels

1. In Focus, click the Files tab, right-click the
FQ29_20100507sig_PSBOXCAR_7 _PSFREDUR.pix file, point to
New, and then click Raster Layer.
2. Add four new 32-bit real channels.

To normalize the decomposition

1. In Focus, click the Tools menu, and then click EASI Modeling.
2. In the Input File list, click the FQ29_20100507sig_PSBOXCAR_7
_PSFREDUR.pix file.
3. In the box below the Input File list, enter the code as shown in the
following figure.
Note: Comments indicated by "!" can be omitted
The code can be loaded from:
~\SAR_Training\EASI_script\Module3_FreDur_Norm.eas

PCI Geomatics Page 79

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

4. Click Run.
Both versions can now be compared; each has its own utility. The power version is
often preferred if the components are to be used in a classification, while the
normalized version helps to better understand the composition of backscattered
power in mixing of scattering mechanisms.

Power version (RGB) Normalized version (RGB)

R: Double G: Volume B: R: % Double G: %Volume Total power (span)
Surface B: % Surface

Page 80 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

% Double % Surface % Volume

Power, Double Power, Surface Power, Volume

HH VV HV

Figure 27. PSFREDUR, Freeman-Durden decomposition results

PCI Geomatics Page 81

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Ingested a RADARSAT-2 fully polarimetric image using SARINGEST
• Applied a PSBOXCAR filter on the ingested SAR data
• Used PSFREDUR to perform the Freeman-Durden decomposition
• Used the EASI Modeling feature in Focus to create a normalized version
of the Freeman-Durden parameters

Page 82 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Lesson 3.3: Cloude-Pottier decomposition

In this lesson, you will:
• Use PSEABA to perform the Cloude-Pottier decomposition
• Use the EASI Modeling feature in Focus to analyze the Cloude-Pottier
parameters

Performing the Cloude-Pottier decomposition

To perform the Cloude-Pottier decomposition, you use the PSEABA algorithm.

To open the PSEABA algorithm

To set up the PSEABA algorithm

1. Under Input Ports, click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
2. Select the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Select each raster of the imported file.
4. Under Output Ports, click Browse, and then enter
FQ29_20100507sig_PSBOXCAR_7_PSEABA.pix as the name of the
output file.
5. Click the Input Params 1 tab, and then select the Retrieve
eigenvalues and eigenvectors check box.
6. Click Run.

Exercise 3: Examining the PSEABA decomposition results using thresholds.

Examine the Cloude-Pottier decomposition parameters. Remember that the entropy
is stored in channel 1 (%1), the anisotropy in channel 2 (%2), the alpha angle in
channel 3 (%3), and the beta angle in channel 4 (% 4).
1. Find the targets characterized by a double-bounce (  ≥ 55°).
2. Are these double-bounce targets characterized by low entropy? What is
the mean entropy of these targets (mean = 0.61±0.2, med = 0.63)?
3. Is the double-bounce scattering generally characterized by one dominant
or two scattering mechanisms (  ≥ 55° U A≥0.8)?
4. What is the mean entropy of  ≥ 55° U A ≥0.8 and  ≥ 55° U A< 0.8?

PCI Geomatics Page 83

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Alpha angle (  ) Beta angle (  ) Entropy (H)

Red=  ≥55° and A<0.8

Anisotropy (A) Red =  ≥55°
Green =  ≥55° and A≥0.8

Figure 28. Cloude-Pottier parameters and thresholding result

It appears that there are two classes of double-bounce scatterers: one class
appears in red and is characterized by only one dominant scattering mechanism
(H), and a second class appears in green and is the result of two scattering
mechanisms (with a high anisotropy). The entropy of two classes is 0.63±0.2 and
0.39±0.15, respectively.
Extract the forested areas that correspond, in theory, to a dipole scattering with
medium-to-high entropy (zones 5 and 2). Do you observe only forested areas in
these zones? Are there forested areas outside these zones?

Note The following script can be loaded from:

~\SAR_Training\EASI_script\Module3_ForestedAreas.eas.
Also, the Landsat-5 mosaic can be used to facilitate the
interpretation,
~\SAR_Training\Landsat\Flevoland_L5_20100906_p198_r
23r24.pix.

Page 84 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

To run this script, you must create two new bitmap layers by right-clicking the file
on the Files tab, and then selecting New > Bitmap Layer. This creates empty
bitmap segments 2 and 3.

Figure 29. EASI Modeling script to extract Cloude-Pottier zones 2 and 5

Figure 30. Green: Zone 5, Red: Zone 2

PCI Geomatics Page 85

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Green: Zone 5, Red: Zone 2 (detail) Landsat-5, R: tm4, G:tm3, B:tm2

(2010-09-06)

Figure 31. Cloude-Pottier zones 2 and 5

Page 86 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Lesson summary
In this lesson, you:
• Used PSEABA to perform the Cloude-Pottier decomposition
• Used the EASI Modeling feature in Focus to analyze the Cloude-Pottier
parameters

PCI Geomatics Page 87

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Lesson 3.4: Touzi decomposition

In this lesson, you will:
• Use the PSTOUZIDEC algorithm to perform the Touzi decomposition
• Compare the Touzi decomposition to the Cloude-Pottier decomposition

• Compare the Touzi  s1 phase to the  hh −  vv phase difference

Performing the Touzi decomposition

To perform the Touzi decomposition, you use the PSTOUZIDEC algorithm.

To open the PSTOUZIDEC algorithm

To set up the PSTOUZIDEC algorithm

1. Under Input Ports, click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
2. Select the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Select each raster of the imported file.
4. Under Output Ports, click Browse, and then enter
FQ29_20100507sig_PSBOXCAR_7_PSTOUZIDEC.pix as the name of
the output file.
5. Click the Input Params 1 tab, and then in the Angle Units list, click
Degrees.
6. Click Run.

Exercise 4: Examine the Touzi decomposition parameters. Unlike Cloude-Pottier

decomposition, the Touzi decomposition does not proceed to the weighted mean of
each parameter (  si ,  s1 , i ,  i ) by their respective eigenvalues (  i , i=1,2,3).

Why do some of the parameters of the second component (i=2) and most of the
parameters of the third component (i=3) look noisy?
The Landsat-5 mosaic can be used to facilitate the interpretation:
~\SAR_Training\Landsat\Flevoland_L5_20100906_p198_r23r24.pix.

Page 88 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Note Some parameters are best viewed in pseudocolor,

especially the angle parameters (  si , i and  i ) and the
phase  s1 .

To view a parameter in pseudocolor

1. In Focus, click the Files tab.
2. Expand the Raster list of the file containing the parameter or parameters
to be viewed in pseudocolor.
3. Right-click the parameter, point to View, and then click As Pseudocolor.
4. Click the Maps tab.
5. Right-click the parameter, and then click PCT Editing.

▪   s1 -90 90 PCT Editing – Stepped

▪ i -45 45 PCT Editing – Stepped

▪  s1 0 90 PCT Editing – Smooth

 s1  s2  s3

  s1   s1 (in pseudocolor)  s 2

PCI Geomatics Page 89

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

1 1 (pseudocolor) 2

1 1 (pseudocolor)
2

1 2 3
(min:0.37, max:0.99) (min:0.23, max:0.48) (min:0.10, max:0.29)

Figure 32. Touzi decomposition parameters (selection)

Page 90 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Note The parameter values are important, but so too are their
spatial distribution (the texture). In some analysis, the
sign of the orientation (ψi ) or helicity (τi ) does not
matter. Calculating their absolute values (  1 |ψi |,  1 )
might facilitate interpretation.

Exercise 5: Compare the Touzi  s1 as1 to the average Cloude-Pottier  .

Find areas in the Flevoland data set where Touzi αs1 differs from Cloude-Pottier  .
Can you explain the observed differences?
Calculate the global Touzi alpha angle (α
̅s ). Do you still observe differences with the
Cloude-Pottier α
̅? If yes, why?

PCI Geomatics Page 91

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

 s1  sg Cloude 
Touzi Touzi

Touzi
 s1
, detail
 sg Cloude  , detail
Touzi , detail

Figure 33. Comparison between Touzi  s1 ,  sg and Cloude 

Page 92 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Exercise 6: Find areas where  1 is low or high. What kinds of land use and land
cover are characterized by high-and-low helicity values? Use EASI Modeling to
create bitmaps containing high helicity with different thresholds, such as 5, 10, and
25 degrees.

Tip You can load the preceding EASI script from:

~\SAR_Training\EASI_script\Module3_Helicity
_thresholding.EAS

Exercise 7: Compare the Touzi dominant orientation angle 1 with the Cloude-
Pottier average beta angle  .

Note In theory, 1 and β̅ both measure the orientation of a

target from the radar line of sight (LOS) (angle  in
Figure 24). While ψ1 is distributed between -45 and 45
̅ is distributed between zero and 90 degrees
degrees, β
  2 • 1 .

Both are also sensitive to bare-soil roughness. Bare soil with a high roughness
tends to produce higher 1 (or  ) values, while smooth surfaces result in 1 (or
 ) values centered around zero degrees.

PCI Geomatics Page 93

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Exercise 8: Locate the urban or industrial areas in the Flevoland image, and
observe the main street orientation from the radar LOS. What are the average 1
and  values when the streets are parallel to the LOS? What are the average  1
and  values when the main streets orientation departs from the LOS?
The following vector file can help to locate the urban area and interpret the relation
between 1 ,  and the LOS: ~\SAR_Training\Vectors\Flevoland_Streets.pix.

Exercise 9: Find some bare fields and observe the 1 and  values.

TOUZI PHASE   s1
The Cloude-Pottier decomposition does not include a phase parameter, as does the
Touzi decomposition with  s1 . This phase is similar, but not equivalent, to the
 hh −  vv phase difference. To better understand the  s1 phase, the  hh −  vv
phase difference will be generated with the PSPHDIFF algorithm.

Generating the  hh −  vv phase difference

You generate the  hh −  vv phase difference using the PSPHDIFF algorithm.

To open the PSPHDIFF algorithm

To set up the PSPHDIFF algorithm

1. Under Input Ports, click Browse, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder, and then select the FQ29_20100507sig_PSBOXCAR_7.pix file.
2. Select each raster of the imported file.
3. Under Output Ports, expand Output: Output phase difference raster,
select Untitled.pix, and then click Browse.
The File Selector window appears.
4. Open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
5. Enter FQ29_20100507sig_PSBOXCAR_7_PSPHDIFF_HHVV.pix as
the name of the output file.
6. Click the Input Params 1 tab, and then in the First input polarization
list, select HH, and in the Second input polarization list, select VV.

Page 94 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

7. In the Angle Units list, select Degrees.

8. Click Run.

Exercise 10: Compare the Touzi  s1 and the  hh −  vv phase difference.

Before the widespread use of polarimetric decompositions, the  hh −  vv phase

difference was sometimes used as a (crude) way to identify the main scattering
mechanisms (see Ulaby et al., 1987). Single-bounce scattering occurs around zero
degrees ([-15, 15], while double-bounce scattering occurs at both ends of the
ϕhh -ϕvv distribution ([-180,100] U ([-100,180]).
The Touzi phase  s1 is similar, but not equivalent to  hh −  vv . The main
differences are:
• Different dynamic range
• A phase representation in the dihedral-trihedral basis instead of the HH-
VV basis

• An independence of  s1 from a polarization basis

For example, if the signal hit a tilted surface or an object with a complex geometry,
the phase of the maximum polarized (backscattered) component will no longer be
in the HH-VV basis. The  hh −  vv will still provide information about the target
structure, but this information might be biased and not be representative of the
target’s real structure. Since the departure of the backscattered signal from the
H-V basis (satellite transmitting antenna) is registered in 1 (or  ) angle, it is
possible to use this information to retrieve an unbiased phase to characterize the
target’s structure.

PCI Geomatics Page 95

Module 3: Polarimetric decompositions Processing SAR data with Geomatica Banff

Exercise 11: Explore the differences between Φαs1 and ϕhh -ϕvv using the
absolute value of  1 as a guide. The anisotropy (A) can also be used.

  s1  hh −  vv  hh −  vv
Touzi ,
Blue: Single-bounce
scattering
Red: Double-bounce
scattering

  s1  hh −  vv  hh −  vv
Touzi , detail , detail ,
Blue: Single-bounce
scattering
Red: Double-bounce
scattering

Figure 34. Comparison between Touzi  s1 phase and  hh −  vv phase difference

Page 96 PCI Geomatics

Processing SAR data with Geomatica Banff Module 3: Polarimetric decompositions

Lesson summary
In this lesson, you:
• Used PSTOUZIDEC to perform the Touzi decomposition
• Compared the Touzi decomposition to the Cloude-Pottier decomposition

• Compared the Touzi  s1 phase to the  hh −  vv phase difference

PCI Geomatics Page 97

Module 4: Using the SAR
Polarimetry Target Analysis tool

About this module

Module 4 has three lessons:
• Lesson 4.1: Introduction to SPTA and the target-selection mode
• Lesson 4.2: Coherent target analysis in SPTA
• Lesson 4.3: Incoherent target analysis in SPTA

Lesson 4.1: Introduction to SPTA and the target-

selection mode
In this lesson, you will:
• Understand the different target-selection modes.
• Draw a target in SPTA
• Understand the target manager
• Learn how to export and import a target using SPTA
With Geomatica SPTA—the SAR polarimetry target analysis tool—you can select
targets in a polarimetric SAR scene, draw a target (or load an existing one),
perform a variety of polarimetric operations on it, and display the results
numerically and graphically.
You can use SPTA to analyze all types of SAR data products. However, to use the
full functionality of SPTA, a fully polarimetric SAR data set in complex format is
required.

To start SPTA
• On the Geomatica toolbar, click SPTA.
The SAR Polarimetry Target Analysis and Target Selection windows
appear.

PCI Geomatics Page 99

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

To open an image
1. In SPTA, click the File menu, and then click Open.
The File Selector window appears.
2. Open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
3. Select the FQ29_20100507sig.pix file, and then click Open.
A pop-up window appears, prompting you to select a georeferencing
system whether from the file or the math model (RPCs).
4. Click Select File.
A second pop-up window appears, prompting you how to display the
image.
5. Click North Up, and then click OK.
The FQ29_20100507sig.pix file is displayed in the Target Selection
window with channels 1, 2, and 4 mapped to RGB, respectively.

Page 100 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Beside RGB, you can change the channels displayed by selecting the channel you
want beside R, G, and B, respectively.
On the toolbar you can:

• Select a different enhancement ( ) or adjust the image contrast ( )

or brightness ( ).
Selecting a different enhancement or modifying the image brightness and
contrast will not change the computed statistics for a particular target.

• Pan ( ) and zoom ( ) the displayed image.

PCI Geomatics Page 101

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

• Change the image interpretation ( ); that is, to display the channels in

intensity, amplitude, or decibels. You can also display only the phase or
the real or imaginary components.

Understanding target-selection modes

There are three target-selection modes available in the SAR Polarimetry Target
Analysis window:
• Pixel plus clutter estimation region: you can choose a single target
pixel at the cursor position and four clutter-estimation regions at the
corners of an overall square region that it is centered on, but excludes,
the target pixel and the side-lobe pixels. Overall size and Gap size can
be modified to estimate the properties of the coherent target against a
larger clutter.
Square neighbor: you can choose a target that includes all pixels within
a square region that is centered on the cursor position.
• Arbitrary region: you can choose either a rectangular or a polygonal
target that includes all pixels within the target.
The selection mode determines the target type and, therefore, the polarimetric
parameters that can be computed. For example, Pixel plus clutter estimation
region determines a point target (you are analyzing the response of the center
pixel in relation to the clutter pixels). Square neighborhood and Arbitrary
region delineate surface targets.
Depending on the type of target, several polarimetric parameters are available. For
example, the symmetric-scattering parameters are only applicable to point targets,
whereas parameters that are applicable to surface targets or require an averaging
of the data (such as entropy) are only available when surface targets (Square
neighborhood or Arbitrary region) are available.
The polarimetric parameters that can be computed also depend on the data set.
Polarimetric parameters that require multilook will be disabled for a single-look
complex image and, similarly, those that require a fully polarimetric data set (such
as polarimetric decompositions) are unavailable with single-polarimetric or dual-
polarimetric data.

To draw a target
1. In SPTA, in the Target Selection window, pan or zoom to the area of
interest, as applicable.
2. In the SAR Polarimetry Target Analysis window, select a target-selection
mode.
3. Draw a target over the area you want.
If the Arbitrary region – polygon option is selected, double-click to
close the shape.
The new target is added to the list in the Target Manager window.

Page 102 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

A) Point target location B) Overall size: 11 ( )

Coordinates: 1380P, 2610L Gap size:3 ( )

C) Overall size: 11 D) Overall size: 11 E) Overall size: 11

Gap size: 1 Gap size:3 Gap size: 5

F) Overall size: 17 G) Overall size: 17 H) Overall size: 17

Gap size: 1 Gap size: 3 Gap size: 5

Figure 35. Target selection mode: pixel plus clutter estimation region

Understanding Target Manager

SPTA keeps a record of the target modes you apply during a session. In the Target
Manager window, you can save one or more target records to a file. You can also
import a saved target, and then apply it to your imagery.
Each time you draw a new target, it is stored temporally in the Target Manager
window. A short description is provided (target 1, target 2...) along with the
selection mode, number of samples it contains, and image coordinates (Figure 36).

Note You can edit the default description by clicking a target

description, and then typing a new value.

PCI Geomatics Page 103

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Figure 36. Target Manager window

Note Targets are not saved automatically; on exiting SPTA, any

unsaved targets will be lost.

To export (save) a target

1. In the Target Manager window, in the Target column of the table, select a
target.
2. Click Export.
The Save Target As window appears.
3. Select a location and enter a file name for the file.
4. Click Save.
The file is saved to disk as a vector file.

Note The exported target (vector) will be in the same

geographic projection (if any) as the image on which it
was drawn. You can then, for other images, open the
target in SPTA or Focus.

Importing and using a target

In the Target Manager window, you can import targets that have been defined on
another image and set one as the current target in a new image. If an imported
target is outside of the bounds of the imagery displayed in SPTA, it cannot be
selected. To be recognized by SPTA, point targets must have the target-type, total-
size, and gap-size attributes for each point.

To import a target for use

1. In the SAR Polarimetry Target Analysis window, click the File menu, and
then click Target Manager.
2. In the Target Manager window, click Import.
3. In the Import Targets window, open the folder containing the target file
you want to import.
4. Select the file, and then click Open.

Page 104 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

5. In the Target Manager window, click the target you want to apply, and
then click Set as Current.
The target displays in the imagery in the Target Selection window.

PCI Geomatics Page 105

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Lesson 4.2: Coherent target analysis in SPTA

In this lesson, you will:
• Select coherent targets in SPTA
• Analyze the scattering characteristics of a coherent target
• Automatically detect coherent targets
The incoherent target analyses are restricted to single-look matrix type. Here, you
will use the s4c format (single-look, nonsymmetrized scattering matrix).

To select a coherent target

1. On the toolbar of the Target Selection window, click Pan and Zoom, as
applicable, to locate a point target.
2. In the SAR Polarimetry Target Analysis window, under Target Selection
Mode, click Pixel plus clutter estimation region.
The default values will suffice for this lesson.
3. In the SAR Polarimetry Target Analysis window, select Numerical
Output, and then do the following:
▪ Under Coherent Target Decomposition, select the Huynen and
Cameron parameters check box.
▪ Under Symmetric scattering parameters, select the Symmetric
scattering characterization check box.
4. Select the target located at 411.5P, 4236.6L (raster coordinates), and
then click Compute.
A series of statistics characterizing the scattering properties of the target
appears in the Results box.

Note The statistics are calculated only if the selected target is

coherent or symmetric according to the values specified
under Coherent Target Decomposition and
Symmetric scattering parameters. To include more
targets in the coherent target analysis, you can decrease
the default values; however, this may produce statistics
that are less meaningful.

Page 106 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Figure 37. Target Selection Mode, Coherent Target Decomposition, and Symmetric
scattering parameters
A second target, located at 1585.5P, 4437.5L can be selected. If an Overall size of
5 and a Gap size of 1 are used, the target does not appear to be symmetric. Set
Overall size to 9 and Gap size to 3, select the target again, and then click
Compute.
The target is now defined against a larger clutter corresponding to a flat ground
surrounding what appears to be a pylon. The symmetric scattering characteristics
can now be estimated.

PCI Geomatics Page 107

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Exercise 1: Compare the statistics of the two selected targets (see Figure 38)
Target 2 – Overall size = 5, Gap Target 2 – Overall size = 9, Gap size=3
size=1

Matrix Representation is s4c Matrix Representation is s4c

==== Huynen and Cameron parameters ==== ==== Huynen and Cameron parameters ====
Target is coherent Target is coherent
Eigenvalues: Eigenvalues:
-1365.166016 + 1757.697144 i, -1365.167969 + 1757.695557 i, -1365.166016 + 1757.697144 i, -1365.167969 +
Eigenvectors: 1757.695557 i,
-14.322257 + 0.000000 i, 0.999996 + 0.000000 i, Eigenvectors:
-1.000005 + 0.000000 i, 14.322258 - 0.000000 i, -14.322257 + 0.000000 i, 0.999996 + 0.000000 i,
Maximum returned power density: 225.422 (linear) -1.000005 + 0.000000 i, 14.322258 - 0.000000 i,
Characteristic angle: 45 deg Maximum returned power density: 225.422 (linear)
Absolute phase of the target: -127.836 deg Characteristic angle: 45 deg
Target skip angle: -1.59984e-05 deg Absolute phase of the target: -127.836 deg
Maximum polarisation orientation relative to the horizontal: - Target skip angle: -1.59984e-05 deg
89.999985 Maximum polarisation orientation relative to the horizontal: -
Maximum polarisation ellipticity: 0.000002504478 89.999985
Rotation angle for maximum symmetrical scattering component: Maximum polarisation ellipticity: 0.000002504478
45 deg Rotation angle for maximum symmetrical scattering component:
Degree of symmetry: 1.0000000 45 deg
Real Component of Co-polarised ratio: 1.0000000 Degree of symmetry: 1.0000000
Imaginary Component of Co-polarised ratio: 0.000000 Real Component of Co-polarised ratio: 1.0000000
Nearest elemental symmetric scattering mechanism: Trihedral Imaginary Component of Co-polarised ratio: 0.000000
Distance to nearest elemental symmetric scattering mechanism: Nearest elemental symmetric scattering mechanism: Trihedral
0.000000 Distance to nearest elemental symmetric scattering mechanism:
0.000000
==== Symmetric Scattering Characterization ====
Target is not symmetric
==== Symmetric Scattering Characterization ====
Degree of Coherence: 0.98805820.988058 deg
Scattering Vector Direction: 0 deg
Phase Difference: -0 deg
Target Sphere Angle (PSI): 0 deg
Target Sphere Angle (CHI): 0 deg
Rotation Angle: 45 deg

Figure 38. SPTA Target Selection window and extracted statistics

Page 108 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Note Coherent targets can be difficult to find by clicking

randomly on an image. In the SAR Radar Analysis folder
of the Algorithm Librarian you can use the PSWHITE or
PSSCM algorithm to help find the coherent targets of an
image. PSWHITE produces a bitmap representing the
coherent point targets. PSSSCM produces a bitmap
representing the coherent point targets, and an image file
containing six channels characterizing the scattering
properties of the coherent targets, where the first channel
represents the degree of coherence.

Bitmaps produced by the PSWHITE and PSSSCM algorithms can be viewed in Focus.

Exercise 2: Using the PSSSCM algorithm, produce a bitmap representing the

coherent point target.

To open the PSSSCM algorithm

To set up the parameters for PSSSCM

1. Under Input Ports, click Browse, and then select the
FQ29_20100507sig.pix file.
2. Under Input Ports, expand Input: Input polarimetric SAR image, and
then select each intensity raster file.
3. Leave Input Targets: Input coherent target bitmap unchanged.
4. Under Output Ports, expand OutputRaster: Output scattering
characteristics raster, select Untitled.pix, click Browse, and then
select the location in which to save the file.
5. In the File Name box, type FQ29_20100507sig_PSSSCM.pix, and
then click Save.
6. Under Output Ports, expand OutputBitmap: Output coherent targets
bitmap, select Untitled.pix, click Browse, and then select the location
in which to save the file.
7. In the File Name box, type FQ29_20100507sig_PSSSCM_CTmap.pix,
and then click Save.
8. Click the Input Params 1 tab, and then change the values of the
parameters according to those shown in Figure 39.
9. Click Run.

PCI Geomatics Page 109

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

The resulting bitmap can be opened in Focus over the FQ29_20100507sig.pix file.
Figure 40 and Figure 41 show the detected coherent point target using various
configurations of window size.
In comparison, the PSWHITE algorithm is less strict in its definition of a coherent
point target. This algorithm relies mainly on a threshold-based detection to
discriminate bright-point targets.

Figure 39. PSSSCM Module Control Panel window

Page 110 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Configuration Target 1 Target 2

No. 1
Parameters

WS=9
CWOW=15
CWCW=3
DST=0.8
DCH=0.8
SCRT=12

No. 2
Parameters

WS=15
CWOW=25
CWCW=15
DST=0.8
DCH=0.8
SCRT=12

Figure 40. Coherent targets detection using PSSSCM (HH channel in background)

Parameters Target 1 Target 2

WS=9
DT=12

Figure 41. Coherent targets detection using PSWHITE (HH channel in background)

PCI Geomatics Page 111

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Selected coherent targets in SPTA
• Analyzed the scattering characteristics of a coherent target
• Automatically detected coherent targets

Page 112 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Lesson 4.3: Incoherent target analysis in SPTA

In this lesson, you will:
• Define and select targets
• Produce numerical and graphical outputs for selected targets
• Analyze numerical and graphical outputs

Exercise 3: You are now ready to select some regions. From the results obtained
in Module 3, Lesson 4.1, and Lesson 4.2, the major land-use and land-cover classes
in the Flevoland region are readily apparent. For this analysis, you will select
regions corresponding to the following land-use and land-cover classes:
• Open water
• Urban area 1 [double-bounce dominated, aligned with the radar LOS]
• Urban area 2 (with volume, multiple scattering, or both)
• Forested areas
• Wetland (meadow, reeds)
• Agriculture (with strong HV, vegetated, rough or both)
• Agriculture (strong HH, double-bounce)
• Agriculture (smooth bare field)

To select regions
1. In the Target Selection window, draw a polygon over an area of open
water in the image.
2. In the Target Manager window, in the Description box, type
a_OpenWater.

3. To save the region as a vector file, click Export.

4. Open the folder in which you want to save the file, and then enter
a_OpenWater.pix as the name of the output file.
The region will be saved as a vector file.
5. Repeat steps 1 to 4 for the remaining land-use and land-cover classes.

PCI Geomatics Page 113

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Tip To get a meaningful analysis of incoherent targets, select

a homogenous region. Also, avoid selecting a region with
too many or too few samples.
All the target polygons are provided in
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29
_20100507\targets.

Figure 42. Location of land-use classes

Page 114 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Producing numerical output

Now that you have defined a set of targets, you can analyze their diffusion
properties with the numerous numerical output tools available in SPTA by
checking the corresponding parameters of interest. Note that the available options
vary depending on the selected target type (box A).
For an incoherent target represented by an arbitrary region, like the ones you have
defined in the previous section, the following options are available:
• Matrix Representations (box B)
• Parameters and Discriminators (box C)
• Channel Comparisons (box D)
After selecting a target and checking the output, click Compute (box F) to get the
numerical results.

To save the numerical-output results to disk

1. In SPTA, click the File menu, and then click Save Numerical Output.
2. In the File Name box, enter a name for the file, select a folder in which to
save the file, and then click Save.
The results can be saved to disk by copying them from the Results box to a
document of your choice (.txt, .doc, .docx).

Note All of the options in boxes B, C, and D have their

equivalent in the Radar Analysis toolbox (see Module 3),
which can be applied to the whole image instead of a
selected region.

PCI Geomatics Page 115

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Exercise 4: Select open water and forested areas and compare their scattering
characteristics by selecting some of the available numerical-output options. You will
use the targets defined in the previous lesson.

Figure 43. SAR Polarimetry Target Analysis window, Numerical Output options

Producing graphical output

SPTA provides three options to visualize your data. Each can be configured
(customized) in various ways:
• Histogram: examine the distribution of one element (or channel)
• Scatter plot: examine the relationship between two (2-D) or three (3-D)
elements
• Response plot: examine the polarimetric properties of one target

Page 116 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

Figure 44. Histogram, Scatter Plot, and Response Plot

PCI Geomatics Page 117

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Exercise 5: With the urban and the forest target, produce and compare the:
1. Copolarized response plot using the normalized scaling
2. Scatter plot using the HH and VV channels
3. Copolarized response plot with the ones in Figure 24 for canonical targets

To produce a response plot

To produce a scatter plot

1. In the Target Manager window, select the urban target
(b_UrbanArea_DB_LOS.pix), and then click Set as Current.
2. In the SAR Polarimetry Target Analysis window, under Graphical Output,
click the Scatter Plot tab.
3. In the Parameter 1 list, click Intensity, and in the Element 1 list, click
HH (or Element 1,1 of covariance matrix).
4. In the Parameter 2 list, click Intensity, and in the Element 2 list, click
VV (or Element 4, 4 of covariance matrix).
5. Leave the other options blank.
6. Click Plot.
Repeat the previous steps for the forest target (d_Forest.pix).

b_UrbanArea_DB_LOS d_Forest
response plot (normalized) response plot (normalized)

Page 118 PCI Geomatics

Processing SAR data with Geomatica Banff Module 4: Using the SAR Polarimetry Target Analysis tool

b_UrbanArea_DB_LOS d_Forest HH-VV

HH-VV scatter plot HH-VV scatter plot

Figure 45. Examples of polarimetric response and scatter plot for an urban and a
forest target

Note You can customize the graphical output using the

standard Gnuplot command lines
(http://www.gnuplot.info/). To customize a graphic,
switch to Interactive mode. In the SAR Polarimetry
Target Analysis window, click the Preferences menu,
point to Plot Options, and then click Interactive.

PCI Geomatics Page 119

Module 4: Using the SAR Polarimetry Target Analysis tool Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Defined and selected targets
• Produced numerical and graphical output for the selected targets
• Analyzed the numerical and graphical output

Page 120 PCI Geomatics

Module 5: Classifying SAR data

About this module

Module 5 has three lessons:
• Lesson 5.1: Classification of polarimetric data, possible workflows
• Lesson 5.2: Unsupervised Wishart classifications
• Lesson 5.3: Supervised Wishart classification (with targets defined in
SPTA)

Lesson 5.1: Classification of polarimetric data,

possible workflows
Applying polarimetric decompositions to fully polarimetric data (Module 4) is not an
end unto itself, but usually part of a workflow (an application). An obvious use of
polarimetric decomposition parameters is for land-cover and land-use classification.
In this regard, Geomatica provides several supervised and unsupervised
classification algorithms designed primarily for optical imagery. While some
polarimetric parameters can be used directly in the classification algorithms (for
example entropy, eigenvalues), others may require further processing (scaling,
thresholding) before being suitable for use.
Another option is to use a parameter or a set of parameters to create bitmaps to
restrain (or exclude) a classification to specific areas using different parameters.
Polarimetric decomposition tends to focus more on scattering-type characterization
and identification while the information in the backscattered power is sometimes
neglected or not used fully. With a fully polarimetric image (s4c), it is also possible
to generate many polarimetric discriminators based on an extensive analysis of the
scattered power (using the PSPOLSYN, PSPOLSYNC, and PSPOLSYNR algorithms).
This topic is covered in Appendix B.
All generated channels and bitmaps can be combined in a single PCIDSK file and
then classified with a conventional classification algorithm (MINDIS, MCL,
ISOCLUS).

PCI Geomatics Page 121

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

Figure 46. High-level workflow for classification of polarimetric data using

supervised or unsupervised classifications
The Geomatica radar suite includes algorithms that are designed specifically for
classifying polarimetric SAR data sets. The PSSWIS and PSUWIS (supervised and
unsupervised Wishart classification) algorithms use class distances customized for
polarimetric data. The training classes are determined from an existing polarimetric
decomposition result or based on polygons and bitmaps you provided representing
different land-use or land-cover classes. This topic will be covered in Lesson 5.2
and Lesson 5.3.

Figure 47. High-level workflow for classification of polarimetric data using

supervised or unsupervised Wishart classifications

Page 122 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

Lesson 5.2: Unsupervised Wishart classifications

In this lesson, you will:
• Run a Wishart unsupervised classification
• Analyze the results of the unsupervised Wishart classification

Unsupervised Wishart classification

The Wishart unsupervised classification is performed on polarimetric SAR (POLSAR)
data by running the PSUSWIS algorithm. The classification is similar to the method
used in the supervised maximum-likelihood classification for optical imagery,
except that the distance measured is customized for polarimetric data.
With the Wishart unsupervised classification, the training classes are determined
from an existing polarimetric decomposition: either a Freeman-Durden or Cloude-
Pottier decomposition. The output image is a single-channel image with pixel values
representing the class number, which can be between 1 and 9 for a Freeman-
Durden input, and between 1 and 16 for a Cloude-Pottier input. The use of iteration
refines the classification, but increases the processing time.
To complete this exercise you will use:
• Output created by the PSBOXCAR algorithm in Lesson 2.2
• Freeman-Durden decomposition output created in Lesson 3.2
• Cloude-Pottier image classification produced by the PSCLOPOT algorithm

To open the PSUSWIS algorithm

Performing an unsupervised Wishart classification using the

Freeman-Durden decomposition
Here, you will use the Wishart classification with the Freeman-Durden
decomposition.

To set up the PSUSWIS parameters

1. Under Input Ports, expand Input: Input polarimetric SAR image,
click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
2. Select the FQ29_20100507sig_PSBOXCAR_7.pix file, and then select
each raster in the file.

PCI Geomatics Page 123

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

3. Expand Input: Input Freeman-Durden or Cloude-Pottier classified

raster (Decomposition).
4. Click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
5. Select the FQ29_20100507sig_PSBOXCAR_7_PSFREDUR.pix file, and
then select each raster in the file.
6. Under Output Ports, expand Output: Output Wishart classified
raster, select Untitled.pix, and then click Browse.
The File Selector window appears.
7. In the File Name box, type
FQ29_20100507sig_PSBOXCAR_7_PSFREDUR_WIS.pix, and then
click Save.
The new path and file name appear in the PSUSWIS Module Control Panel
window.
8. Click the Input Params 1 tab.
9. In the Number of iterations list, click 10.
10. Click Run

Performing an unsupervised Wishart classification using the

Cloude-Pottier decomposition
Before running the Wishart classification using the Cloude-Pottier decomposition,
you must run the PSCLOPOT algorithm.
PSCLOPOT performs an unsupervised Cloude-Pottier classification of a fully
polarimetric SAR (POLSAR) data set. The classification is based on the partitioning
of the entropy, alpha angle, and anisotropy feature space using a series of
thresholds that can you can set. The output image has one channel, where the pixel
values indicate the class number assigned to each pixel.

To set up the PSCLOPOT parameters

1. Under Input Ports, expand Input: Input polarimetric SAR image.
2. Click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
3. Select the FQ29_20100507sig_PSBOXCAR_7.pix file, and then select
each raster in the file.
4. Under Output Ports, expand Output: Output Cloude-Pottier
classified raster, select Untitled.pix, and then click Browse.
The File Selector window appears.
5. In the File Name box, type
FQ29_20100507sig_PSBOXCAR_7_PSCLOPOT.pix, and then click
Save.
The new path and file name appear in the PSCLOPOT Module Control Panel
window.

Page 124 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

6. Click Run.

To set up the PSUSWIS parameters

1. Under Input Ports, expand Input: Input polarimetric SAR image.
2. Click Browse, and open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
3. Select the FQ29_20100507sig_PSBOXCAR_7.pix file, and then select
each raster in the file.
4. Expand Input: Input Freeman-Durden or Cloude-Pottier classified
raster.
5. Click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
6. Select the FQ29_20100507sig_PSBOXCAR_7_PSCLOPOT.pix file, and
then select each raster in the file.
7. Under Output Ports, expand Output: Output Wishart classified
raster, select Untitled.pix, and then click Browse.
The File Selector window appears.
8. In the File Name box, type
FQ29_20100507sig_PSBOXCAR_7_PSCLOPOT_WIS.pix, and then
click Save.
The new path and file name appear in the PSUSWIS Module Control Panel
window.
9. Click the Input Params 1 tab.
10. In the Number of iterations list, click 10.
11. Click Run.

PCI Geomatics Page 125

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

Comparing the unsupervised Wishart classifications

Exercise 1: Using the Landsat-5 image

(Flevoland_L5_20100906_p198_r23r24.pix) and the filtered FQ29 Flevoland image
(FQ29_20100507sig_PSBOXCAR_7.pix), evaluate and compare the Freeman-
Durden unsupervised classification with the Cloude-Pottier. What are the
similarities? What are the main differences? (Hint: Compare the urban areas.)

Figure 48. Unsupervised Wishart classifications using the Freeman-Durden

decomposition (7 x 7)

Figure 49. Unsupervised Wishart classifications using the Cloude-Pottier

decomposition (7 x 7)

Page 126 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

Lesson summary
In this lesson, you:
• Ran a Wishart unsupervised classification
• Analyzed the results of the unsupervised Wishart classification

PCI Geomatics Page 127

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

Lesson 5.3: Supervised Wishart classification (with

targets defined in SPTA)
SPTA provides several tools you can use to perform an in-depth analysis of the
scattering properties of point (Lesson 4.1) and distributed targets (Lesson 4.2),
based on regions you define. You can then interactively use SPTA and the Focus
SAR Radar Analysis toolbox to perform a complete polarimetric analysis.
In this lesson, you will use targets defined in SPTA to train the supervised Wishart
classification of the FQ29_20100507sig_PSBOXCAR_7.pix file in Focus.
In this lesson, you will:
• Set up preclassification tasks
• Import and edit training sites
• Run a Wishart supervised classification
• Analyze the results of the Wishart classification
• Compare the results of the supervised and unsupervised Wishart
classifications

Supervised Wishart classification

The Wishart supervised classification is performed on polarimetric SAR (POLSAR)
data sets using the PSSWIS algorithm. The classification is similar to the method
used in the supervised maximum-likelihood classification for optical imagery,
except that the distance measured is customized for polarimetric data. In the
Wishart supervised classification, you provide the training classes.
You specify a file containing the classification results in a single 8-bit channel. Its
pixel values specify the class number that is assigned to each pixel. Pixels that
cannot be classified (usually with zero values in all input channels) are represented
by a pixel value of zero. The classification file can hold up to 255 classes.

Note The PSSWIS algorithm was enhanced in Geomatica 2016

to support training sites in vector and bitmap format. In
previous versions of Geomatica, training sites were
supported only in bitmap format.

Preclassification tasks
The supervised Wishart classifier (the PSSWIS algorithm) requires as input a series
of training areas representing each class to classify. These training sites can be in
bitmap or vector format and created in Focus and then imported.

Importing the training sites from SPTA (vector format)

In Lesson 4.2, a series of targets (polygons) representing some land-use and land-
cover classes has been created already; these targets will be used as training sites.
You will find the targets in the folder:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\targets

Page 128 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

The first task is to put all the training sites in the same PCIDSK file. Each class will
be stored in a different segment. When the training sites are imported from
multiple sources, you must ensure that each shares the same projection as the
image to be classified. If bitmaps are used as training sites, they must have the
same number of lines and columns as the image to be classified.

Importing the training areas into one PCIDSK file

1. In Focus, click the Tools menu, and then click Data Merge.
2. Click Browse, select all the targets in
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
\targets, and then click Open.
3. Under Available Files/Layers, select all of the target files, and then click
Next.
4. Enter a name for the output file to which to write all of the training areas
after the merge.
For example, enter Flevoland_20100507_Training_Sites.pix.
Be sure to retain PCIDSK as the output format.
5. Under Geo-referencing Setup:
▪ Under Projection, click From File, and then select a target from the
list.
Notice that the projection information is updated.
▪ Under Extents, click From File, click Browse, and then select
FQ29_20100507sig_PSBOXCAR_7.pix.
This is image you generated in Lesson 2.2.
6. Under Resolution, retain the default values (make no changes).
Because the vector data is merged, the output resolution can be ignored.
7. Under Reprojection, you can also retain the default values.
8. Click Next.
9. Under Select Layer(s), confirm the file or files you selected, and then
click Finish.
The training sites are now merged into a single PCIDSK file.

Creating the training sites in Focus (vector format)

Alternatively, you can create the training sites in Focus.
1. In Focus, click the File menu, click Open, select the image you want to
classify, (FQ29_20100507sig_PSBOXCAR_7.pix), and then click Open.
2. Click the Maps tab, right-click New Area, and then click New Vector
Layer.
The New Vector window appears.
3. Under Layer Type, click Unstructured.
4. Under Georeferencing, click Use Layer Georeferencing.
Verify that the FQ29_20100507sig_PSBOXCAR_7.pix file is selected and
then to create the new vector layer, click OK.

PCI Geomatics Page 129

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

5. On the Focus toolbar, click the arrow beside New Shapes ( ), click
the type of shape you want (Polygon, Rectangle, Ellipse, or Trace),
and then draw a shape over an urban area on the new vector layer.

6. Save the vector you created as

Flevoland_20100507_Training_Sites.pix.
7. To add a second vector layer, click the Files tab, right-click
Flevoland_20100507_Training_Sites.pix, and then click New Vector
Layer.
8. Right-click the vector you created, and then click View.
9. Click the Maps tab.
10. Draw a new shape representing the desired land-use or land-cover class
using any of the shape commands you used in step 5.
Tip: Be sure to select the new vector layer before editing.
11. After editing, right-click the bitmap layer, and then click Save.
12. Repeat steps 7 to 11 for each new class.
13. Click the Files tab, and then select the
Flevoland_20100507_Training_Sites.pix file.
14. Expand Vectors, double-click a vector layer, and then rename it to better
describe its content.

Note All shapes created on the same layer must belong to the
same class.

Performing a supervised Wishart classification

You will now perform a supervised Wishart classification using the PSSWIS
algorithm.

To open the PSSWIS algorithm

Page 130 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

4. Double-click PSSWIS.
The PSSWIS Module Control Panel window appears.

To set up the PSSWIS parameters

1. Under Input Ports, expand Input: Input polarimetric SAR image.
2. Select each raster in the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Expand InputBitmaps: Input sample data.
4. Select each bitmap in the Flevoland_20100507_Training_Sites.pix file.
5. Under Output Ports, expand Output: Output Wishart classified
raster, select Untitled.pix, and then click Browse.
The File Selector window appears.
6. In the File Name box, type
FQ29_20100507sig_PSBOXCAR_7_PSSWIS.pix, and then click Save.
The new path and file name appear in the PSWIS Module Control Panel
window.
7. Click Run.

Interpreting the PSSWIS classification results

After running the PSSWIS algorithm, you can analyze the results.

To interpret the PSSWIS classification results

1. In Focus, open the FQ29_20100507sig_PSBOXCAR_7_PSEABA.pix file
created in Lesson 3.3.
If you do not have the open already, on the File menu, click Open, and
then open the file from its location.
2. To facilitate the interpretation of the supervised Wishart classification, you
can edit the legend names and colors, as shown in Figure 50.

PCI Geomatics Page 131

Module 5: Classifying SAR data Processing SAR data with Geomatica Banff

Exercise 2: What are the main differences between these two classifications? You
can use an RGB composite made of the HH, HV, and VV channels of the
FQ29_20100507sig_PSBOXCAR_7.pix file to help in your interpretation.

Unsupervised Wishart Supervised Wishart HH-HV-VV

(PSSWIS)

Unsupervised Wishart legend Supervised Wishart legend

Figure 50. Classifications comparison and legends

Page 132 PCI Geomatics

Processing SAR data with Geomatica Banff Module 5: Classifying SAR data

Lesson summary
In this lesson, you:
• Set up preclassification tasks
• Ran a Wishart supervised classification
• Analyzed the results of the Wishart classification
• Visually compared the results of the Wishart supervised and unsupervised
classifications

References
Ulaby, F.T., Held, D., Dobson, M.C., McDonald, K.C., Senior, T.B.A (1987). Relating
Polarization Phase Difference of SAR Signals to Scene Properties. IEEE Transactions
on Geoscience and Remote Sensing. Vol. GE-25, no.1, p.83-92.
Lee, Jong-Sen., Pottier, Eric (2009). Polarimetric Radar Imagining: from basics to
applications. CRC Press, Taylor & Francis Group, Boca Raton, Florida, USA. 398
pages.
Lo, C.P. (1998). Applications of Imaging Radar to Land Use and Land Cover
Mapping. Published in: Manual of remote sensing: principles and applications of
imaging radar. R.A. Ryerson (ed). John Wiley & Sons, Inc., New York. 896 pages.
van Zyl, J.J., Ulaby, F.T (1990). Chapter 2: Scattering Matrix Representation for
simple targets. Publié dans: Radar Polarimetry for Geoscience Applications. F.T.
Ulaby. C. Elachy, editors. Artech House, Norwood, MA, USA. 388 pages.

PCI Geomatics Page 133

Module 6: Change detection using
polarimetric data

About this module

Module 6 has five lessons:
• Lesson 6.1: Examples of possible workflows for change detection
• Lesson 6.2: Performing an intensity-change detection
• Lesson 6.3: Wishart change detection
• Lesson 6.4: Performing a phase-change detection
• Lesson 6.5: Incoherent change-detection workflow

Lesson 6.1: Examples of possible workflows for

change detection
In this lesson, you will:
• Learn about the various options for change detection in Geomatica
• Examine examples of possible workflows for coherent and incoherent
change detection

Types of changes and detection techniques

Detecting and monitoring changes between two images, or from a time series of
images, is an important use of SAR data. There are many types of changes
possible, such as:
1. Permanent (noncyclic) changes:
▪ Short-term or abrupt changes: disaster monitoring (flood, fire),
detecting movement of ships and other vessels, vehicles, and so forth
▪ Medium-term, long-term, or gradual changes: mostly land-use, land-
cover changes like deforestation and reforestation, urban sprawl,
glacier recession, wetland and forest monitoring, and so forth
2. Seasonal (cyclic) changes, crop growth and rotation, leaf cycle in forests

PCI Geomatics Page 135

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

The success of applying change detection depends on the adequacy between the
nature of change you want to detect and the sensor spatial, spectral, radiometric,
and temporal resolution. The techniques you select will have a significant bearing
on the results you obtain.
There are many techniques for detecting change with remote-sensing data. At one
end of the spectrum there is hard-change detection based on comparison of land-
use or land-cover classification and the resulting changes in discrete (qualitative)
categories.
At the other end, it is possible to directly compare two images using a band or ratio
technique or a principal-component analysis, for example. These techniques reveal
the magnitude of change, but provide little information on the nature of the
changes.
Between each end of the spectrum, a multitude of hybrid techniques exists with
varying levels of sophistication10.
The Geomatica radar suite has three coherent change-detection algorithms that can
extract the magnitude of change between two images (Table 10).

Table 10. Change-detection algorithm included in the Geomatica radar suite

Name Input type Description

CCDINTEN ▪ Detected or CCDINTEN measures the change in total
complex radar backscattering between the test
▪ Filtered or non and reference SAR images. If the image
is SLC, the intensity is determined and
▪ Single, dual or written to the output file. You can use
fully polarimetric integer or real-valued, speckle-filtered
(symmetrized or data input to use for testing or as
non) reference. You can apply several speckle
filters to your files before running
CCDINTEN.
CCDWISH ▪ Complex CCDWISH uses Wishart statistics to test
▪ SLC (non filtered) the equality of two complex covariance
matrices that are assumed to follow a
▪ Fully polarimetric Wishart distribution. When the
(symmetrized or backscatter coefficients are very close,
non) but either the phase or the correlation
coefficient between the copolarization has
changed, the Wishart test statistic is
much more sensitive to the differences
than test statistics based only on the
intensity. That is, CCDWISH can highlight
changes not detected by CCDINTEN.
The total number of channels to process
must be four or fewer.

10 See Jensen (2005) and Hussain et al. (2013) for a review.

Page 136 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Name Input type Description

CCDPHASE ▪ Complex CCDPHASE measures change based on
▪ Nonfiltered the average coherence (over a specified
window size) between two coregistered
▪ Fully-polarimetric SLC SAR images. Coherence values will
(symmetrized or range between zero and one, and areas
non) of change will have coherence values
close to zero (coherent).
The total number of channels to process
must be four or fewer.

Each change-detection algorithm in the preceding table automatically extracts the

overlap area between each image pair and writes four channels to the output file:
• Intensity value (or sum of intensities) of the input test data
• Intensity value (or sum of intensities) of the reference data
• Change metric (absolute intensity ratio, Wishart metric, or the absolute
intensity ratio of the coherence)
• Change, as percentiles, ranked from zero (no change) to 100 percent
(maximum change)
Image quality in detecting change in SAR data depends on the comparability of the
source data. As much as is possible, source images should have the same
calibration applied, the same georeferencing, pixel size, viewing geometry, and
polarization. The quality of the coregistration of each image pair used as input is
also a significant factor.

Tip To check pixel alignment before performing a change

analysis, run AUTOSHIFT. In this lesson, the scenes are
well aligned, so running AUTOSHIFT is not necessary.

Examples of possible workflows

Typically, applying the same calibration type to all of the data (sigma-nought, beta-
nought, gamma-nought) is more important than the calibration type. You must also
verify the alignment of all of the data. If the data is not well aligned, you can use
several workflows, depending on the severity of the misalignment and the data
type.
If orbit segments are available, you can collect GCPs and orthorectify the data in
OrthoEngine. If an orthorectification is inappropriate or there is insufficient
information on the space-borne or airborne sensor, you can use a different
technique (polynomial, thin-plate spline) to geocode the data. If the misalignment
between an image pair is slight, you can coregister them by running the
AUTOSHIFT algorithm.
Any geocoding applied to complex data is retained after it is geocoded. By doing so,
the data remains compatible with the change-detection algorithms that require
complex data as input (CCDPHASE and CCDWISH).

PCI Geomatics Page 137

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

To detect change, you can run CCDINTEN, CCDPHASE, and CCDWISH individually
or in combination to produce images showing change based on metrics (Table 10).
Similar to the output from polarimetric decompositions (Module 3), the output from
the change-detection algorithms is typically not the end of the workflow, but rather
the beginning.
An option is to extract areas of change based on image thresholding to produce
binary maps that represent changed areas versus non. You can also convert binary
maps to vector layers and overlay or combine them with ancillary data to interpret
change.
Another option, if you are working with fully polarimetric data, is to apply one or
many polarimetric decompositions on the same data set. In this case, the change-
detection algorithms provide the magnitude of change (quantitative) while the
polarimetric decompositions help you to understand the nature of change
(qualitative) by identifying the scattering mechanisms. This strategy is
demonstrated in Lesson 6.4.
Finally, to evaluate the same area, use a similar size processing window for change
detection and polarimetric decompositions.

Figure 51. Typical workflow for detection of incoherent change. Dotted lines indicate
optional steps.
A similar workflow for change detection can be applied to coherent targets; that is,
hard targets that cover from one to a few tens of pixels that can be compared to
canonical targets (see Module 3, polarimetry fundamentals). A coregistration
between each SLC image pair is usually preferred over an orthorectification to
preserve the highest possible spatial resolution and to preserve the signal as close
as possible to its raw (calibrated) values. Because the characterization of coherent
targets relies essentially on the phase, use a smaller window size with algorithms
such as CCDPHASE and CCDWISH.
Distinguishing change that occurs over few contiguous pixels from the speckle can
be a challenge. This is common because (conventional) speckle filtering is not
usually applied to the data before the change-detection analysis or coherent-target
characterization.

Page 138 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Another challenge is to identify pixels that correspond to coherent targets according

to tests (assumptions) that range from a simple thresholding of numeric values—to
extract bright pixels—to more sophisticated tests that include an evaluation of the
target phase, a comparison of the target with its surrounding area (the "clutter", as
described in Lesson 4.2), or both.
The whitening filter, PSWHITE, or the symmetric scattering characterization
algorithm (PSSSCM) can find and characterize such targets. Coherent polarimetric
decomposition is another way to characterize coherent targets for those pixels
identified as such by PSWHITE or PSSSCM.
Finally, coherent-target change can be interpreted in the same way as incoherent
changes by combining the change-detection results with the target-characterization
workflow.

Figure 52. Possible workflow for coherent change-detection.

Dotted boxes and arrows indicate nonmandatory steps.
Today, many sensors, such as Radarsat-2 and TerraSAR-X, have a high-resolution
acquisition mode, so the concept of point target is less relevant. For example, large
ships can be characterized by tens of pixels, more for large buildings, while smaller
targets like cars and poles can encompass tens of pixels and can be treated as
incoherent targets.
On average, however, such targets are more coherent compared to extended
targets (hundreds or thousands pixels). A coherent analysis conducted at full
resolution might reveal a precise characterization of their properties and structure.

PCI Geomatics Page 139

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Learned about the options available for change detection in Geomatica
• Examined examples of possible workflows for detection of coherent and
incoherent change

Page 140 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Lesson 6.2: Performing an intensity-change detection

In this lesson, you will:
• Run an intensity-change detection using the CCDINTEN algorithm on the
HH, HV, and VV channels and the SPAN
• Analyze the results of intensity-change detection
To complete this lesson, you need:
• FQ29_20100507sig.pix, located in:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\
(This is the file you created in Lesson 1.4.)
• To ingest the Radarsat-2 scene in
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100531\, select the
sigmanought calibration, and then name the file
FQ29_20100531sig.pix, as described in Lesson 1.4.
CCDINTEN measures the change in total radar backscatter between the test and
reference SAR images by comparing the sum of the intensities of the input
channels. Given two registered detected or SLC SAR images, CCDINTEN determines
the overlap area, normalizes the intensity values based on the total span, and
calculates the change as a percentile.
If the image is SLC, the intensity is determined and written to the output file.
CCDINTEN is NoData-aware, meaning that pixels that contain a value defined as
NoData in the metadata of the input file are excluded from processing. The input
and reference layers must be correspondingly comparable; that is, the first layer
from each image must have the same polarizations.

To open the CCDINTEN algorithm

To set up the CCDINTEN algorithm

1. Under Input Ports, expand Input: Input complex or detected SAR
image, click Browse, and then open the FQ29_20100507sig.pix file.
2. Select the HH channel.
3. Under Input Ports, under Reference: Reference complex or detected
SAR image, click Browse, and then open FQ29_20100531sig.pix.
4. Select the HH channel.
5. Under Output Ports, click Browse.
6. In the File name box, enter
FQ29_20100507_0531sig_CCDINTEN_15x15_HH.pix as the name of
the output file.

PCI Geomatics Page 141

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

7. Click the Input Params 1 tab, and then in the Window size list, click
15.
8. Click Run.

Exercise 1: Repeat steps 1 to 7, by selecting the HV channel instead of HH.

Specify FQ29_20100507_0531sig_CCDINTEN_15x15_HV.pix as the name of
the output file.
Repeat steps 1 to 7, by selecting the VV channel instead of HH. Specify
FQ29_20100507_0531sig_CCDINTEN_15x15_VV.pix as the name of the
output file.
Repeat steps 1 to 7, by selecting all the raster channels instead of HH. Specify
FQ29_20100507_0531sig_CCDINTEN_15x15_span.pix as the name of the
output file.

Exercise 2: Interpret and compare the results of the CCDINTEN change detection
for HH, HV, VV, and the span.

HH/R: 0507 HH HH
/G:0531/B:0507 Change metric 15 x 15 Ranked change 15 x 15

Page 142 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

HV/R: 0507 HV HV
/G:0531/B:0507 Change metric 15 x 15 Ranked change 15 x 15

VV/R: 0507 VV VV
/G:0531/B:0507 Change metric 15 x 15 Ranked change 15 x 15

Span Span Span

/R: 0507 /G:0531 /B:0507 Change metric 15 x 15 Ranked change 15 x 15

Figure 53. CCDINTEN results using a processing window of 15 x 15

PCI Geomatics Page 143

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Ran intensity-change detection on SAR data
• Analyzed the results of the intensity-change detection

Page 144 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Lesson 6.3: Wishart change detection

In this lesson, you will:
• Run an intensity-change detection using the CCDWISH algorithm on the
SPAN
• Analyze the results of Wishart change detection
To complete this lesson, you will need:
• FQ29_20100507sig.pix, located in:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\
• FQ29_20100531sig.pix, located in:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100531\
CCDWISH measures change between two SLC SAR images, based on modified
Wishart statistics. The statistical probability of the area of interest being "the same"
is computed in the defined analysis window. Areas with a high probability are
considered unchanged, while areas with a low probability are considered changed;
that is, they have a low probability of being the same.
The Wishart-test statistic is much more sensitive to the differences than test
statistics based only on the intensity; that is, CCDWISH can indicate changes not
detected by CCDINTEN.

To open the CCDWISH algorithm

To set up the CCDWISH algorithm

1. Under Input Ports, under Input: Input complex SAR image, click
Browse, and then open FQ29_20100507sig.pix.
2. Select each raster in the FQ29_20100507sig.pix file.
3. Under Input Ports, under Reference: Reference complex SAR image,
click Browse, and then open FQ29_20100531sig.pix.
4. Select each raster in the file FQ29_20100531sig.pix.
5. Under Output ports, click Browse, and then in the File Name box enter
FQ29_20100507_0531sig_CCDWISH_15x15_span.pix as the name
of the output file.
6. Click the Input Params 1 tab, and then in the Window size list, click
15.
7. Click Run.

PCI Geomatics Page 145

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Exercise 4: Compare the change-detection results for the span obtained with
CCDINTEN and CCDWISH. Which type of land use and land cover shows the most
difference, and which one shows the least difference?

CCDWISH change metric CCDWISH ranked change R: CCDINTEN ranked

G: CCDWISH ranked
B: CCDINTEN ranked

Figure 54. CDWISH change-detection results for span (15 x 15 window)

Exercise 5: The differences in results between CCDINTEN and CCDWISH shown in

Figure 54 may be linked to a change that occurred in the HH-VV correlation or
phase difference instead of a change in intensity. To verify this, use the PSCC
algorithm to calculate the correlation coefficient between the HH and VV channels.

To open the PSCC algorithm

To set up the PSCC algorithm

1. Under Input Ports, expand Input: Input polarimetric SAR image,
click Browse, and then open _20100507sig_PSBOXCAR_7.pix.
2. Select each raster in the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Under Output Ports, expand Output: Output correlation coefficient
raster, click Browse, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\
folder, and then specify
FQ29_20100507sig_PSBOXCAR_7_PSCC_HH-VV.pix as the name of
the output file.

Page 146 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

4. Click the Input Params 1 tab.

5. In the First input polarization list, click HH in the Second input
polarization list, click VV, and then in the Angle measurement list,
click Degrees.
6. Click Run.
Repeat steps 1 to 5 selecting FQ29_20100531sig_PSBOXCAR_7.pix in Input
Ports and specifying FQ29_20100531sig_PSBOXCAR_7_PSCC_HH-VV.pix as
the name of the output file.

Correlation coefficient Correlation coefficient R: PSCC 05-31

magnitude magnitude G: PSCC 05-07
(2010-05-07) (2010-05-31) B: PSCC 05-31

Figure 55. PSCC correlation coefficient ([0-1]) between channels HH and VV

PCI Geomatics Page 147

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Ran a Wishart change detection on SAR data
• Analyzed the results of a Wishart change detection
• Compared the results of a Wishart change detection and an intensity-
change detection
• Compared the results of a Wishart change detection with a change
observed in the HH-VV correlation coefficient

Page 148 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Lesson 6.4: Performing a phase-change detection

In this lesson, you will:
• Run a phase-change detection using the CCDPHASE algorithm
• Analyze the results of the phase-change detection
For this lesson, you will need:
• FQ29_20100507sig.pix, located in:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507\
• FQ29_20100531sig.pix, located in:
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100531\
CCDPHASE measures change based on the average coherence (over a specified
window size) between two coregistered SLC SAR images. Coherence values will
range between zero and one, where areas of change will have coherence values
close to zero (incoherent) while unchanged areas will have higher values
(coherent).

To open the CCDPHASE algorithm

To set up the CCDPHASE algorithm

1. Under Input Ports, expand Input: Input complex SAR image, click
Browse, and then open FQ29_20100507sig.pix.
2. Select each raster in the FQ29_20100507sig.pix file.
3. Under Input Ports, under Reference: Reference complex or detected
SAR image, click Browse, and then open FQ29_20100531sig.pix.
4. Select each raster in the FQ29_20100531sig.pix file.
5. Under Output ports, click Browse, and then enter
FQ29_20100507_0531sig_CCDPHASE_span.pix as the name of the
output file.
6. On the Input Params 1 tab, in the Window size list, click 15.
7. Click Run.

PCI Geomatics Page 149

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Span Span
Change metric Ranked change

Figure 56. CCDPHASE change-detection results (15 x 15 window)

Exercise 5: Run the CCDPHASE detection algorithm for HH, HV, and VV channels.

HH HH
Change metric 15 x 15 Ranked change 15 x 15

Page 150 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

HV HV
Change metric 15 x 15 Ranked change 15 x 15

VV VV
Change metric 15 x 15 Ranked change 15 x 15

Figure 57. CCDPHASE results (15 x 15 window)

Exercise 6: Compare the CCDPHASE change-detection results to the HH-VV phase

difference.
PSPHDIFF determines the phase-angle difference between two polarizations of a
polarimetric SAR (POLSAR) data set. The dual-polarization and quad-polarization
polarimetric data is accepted for processing.

To open the PSPHDIFF algorithm

PCI Geomatics Page 151

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

To set up the PSPHDIFF algorithm

1. Under Input Ports, expand Input: Input polarimetric SAR image,
click Browse, and then open FQ29_20100507sig_PSBOXCAR_7.pix.
2. Select each raster in the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Under Output ports, click Browse, and then enter
FQ29_20100507sig_PSBOXCAR_7_PSPHDIFF_HHVV.pix as the
name of the output file.
4. Click the Input Params 1 tab.
5. In the First input polarization list, click HH, in the Second input
polarization list, click VV, and then in the Angle units list, use the
default setting Degrees.
6. Click Run.
Repeat steps 1 to 5 using FQ29_20100531sig_PSBOXCAR_7.pix in step 1 and
specifying FQ29_20100531sig_PSBOXCAR_7_PSPHDIFF_HHVV.pix in step 2.

2010-05-07, 2010-05-31, R:05-07 / G:05-31

HH-VV phase HH-VV phase B:05-07

Figure 58. PSPHDIFF phase difference between HH and VV channels

Page 152 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Lesson 6.5: Incoherent change-detection workflow

In this lesson you will:
• Use CCDWISH to produce a map of changes
• Extract areas of change using channel thresholding
• Export the areas of change to a vector file using BIT2POLY
• Interpret the nature of change using a polarimetric decomposition
(PSTOUZIDEC)

Data preparation
To complete this lesson you need:
• The FQ29_20100507_0531sig_CCDWISH_15x15_span.pix file produced in
Lesson 6.3. If this file has not been produced, it can be generated with
CCDWISH (15 x 15 processing window) using as input the two Radarsat-2
Flevoland data sets ingested with the sigma nought calibration
(FQ29_20100507sig.pix and FQ29_20100531sig.pix).
• The results from the Touzi decomposition (PSTOUZIDEC) applied on
FQ29_20100507sig.pix and FQ29_20100531sig.pix. Remember to first
filter the files using PSBOXCAR with a 7 x 7 processing window. The
outputs can be named:
FQ29_20100507sig_PSBOXCAR_7_PSTOUZIDEC.pix and
FQ29_20100531sig_PSBOXCAR_7_PSTOUZIDEC.pix.

Area-of-change extraction using band thresholding

After producing the change-metric file, extract the areas of change. Here, there are
several possibilities. Image thresholding is a simple technique in which numeric
values above a certain value (threshold) are considered as change.
The question is how to determine the threshold value? While there is no definitive
answer—the threshold value depends on the situation, the application objective,
and the size of the processing window—a good starting point is to look at the
histogram of the change metric.

To display the histogram of a channel

1. In Focus, open the file in question.
2. Click the Files tab, and then expand the Rasters list of
FQ29_20100507_0531sig_CCDWISH_15x15_span.pix.
3. Right-click the third channel (CCDWISH: Change metric), and then click
Histogram with statistics.
The Histogram with Statistics window appears.
The histogram for the CCDWISH change metric is shown in Figure 59. The
histogram of the same metric produced with a 7 x 7 processing window is also
shown for comparison.

PCI Geomatics Page 153

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

As a reminder, the change metric from CCDWISH (and CCDPHASE and CCDINTEN)
does not consider the direction of change; therefore, the higher the number, the
stronger the change is according to the metric. Based on a visual inspection of the
histogram, the majority of pixels are distributed between 24.1 and 31 with a mean
of 25.52. The shape of the histogram suggests that a value of approximately 26
may be a good starting point for setting the threshold value.

A) CCDWISH: Change metric, 15 x 15 B) CCDWISH: Change metric, 7 x 7

Figure 59. CCDWISH change-metric histograms

Using Focus EASI Modeling to set threshold value and extract areas of
change
The change metric will be thresholded and the result stored in a bitmap layer.
1. If not open already, in Focus, open
FQ29_20100507_0531sig_CCDWISH_15x15_span.pix.
2. Click the Files tab, and then select the
FQ29_20100507_0531sig_CCDWISH_15x15_span.pix file.
3. Right-click, point to New, and then click Bitmap Layer.
4. Load the (empty) bitmap.
5. Right-click the bitmap, and then click View.
The bitmap is displayed automatically at the top on the Map tab.
6. On the Tools menu, click EASI Modeling.
You are now ready to extract the areas of change.
7. Using the following script, extract all pixels that are above 26 and output
the results in the new bitmap (%%2).

Page 154 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

A visual inspection indicates that a value of 26 may to be too permissive, especially

over the water areas (Figure 60, A and C). Inland, most of the detected changes
are associated with agricultural land and wetlands. None of the urban areas are
showing change (except for some localized targets), which is expected because of
the short lapse in time between the two acquisitions. This also indicates that the
coregistration between the two images is good.
In comparison, a threshold value of 27.5 produced better results (Figure 60, B and
D) by restricting the change to the most important ones while reducing the amount
of change over the water area.

A) CCDWISH >26 B) CCDWISH >27.5

PCI Geomatics Page 155

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

C) CCDWISH >26 (detail) D) CCDWISH >27.5 (detail)

Figure 60. Thresholding of the CCDWISH change metric

Converting a bitmap layer to a vector layer

The bitmap representing the areas of change can now be exported to a vector-
polygon layer. The benefits of this are:
• A bitmap layer is characterized by a pixel-and-line-coordinate system tied
to the raster file from which it is created. A vector layer is referenced
according to a geographical projection and not attached to a specific layer.
• It is easier to overlay vectors on relevant-image (data) layers to interpret
change
• It is possible to filter the polygons of change based on their areas or
shape, if necessary
• It is possible to automatically collect statistics for each polygon in a vector
layer

To convert a bitmap layer to a polygon layer

1. In Focus, click the Tools menu, and then click Algorithm Librarian.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find what box, type BIT2POLY, and then click Find Next.
4. Double-click BIT2POLY.
5. If not already open, under Input Ports, click Browse, and then select
the FQ29_20100507_0531sig_CCDWISH_15x15_span.pix file.
6. Under Input: Bitmap Layer, select the bitmap you want to convert to a
polygon layer.
For this exercise, select the bitmap that corresponds to CCDWISH>27.5.
7. Under Output Ports, click Output: Polygon Layer, click Untitled.pix,
and then click Browse.

Page 156 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

8. In the File Selector window, enter a file name and a location.

For this exercise, name the file
polygons_CCDWISH_15_Change_sup27.5.pix
9. Click the Input Params 1 tab.
Optionally, you can select the Smooth Vector option.
10. Click Run.
Each contiguous region of on pixels; that is, values of 1, will be grouped
together into a single polygon. Only pixels with a value of 1 in the input
bitmap result in polygons in the output.
Create a copy of polygons_CCDWISH_15_Change_sup27.5.pix.

Filtering and extracting statistics from polygons

For each polygon, BIT2POLY automatically creates four fields:
• Shape ID (unique identifier)
• Area (sq m)
• Perimeter (m)
• PixelValue (inherited from the bitmap layer)
View the attributes of the vector file.

To view a table of vector-file attributes

1. In Focus, click the Maps tab.
2. Right-click the vector file for which you want to view the attributes, and
then click Attribute Manager.
The Attribute Manager window appears.

In the following procedure, you will use a filter to remove all polygons less than
3,000 meters squared in area, which corresponds approximately to 110 pixels.

PCI Geomatics Page 157

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

To filter the polygons based on area

1. In the Attribute Manager window, click the Record menu, point to Query
By, and then click Example.
2. In the Attributes box, select Area.
3. Click the menu, and then select the Less than operator (<).
4. In the New statement box, enter 3000, and then click Add.
5. Click Apply.
All polygons less than 3,000 meters squared in area (4458 of 6116
polygons) are selected (Figure 61).
6. In the Attribute Manager window, click the Record menu, and then to
remove all the selected polygons, click Delete.
7. On the Layer menu, click Save.
8. Rename the vector file
polygons_CCDWISH_15_Change_sup27.5_sup3000sqm.pix.

Figure 61. Polygons greater than 3,000 meters squared (white) superposed on the
original bitmap layer in red (CCDWISH>27.5)

Page 158 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

Extracting statistics to the polygon attributes table

For each polygon, you will now extract statistics to help you understand the nature
of change. In this example, you will use the polarimetric parameters of the Touzi
decomposition (see Lesson 3.4). For each polygon, the mean of the following
parameters will be calculated:

•  s1 , dominant symmetric-scattering type

• Span, which corresponds to the sum of the eigenvalues ( 1 + 2 + 3 )

• 1 N , normalized first eigenvalue, which corresponds to 1 /span

Before proceeding, you must apply the Touzi decomposition (PSTOUZIDEC) to the
Radarsat-2 Flevoland data set. Remember to add extra 32R channels to each file
and compute the normalized version of the first eigenvalue (see Lesson 3.4).
Next, use the Focus Overlay Wizard to compute the mean of each parameter of
each polygon in polygons_CCDWISH_15_Change_sup27.5_sup3000sqm.pix. This
example calculates the mean of the dominant scattering type (  s1 ) of the May 7
image.

To use the Overlay Wizard

1. In Focus, click the Analysis menu, and then click Overlay.
The Overlay Wizard window appears.
2. Under Select type of overlay, click Statistical - Computes statistics
by virtually overlaying them, and then click Next.

PCI Geomatics Page 159

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

3. Under Primary Input, under Layer, click the File list, and then select
the vector file to add the new attributes (statistics); that is, select
polygons_CCDWISH_15_Change_sup27.5_sup3000sqm.pix.
4. Under Secondary Input, under Layer, click the File list, and then select
the file containing the layer of interest; that is, select
FQ29_20100507sig_PSBOXCAR_7_PSTOUZIDEC.pix.
If the file is already open in Focus it will be available in the list; otherwise,
to open the file, click Browse.
5. Click the Layer list, select (3 [32R] Dominant Touzi Alpha_S
Parameter, and then click Next.
6. In the table, select one or more attributes to compute (you may need to
click Advanced to see all of them), and then click Finish.
The selected attributes are calculated.
7. Repeat steps 1 to 7 for each layer of interest.
After collecting all of the statistics, the vector file should contain
approximately 1660 records(polygons) and 10 fields. This file can be
found in ~\SAR_Training\Radar\Vectors\
(polygons_CCDWISH_15_Change_sup27.5_sup3000sqm.pix).

Tip You can export the attributes to other formats, such as a

Microsoft Excel spreadsheet or an Oracle configuration
file. To do so, on the Layer menu, click Export
Attributes.

Interpreting change
The polygons representing the areas of change, the statistics collected for each
polygon, the original parameters from the Touzi decomposition, and the Landsat
images can all be used to interpret and understand change.

Page 160 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data

In Figure 60, you can see that most change is associated with surface scattering in
agricultural fields. This is confirmed in Figure 62, where  s1 shows values less than
40 degrees for the two dates for most of the polygons. In the same figure, (  s1 ),
most of the polygons are clustered along the 1:1 line and do not show a big change
in the dominant scattering mechanisms, which, at first, might look counterintuitive.
In comparison, as expected, most polygons show a strong difference in the span
scatter plot because CCDWISH uses the trace of the coherency matrix (Figure 62,
12-D). A possible explanation is that soil moisture increases the backscatter power,
but with little effect on the phase relationship between the two orthogonal
polarizations used for signal Tx/Rx (usually HH-VV). The backscattering mechanism
remains the same (wet soil versus dry), especially because the backscattered
power is normalized in the calculation of  s1 .

A)  s1 , dominant symmetric scattering B) 1 N , normalized first eigenvalue

type

PCI Geomatics Page 161

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

C) Span (intensity) D) Span (Intensity, axis cut at 0.5)

Figure 62. Change-detection results, agricultural area

Some polygons are characterized by a change in both  s1 and the span. A typical
situation is fields occupied by row crops in which the combination of the plant
growth and rows increase  s1 (from surface toward dipole or a week double-
bounce scattering) while the span has decreased due to soil drying between May 7
and May 31 (Figure 62, A and B; Figure 63, A and B versus C).
Figure 64 and Figure 65 show change associated with moving targets over land or
water. Typically, these targets produce the most noticeable changes in the
scattering type, the span and the purity of the backscattered power represented
here by 1 N .
You can see in Figure 64, E) that the workflows discussed in this section worked
well in urban areas to filter out the few changes that were associated with small
differences in the viewing geometry between the two images. In the same figure,
you can also see the effect of using a large window size (15 x 15) in CCDWISH:
change associated with small targets has expanded to the areas that surround
them.

Page 162 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data


A) May 7, 2010, s1 ,

B) May 31, 2010, s1 ,
C) Span
(R: May 7, G: May 31, B:
dominant symmetric dominant symmetric May 7)
scattering type scattering type

The polygons correspond to

CCDWISH greater than 27.5
and greater than 3,000
meters squared.

D) May 7, 2010, 1 N , first E) May 31, 2010, 1 N , first

eigenvalue eigenvalue

Figure 63. Change-detection results, agricultural area

PCI Geomatics Page 163

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff


A) May 7, 2010, s1 ,

B) May 31, 2010, s1 ,
C) Span
(R: May 7, G: May 31, B:
dominant symmetric dominant symmetric May 7)
scattering type scattering type

The polygons correspond to

CCDWISH greater than 27.5
and greater than 3,000
meters squared.

D) High-resolution imagery E) CCDWISH, original

(from Google Earth) change-detection bitmap in
red

Figure 64. Change-detection results, urban environment.

Page 164 PCI Geomatics

Processing SAR data with Geomatica Banff Module 6: Change detection using polarimetric data


A) May 7, 2010, s1 ,

B) May 31, 2010, s1 ,
C) Span
(R: May 7, G: May 31, B:
dominant symmetric dominant symmetric May 7)
scattering type scattering type

The polygons correspond to

CCDWISH greater than 27.5
and greater than 3,000
meters squared.


D) May 7, 2010, 1N , first

E) May 31, 1N , first
eigenvalue eigenvalue

Figure 65. Change-detection results, ship detection

PCI Geomatics Page 165

Module 6: Change detection using polarimetric data Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Used CCDWISH to produce a map of change
• Extracted areas of change using channel thresholding
• Exported the areas of change to a vector file using BIT2POLY
• Interpreted the nature of change using a polarimetric decomposition
(PSTOUZIDEC)

References
Masroor Hussain, Dongmei Chen, Angela Cheng, Hui Wei, David Stanley (2013).
Change detection from remotely sensed images: From pixel-based to object-based
approaches. ISPRS Journal of Photogrammetry and Remote Sensing, Vol.80, p.91-
106.
Jensen, J. R (2005). Introductory Digital Image Processing: A Remote Sensing
Perspective. 3rd ed. Prentice-Hall Series in Geographic Information Science.
Pearson/Prentice Hall, Upper Saddle River, N.J., USA, 526 pages.

Page 166 PCI Geomatics

Appendix A

Processing with single-pol or dual-pol detected data

Appendix A has three lessons:
• Applying a band ratio or a band difference to a pair of detected images
using the EASI Modeling feature in Geomatica Focus
• Performing intensity-change detection
• Computing SAR texture measures
This appendix shows how Focus can increase the information content of a detected
single-polarization or dual-polarization image. A detected, dual-polarization image
(simulated) acquired over the Flevoland region of the Netherlands is used.

Applying ratio or difference to detected channels

using EASI Modeling
In this lesson you will:
• Apply a band ratio and a band difference between two polarimetric
channels using the EASI Modeling feature
• Compare the band-ratio and band-difference results

Band ratio and band difference

The band-ratio and band-difference techniques can be used to extract:
• Change between two images acquired on different dates
• Difference between two polarization channels of the same image
You can use various strategies to extract change or difference between two images
or channels. First, you can use a pixel-by-pixel ratio or difference, and then filter
the result using an average (or median) filter. You can also compute the mean (or
other statistics) over a window, return the result to the central pixel, and then
proceed to the ratio or difference. Another strategy is to first filter the data, and
then proceed to a pixel-by-pixel comparison between the two channels or images.

PCI Geomatics Page 167

Appendix A Processing SAR data with Geomatica Banff

Data preprocessing
Using the concepts learned in Lesson 1.4, Lesson 1.5, and the Flevoland scene
located in ~\SAR_Training\ Radar\Flevoland_RS2_SLC\FQ29_20100507:
1. Ingest the product.xml file into a PCIDSK file using sigma as the
calibration type.
Name the file FLE_FQ29_20100507sig.
2. Use the PSIQINTERP algorithm to convert complex data to detected data.
Convert the HH and HV channels to intensity.
Name the file FLE_FQ29_20100507sig_HH-HV.

To preprocess an image when applying a speckle filter

1. In Focus, click the Tools menu, and then click Algorithm Librarian.
2. In the Algorithm Librarian window, click Find.
The Find Algorithm window appears.
3. In the Find what box, type FSPEC, and then click Find Next.
4. Double-click FSPEC.
The FSPEC Module Control Panel window appears.
5. Under Input Ports, open the ~\SAR_Training\Radar\Flevoland_RS2
folder, and then select the FLE_FQ29_20100507sig_HH-HV file.
6. Under Output Ports, select Untitled.pix, and then click Browse.
The File Selector window appears.
7. In the File name box, type FLE_FQ29_20100507sig_HH-
HV_FEfrost_7.pix.
8. Click the Input Params 1 tab.
9. In the Image Units list, click Power, in the Filter Type list, click
Enhanced Frost, in the Filter X Size list and Filter Y Size list,
respectively, click 7.
Leave the other options set to the default values.
10. Click Run.

To perform a channel ratio and a channel difference using EASI Modeling

1. Add two new raster layers: in Focus, click the Files tab, right-click the
FLE_FQ29_20100507sig_HH-HV_FEfrost_7.pix. file, point to New,
and then click Raster Layer.
The Add Image Channels window appears.
2. Add two (2) 32 bit real channels, and then click Add.
3. On the Tools menu, click EASI Modeling.
The EASI Modeling window appears.
4. In the Input File list, click the FLE_FQ29_20100507sig_HH-
HV_FEfrost_7.pix file.
5. Enter the code to perform the ratio and difference (the comments can be
omitted), as shown in the following figure.

Note The modeling results are saved automatically.

Page 168 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Compare the ratio and difference results to the original HH and HV channels.

PCI Geomatics Page 169

Appendix A Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Applied a band ratio and a band difference between two polarimetric
channels using EASI Modeling
• Compared the band-ratio and band-difference result

Page 170 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Applying an intensity-change detection on detected

data
In this lesson you will:
• Apply intensity-change detection between two polarimetric channels
• Analyze the results
The CHDET algorithm detects change in SAR imagery. Given two registered
amplitude or intensity SAR images, CHDET calculates a difference image, removes
the speckle noise, and generates bitmap overlays, which display changes greater
than a specified threshold and the number of pixels.
CHDET is intended to detect change between two single-channel images acquired
on different dates. However, you can use this program to detect change between
two channels of a dual-pol image.

Applying an intensity-change detection

You can apply an intensity-change detection with the CHDET algorithm.

To open the CHDET algorithm

To set up the parameters for CHDET

1. Open the nonfiltered Fle_FQ29_20100507sig_HH-HV.pix file.
2. Under Input Ports, expand InputA: Layer A, and then select the HH
channel.
3. Expand InputB: Layer B, and then select the HV channel.
4. Under Output Ports, expand Output: A-B Difference Layer, and then
select the Untitled.pix check box.
5. Click Browse, open the folder in which you want to save the file, and then
in the File name box, type Fle_FQ29_20100507sig_CHDET-avg7.pix.
6. Click the Input Params 1 tab.
7. Do the following:
▪ In the Filter Type list, click AVG.
▪ In the Positive Threshold box and the Negative threshold box,
type 5 and -30 (decibels), respectively.
▪ In the Minimum Number of Neighbours box, click 0
▪ In the Image Format list, click Power.
▪ In the Filter Size list, click 7.

PCI Geomatics Page 171

Appendix A Processing SAR data with Geomatica Banff

▪ In the Number of Looks box, type 1

▪ In the Data Type list, click 32R.
8. Click Run.

Exercise 2: For which kind of land use and land cover is HH higher than HV? For
which kind of land use and land cover is HH similar to HV?
Compare the results from the previous lesson (ratio). What are the main
differences you can observe between these two techniques? Are some differences
related to type of filter used?
To facilitate the interpretation of the results, a Landsat-5 image mosaic of the
region is provided. Select the Flevoland_L5_20100906_p198_r23r24.pix file located
in the ~\SAR_Training\Landsat folder.

Page 172 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

HH (nonfiltered) HV (nonfiltered)

HH/HV ratio, Enhanced Frost filter 7 x 7 CHDET HH/HV ratio, Average 7 x 7

(Lesson 3.1) (Lesson 3.2)

Figure 66. Comparison of band-ratio results

PCI Geomatics Page 173

Appendix A Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Applied and intensity-change detection between two polarimetric channels
• Analyzed the results

Page 174 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Computing SAR texture measures

In this lesson, you will:
• Calculate SAR-specific texture measures by using the SARTEX algorithm
• Calculate texture measures based on co-occurrence matrices by using the
TEX algorithm
SARTEX creates a set of texture images from a single channel in an input image.
The radar-specific measures are based on pixel-value statistics in a window
surrounding each pixel. With SAR images, a single pixel conveys very little
information about the underlying ground cover. Only by looking at the values of
many pixels can some interpretations be made. The statistics that apply best to a
given area depend on the formation process of the actual image.
The extracted texture measures can be used as input for classification algorithms.
The SAR-specific texture measures in this module account for radar-image
formation and statistical properties of radar speckle. Therefore, it is essential that
no prior filtering has been performed on the input channel.
Texture measures extracted by SARTEX are based on SAR-specific statistics of pixel
values in the rectangular window using dimensions you specify. They complement
measures extracted by using the TEX algorithm, based on a gray-level co-
occurrence matrix (GLCM), and with the HISTEX algorithm, based on the histogram
of pixel values in a window. The three types of measures can be combined for
further analysis, such as segmentation and classification of the image.
The optimal window size depends on image characteristics (terrain roughness,
land-cover type, imaging parameters). With large windows, some blocky artifacts
can appear in texture images, if a small, bright target is enclosed fully by the
window. Therefore, a small-to-moderate window size is recommended.

Creating SAR-specific texture images

To create SAR-specific texture images, you run the SARTEX algorithm.

To open the SARTEX algorithm

To set up the parameters for SARTEX

1. Under Input Ports, expand Input: Input detected SAR band.
2. From the Fle_FQ29_20100507sig_HH-HV.pix file, select the HH
channel.
3. Under Output Ports, expand Output: Output texture measure
raster, select Untitled.pix and then click Browse.
The File Selector window appears.

PCI Geomatics Page 175

Appendix A Processing SAR data with Geomatica Banff

4. Open the Radar Data folder.

5. In the File name box, type Fle_FQ29_20100507sig_HH-
HV_SARTEX_7.pix, and then click Save.
6. Click the Input Params 1 tab.

7. Under Basis of SAR Texture Measure, select all four of the check boxes.
8. In the Horizontal Window Size list, click 7.
9. In the Vertical Window Size list, click 7.
10. In the Image Units list, click Power.
Note: It is important to select the correct image format for the input
layer. Any required conversions are performed internally to use the
correct values for each computed texture measure.
11. Click Run.

Page 176 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Note To solve a specific problem, such as distinguishing

between several land-cover types, for example, a typical
approach is to compute all of the texture measures, and
then use the measure that provides the best result.

Table 11. SAR-specific texture measures

Texture measure Explanation

Normally distributed This is a ratio of the mean of squared intensity to the mean
intensity (VI) intensity squared. It is applicable when pixel intensity has a
Gaussian distribution.
Normally distributed This is a ratio of mean intensity to the squared mean
amplitude (VA) amplitude. It is applicable when pixel amplitude has a
Gaussian distribution.
Lognormal distributed This is a difference of the mean value of the squared
intensity (VL) intensity logarithm and the square of the mean intensity
logarithm. It is applicable when pixel intensity has a
lognormal distribution.
K-distributed intensity (U) This is a normalized log measure of texture. It
approximates k-distribution when the number of looks in
the SAR image is high.

Exercise 3: Compare the different texture measures from SARTEX.

VI, intensity distributed normally VI, detail

PCI Geomatics Page 177

Appendix A Processing SAR data with Geomatica Banff

VA, amplitude distributed normally VA, detail

VL, intensity distributed lognormally VL, detail

U, intensity k-distributed U, detail

Figure 67. SARTEX results (7 x 7)

Page 178 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Creating texture images based on co-occurrence matrices by

using TEX
The TEX algorithm calculates a set of texture measures for all pixels in an input
image. The measures are based on second-order statistics computed from the
gray-level co-occurrence matrices. Texture measures for a specific direction or
directional-invariant can be computed. You can use the texture measures as input
for classification algorithms.

To open the TEX algorithm

To set up the parameters for TEX

1. Under Input Ports, expand Input: Image Layer.
2. From the Fle_FQ29_20100507sig_HH-HV.pix file, select the HH
channel.
3. Under Output Ports, expand Output: Output texture measure
Layer(s), select Untitled.pix, and then click Browse.
The File Selector window appears.
4. Open the Radar Data folder.
5. In the File name box, type Fle_FQ29_20100507sig_HH-
HV_TEX_11.pix, and then click Save.
6. Click the Input Params 1 tab.
7. Under TEXTURE MEASURE PARAMETERS, select the Homogeneity,
Variance, Entropy, Angular Second Moment, and Correlation check
boxes.
8. In the Filter X Size list, click 11.
9. In the Filter Y Size list, click 11.
10. In the Number of Gray Levels list, click 32.
11. Click Run.

PCI Geomatics Page 179

Appendix A Processing SAR data with Geomatica Banff

Exercise 4: Compare the different textures measured with TEX.

Homogeneity, 11 x 11, 32 gray levels Homogeneity, details

Page 180 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix A

Standard Deviation, 11 x 11, 32 gray levels Standard Deviation, details

Entropy, 11 x 11, 32 gray levels Entropy, details

Angular second moment, 11 x 11, 32 gray Angular second moment, details

levels

Figure 68. TEX results (11 x 11)

PCI Geomatics Page 181

Appendix A Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Calculated SAR-specific texture measures by using SARTEX
• Calculated texture measures based on co-occurrence matrices by
running TEX

Page 182 PCI Geomatics

Appendix B

Polarimetric discriminators
Polarimetric decompositions are useful in identifying the scattering mechanisms
characterizing a point or a distributed target. However, they only characterize a
part of the polarimetric information available for a given target. It is possible to
deepen the characterization of a target by analyzing its eigenvalues ( i , i=1, 2, 3)
produced by a polarimetric decomposition (Module 3).
The polarimetric response plots introduced in Module 3 and Module 4 can also be
used to identify the backscattering mechanisms characterizing a target. These
responses also provide an analysis of the backscatter power that is often
overlooked in the polarimetric decompositions.
It is particularly interesting to analyze the peaks and valleys of a polarimetric
response plot:
"the polarization plots have peaks at polarizations that give rise to maximum
received power, and valleys where the received power is smallest, in agreement
with the concept of Huynen's polarization fork in the Poincaré sphere"
—(CCRS, 2007, Boerner, et al. 1998, fig.5-3-9)
The polarimetric response plot corresponds to a projection of the Poincaré sphere,
where the orientation angle ( ) represent the longitude and the ellipticity angle (
 ) represents the latitude. Using an increment of one degree, there are more than
16,000 possible combinations of  and  for signal transmission, and a similar
number for reception (16,000 x 16,000 possible transmit-receive combinations).
Fortunately, the SAR Analysis toolbox provides algorithms to automatically analyze
the properties of a target using the concept of the Huynen fork on the Poincaré
sphere.
The PSPOLDIS algorithm calculates a number of polarimetric discriminators for a
fully polarimetric SAR (POLSAR) data set. The PSPOLSYN algorithm creates a
synthesized-backscatter SAR image for arbitrary transmit and receive polarizations.
The PSPOLSYNC algorithm creates a synthesized-backscatter image to maximize
the contrast between two targets.
Appendix B has four lessons:
• Generating polarimetric discriminators based on coherency matrix
eigenvalues
• Generating polarimetric discriminators based on analysis of the Poincaré
sphere

PCI Geomatics Page 183

Appendix B Processing SAR data with Geomatica Banff

• Synthesizing a backscatter SAR image for arbitrary transmit and receive

polarizations
• Maximize the contrast between two targets

Generating polarimetric discriminators from

coherency matrix eigenvalues
In this lesson, you will:
• Generate some polarimetric discriminators based on the coherency matrix
eigenvalues using the EASI Modeling feature in Focus
• Compare the resulting polarimetric discriminators
The entropy and the anisotropy are calculated from the three eigenvalues of the
coherency matrix. Several parameters can be determined from these eigenvalues
to perform an in-depth analysis of the scattering properties of the scene to be
examined. In this lesson, you will use the EASI Modeling feature to calculate a
series of polarimetric discriminators.

Data preprocessing
Before you can calculate polarimetric discriminators, you must preprocess the data.

To preprocess the data

1. Ingest into a pix file the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507 file
using product.xml as the key name file.
Select sigma as the calibration type, and then name the file
Fle_FQ29_20100507sig.pix.
2. Filter the ingested image using the PSBOXCAR algorithm.
Select a filter size of 7 x 7, and then name the file
Fle_FQ29_20100507sig_PSBOXCAR_7.pix.
3. Using Fle_FQ29_20100507sig_PSBOXCAR_7.pix as the input file,
apply the Cloude-Pottier decomposition (PSEBA).
4. Click the Input Params 1 tab, and then select the Retrieve
eigenvalues and eigenvectors check box.
5. Click the Files tab, and then name the output:
Fle_FQ29_20100507sig_PSBOXCAR_7_PSEABA.pix
6. From Fle_FQ29_20100507sig_PSBOXCAR_7_PSEABA.pix, extract
channels 1, 2, 5, 6 and 7.
Name the output Fle_FQ29_20100507sig_eigen.pix.
This new file should contain only five channels.

To calculate a series of polarimetric discriminators

1. In Focus, click the File menu, click Open, and then from the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder, open the Fle_FQ29_20100507sig_eigen.pix file.

Page 184 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

2. Right-click the Fle_FQ29_20100507sig_eigen.pix file, point to New,

and then click Raster Layer.
The Add Image Channels window appears.
3. Add seven (7) 32-bit real channels, and then click Add.
4. In Focus, click the Tools menu, and then click EASI Modeling.
The EASI Modeling window appears.
5. In the Input File list, click Fle_FQ29_20100507sig_eigen.pix.
6. Enter the following script, and then click Run.
Note: You can load the script from ~\SAR_Training\EASI_script\
AppendixB_PolarimetricDiscriminator.eas.

Exercise 1:
Compare the different eigenvalues based on polarimetric discriminators.
There is some correlation between anisotropy, entropy, and the polarimetric
discriminators based on the coherency matrix eigenvalues. Which polarimetric
discriminators are less correlated and which are more correlated?

Figure 69. EASI Modeling window, eigenvalues polarimetric discriminators

PCI Geomatics Page 185

Appendix B Processing SAR data with Geomatica Banff

Table 12. Correlation coefficients between the polarimetric discriminators (channel

description given in Figure 69, for entire image except open-water areas)

%1 %2 %6 %7 %8 %9 %10 %11

%2 -0.28 x x x x x x x

%6 -0.86 -0.19 x x x x x x

%7 0.80 0.73 0.37 x x x x x

%8 0.29 0.81 -0.59 0.18 x x x x

%9 0.79 -0.78 -0.46 -0.89 -0.35 x x x

%10 -0.89 0.61 0.65 0.84 0.15 -0.96 x x

%11 0.89 -0.61 -0.65 -0.84 -0.15 0.96 -1.00 x

%12 0.88 -0.49 -0.70 -0.76 -0.05 0.89 -0.98 0.98

Page 186 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

%1, entropy (H) %2, anisotropy (A) %6, dominant point target

%7, more than one %8, two strong scattering %9, fully diffused scattering
scattering mechanism mechanisms

%10,polarization fraction %11, radar-vegetation index %12, pedestal height

Figure 70. Polarimetric discriminators from the coherency matrix eigenvalues

PCI Geomatics Page 187

Appendix B Processing SAR data with Geomatica Banff

Lesson summary
In this lesson, you:
• Generated polarimetric discriminators based on the coherency matrix
eigenvalues
• Compared the polarimetric discriminators produced

Page 188 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

Generating polarimetric discriminators from analysis

of the Poincaré Sphere
In this lesson, you will:
• Generate different polarimetric discriminators by using the PSPOLDIS
algorithm
• Compare the polarimetric discriminators produced
PSPOLDIS calculates a number of polarimetric discriminators for a fully polarimetric
SAR (POLSAR) data set. Polarimetric discriminators are useful in characterizing the
features in your image by identifying the various types of scattering mechanisms.
The discriminators are based on the polarimetric synthesis and describe the
polarimetric response of features in the image.
These include:
• Minimum and maximum polarization
• Minimum and maximum completely polarized component
• Minimum and maximum completely unpolarized component
• Minimum and maximum received power
• Minimum and maximum scattered intensity
• Coefficient of variation
• Fractional polarization
The orientation and ellipticity angles for the extrema of the completely polarized
component are also written to the output file.
Because PSPOLDIS searches for the extrema of the response at every pixel, the
processing can take a long time, depending on the number of lines and pixels in the
image, and the step-size values. Reducing step-size values for orientation and
ellipticity increases the processing time, but makes the results more accurate. For
example, changing STEPPSI and STEPCHI from the default values of 10 degrees
to 5 almost quadruples the processing time, but the extrema determined are
accurate to within plus or minus 2.5 degrees.

PCI Geomatics Page 189

Appendix B Processing SAR data with Geomatica Banff

Table 13. PSPOLDIS parameters

No. Parameters Min. Max. Unit

Ratio of the intensity of the maximum polarized part
1 0 1 N/A
on total power
Ratio of the intensity of the minimum polarized part
2 0 1 N/A
on total power
Maximum intensity of the of the completely
3 0 N.A Intensity
polarized component
Orientation angle ( ) at the maximum of the
4 -90 90 Degree
completely polarized component
Ellipticity angle (  ) at the maximum of the
5 -45 45 Degree
completely polarized component
Minimum intensity of the completely polarized
6 0 N.A Intensity
component
Orientation angle ( ) at the minimum of the
7 -90 90 Degree
completely polarized component
Ellipticity angle (  ) at the minimum of the
8 -45 45 Degree
completely polarized component
Maximum intensity of the completely unpolarized
9 0 N.A Intensity
component
Minimum intensity of the completely unpolarized
10 0 N.A Intensity
component
11 Maximum of the received power 0 N.A Intensity
12 Minimum of the received power 0 N.A Intensity
13 Maximum of the scattered intensity 0 N.A Intensity
14 Minimum of the scattered intensity 0 N.A Intensity
15 Coefficient of variation 0 1 Unitless
16 Coefficient of fractional polarization 0 1 Unitless

To open the PSPOLDIS algorithm

To set up the parameters for PSPOLDIS

1. Under Input Ports, expand Input: Input polarimetric SAR image.
The tree expands to show the array segments.
2. Click Browse, open the folder
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507, and
then select the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Select each raster for this file.

Page 190 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

4. Under Output Ports, expand Output: Output polarimetric

discriminator, select Untitled.pix, and then click Browse.
The File Selector window appears.
5. Open the ~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29 folder.
6. Type FQ29_20100507sig_PSBOXCAR_7_PSPOLDIS.pix as the file
name.
7. Click Save.
The path and file name are updated in the PSPOLDIS Module Control Panel
window.
8. Click the Input Params 1 tab.
9. In the Orientation angle step size box and the Ellipticity angle step
size box, respectively, type 5.
10. Click Run.
The PSPOLDIS analysis runs and the results appear in the Focus viewer.

Exercise 2: Compare the various polarimetric discriminators generated by

PSPOLDIS.

% 3, Max. Int. C.P.C %6, Min. Int. C.P.C % 1, Int. Max. Pol.
part/total power

% 4,  at max. C.P.P % 5,  at max. C.P.P % 2, Int. Min. Pol. part/total

power

PCI Geomatics Page 191

Appendix B Processing SAR data with Geomatica Banff

%9, Max. Int. C.U.C %10, Min. Int. C.U.C %16, Coef. Frac. Pol.

Figure 71. Polarimetric discriminators from PSPOLDIS (selection)

Page 192 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

Lesson summary
In this lesson, you:
• Generated different polarimetric discriminators by using PSPOLDIS
• Compared the polarimetric discriminators produced

PCI Geomatics Page 193

Appendix B Processing SAR data with Geomatica Banff

Synthesizing a backscatter image for arbitrary

transmit and receive polarizations
In this lesson you will:
• Create synthesized backscatter SAR images for arbitrary transmit and
receive polarizations
The PSPOLSYN algorithm creates a synthesized backscatter SAR image for arbitrary
transmit and receive polarizations, from a fully polarimetric SAR (POLSAR) data set.
Creating an image with polarizations that are different from the original ones may
reveal information that is not apparent in the input data set. The input file must
represent a nonsymmetrized or symmetrized, fully polarimetric (quad-polarization)
SAR data set in the scattering (s4c or S3c), covariance (c4r6c or C3r3c), coherency
(t4r6c or T3r3c), or Kennaugh (k16r or K9r) matrix format.
In Figure 72, an urban target, corresponding to a building not aligned with the
radar line of sight (LOS), has been selected. This misalignment has created a
stronger backscattering in HV, but the backscattering in HH might be lower than
what you would expect for this kind of target.
Target
coordinates
5°38'08.9''E;
52°33'57.8N
HH=0.307
(-5.12dB)
HV=0.193
(-7.14 dB)
VV=0.246
(-6.09 dB)
Nb. pixels = 44

R: HH, G:HV, B:VV Target in red

(R=HH, G=HV, B=VV)

Figure 72. Urban target

In the previous lesson, you used PSPOLDIS to characterize the backscattered power
of the FQ29_20100507sig_PSBOXCAR_7.pix scene. In this lesson, you will try to
enhance the backscattered power by generating a synthesized backscatter SAR
image that corresponds to the target’s orientation ( ) and ellipticity (  ) angles at
the maximum of the completely polarized component (Table 14).

Page 194 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

Table 14. PSPOLDIS parameters (selection) for a small urban target

Description Mean (°) Med (°) S-D (°)

Ellipticity angle (  ) at the maximum of the
46.1 44.9 11.8
completely polarized component
Orientation angle ( ) at the minimum of
12.5 10.0 5.9
the completely polarized component
Ellipticity angle (  ) at the minimum of the
-57.0 -60.0 18.4
completely polarized component
Ellipticity angle (  ) at the maximum of the
-12.9 -10.0 13.4
completely polarized component

To open the PSPOLSYN algorithm

To set up the parameter for PSPOLSYN

1. Under Input Ports, expand Input: Input polarimetric SAR image.
The tree expands to show the array segments.
2. Click Browse, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder, and then select the FQ29_20100507sig_PSBOXCAR_7.pix file.
3. Select each raster for this file.
4. Under Output Ports, expand Output: Output synthesized detected
SAR image, select Untitled.pix, and then click Browse.
The File Selector window appears.
5. Open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
6. In the File name box, type
FQ29_20100507sig_PSBOXCAR_7_PSPOLSYN_t1.pix.
7. Click Save.
The path and file name are updated in the PSPOLSYN Module Control
Panel window.
8. Click the Input Params 1 tab.
9. In the Transmitted orientation angle box and the Transmitted
ellipticity angle box, type 46.1 and 12.5, respectively.
10. In the Received orientation angle box and the Received ellipticity
angle box, type 46.1 and 12.5, respectively.

PCI Geomatics Page 195

Appendix B Processing SAR data with Geomatica Banff

11. Click Run.

The PSPOLSYN algorithm runs and the results appear in Focus.

Exercise 3: A copolarized (transmit=receive) image has been generated. You will

synthesize a second image using the minimum of the completely polarized
component angles.
Repeat the previous steps with the following modifications:
• In step 5, specify RS2_FQ29_Fle_20100507sig_PSPOLSYN_t2.pix
• In step 9, specify -57.0 and -12.9 for the Received orientation angle
and Received ellipticity angle boxes, respectively

Page 196 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

T1
T :46.1

T  :12.5

R :46.1

R  :12.5

Mean = 1.25
(0.96 dB)

T2
T :46.1

T  :12.5

R :-57.04

R  :-12.95

Mean = 0.12
(-9.2dB)

T3
T :-57.0

T  :-12.9

R :-57.0

R  :-12.9

Mean = 0.22
(-6.57dB)

Figure 73. Synthesized backscatter SAR images for various transmit and receive
polarization configurations

PCI Geomatics Page 197

Appendix B Processing SAR data with Geomatica Banff

Lesson summary
In this lesson you:
• Created synthesized backscatter SAR images for arbitrary transmit and
receive polarizations

Page 198 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

Maximizing the contrast between two targets

In this lesson you will:
• Create a synthesized backscatter image to maximize the contrast between
two targets
The PSPOLSYNC algorithm creates a synthesized backscatter image from a fully
polarimetric SAR data set. The polarizations used for synthesis are selected
automatically by PSPOLSYNC to maximize the contrast (ratio) between two targets.
The polarizations are determined by searching the full range of the orientation and
ellipticity angles for the polarizations transmitted and received. The identified
polarizations are then used to synthesize the radar backscatter at each pixel in the
input data set. The synthesis algorithm is the same as that of PSPOLSYN.
The input file must represent a nonsymmetrized or symmetrized fully polarimetric
(quad-polarization, complex) data set in the scattering (s4c or S3c), covariance
(c4r6c or C3r3c), coherency (t4r6c or T3r3c), or Kennaugh (k16r or K9r) matrix
format.
In the following procedure, you will compare the urban target described in the
previous lesson to a forest target.
Details
Forest target:
HH=0.140
(-8.53dB)
HV=0.054
(-12.67 dB)
VV=0.142
(-8.47 dB)
Nb. pixels = 44

R: HH, G:HV, B:VV Target in red: Urban,

in blue: Forest
R:HH, G:HV, B:VV

Figure 74. Targets used as input for the PSPOLSYNC algorithm

PCI Geomatics Page 199

Appendix B Processing SAR data with Geomatica Banff

Create the two targets

Add two to new bitmap layers to FQ29_20100507sig_PSBOXCAR_7.pix
(In FOCUS, click on the file tab, right click on
FQ29_20100507sig_PSBOXCAR_7.pix ->New->Bitmap Layer)

1. Name the first bitmap layer “urban”, name the second bitmap layer “forest”
2. Load the new empty bitmap in FOCUS. Right-click on the layer and then
select view.
3. Click on the FOCUS Maps tab
4. Highlight the “urban” bitmap layer and draw a region similar to the one
presented at figure 74.
Save your “urban” bitmap layer.
5. Repeat step 5 and draw the “forest” bitmap layer.

To open the PSPOLSYNC algorithm

To set up the parameter for PSPOLSYNC

1. Under Input Ports, expand Input: Input polarimetric SAR image.
The tree expands to show the array segments.
2. Click Browse, open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder, select the FQ29_20100507sig_PSBOXCAR_7.pix file, and then
select each raster for this file.
3. Under Input Ports, expand InputTarget1: First input target layer.
4. Select the urban bitmap layer.
5. Under Input Ports, expand InputTarget: Second input target layer.
6. Select the forest bitmap layer.
7. Under Output Ports, expand Output: Output synthesized detected
SAR image, select Untitled.pix, and then click Browse.
The File Selector window appears.
8. Click Browse, and then open the
~\SAR_Training\Radar\Flevoland_RS2_SLC\FQ29_20100507
folder.
9. In the File Name box, type
FQ29_20100507sig_PSBOXCAR_7_PSPOLSYNC_t1.pix.

Page 200 PCI Geomatics

Processing SAR data with Geomatica Banff Appendix B

10. Click Save.

The path and file name are updated in the PSPOLSYNC Module Control
Panel window.
11. Click Run.
The PSPOLSYNC analysis runs and the results appear in the Focus viewer.

To view the metadata

1. In Focus, click the Files tab.
2. Right click the
FQ29_20100507sig_PSBOXCAR_7_PSPOLSYNC_t1.pix file, and then
click Properties.
The File Properties window appears.
3. Click the MetaData tab.
The metadata shows that the maximum contrast between the two targets
is obtained with the following configuration:
Optimum Contrast Receive Ellipticity: 26.4°
Optimum Contrast Receive Orientation: 7.5°
Optimum Contrast Transmit Ellipticity: 16.0°
Optimum Contrast Transmit Orientation: 11.0°.
Details
Red: Target 1,
urban
0.464 (-3.33dB)

Blue: Target 2,
forest
0.054 (-12.7dB)

Figure 75. Synthesized backscatter image maximizing the contrast between the two
targets

PCI Geomatics Page 201

pcigeomatics.com
[email protected]

Oracle Exadata Hardware Installation & Software Configuration Online Assessment (2021)
No ratings yet
Oracle Exadata Hardware Installation & Software Configuration Online Assessment (2021)
40 pages
10 STPP 3D Coordinate Transformation
No ratings yet
10 STPP 3D Coordinate Transformation
72 pages
WGS 84 (G2296)
No ratings yet
WGS 84 (G2296)
10 pages
WASSP Spec Sheet S3 Final
No ratings yet
WASSP Spec Sheet S3 Final
2 pages
SATLAB Field Tools Manual SL55
No ratings yet
SATLAB Field Tools Manual SL55
18 pages
Operation Manual (NTS-320 Series Total Station)
No ratings yet
Operation Manual (NTS-320 Series Total Station)
140 pages
Fitzpatrick Dermatology
100% (4)
Fitzpatrick Dermatology
2,576 pages
EnvisionProject Quick Start Guide
100% (1)
EnvisionProject Quick Start Guide
48 pages
TrainingGuide Geomatica OrthoEngine PDF
No ratings yet
TrainingGuide Geomatica OrthoEngine PDF
174 pages
ADS80 Datasheet en
No ratings yet
ADS80 Datasheet en
2 pages
Geodesy Chapter 3-2
No ratings yet
Geodesy Chapter 3-2
20 pages
Manual Topcon Link
No ratings yet
Manual Topcon Link
240 pages
Transformation Between ITRF2000 and WGS84
100% (1)
Transformation Between ITRF2000 and WGS84
12 pages
GPS 3 Upd PDF
100% (1)
GPS 3 Upd PDF
100 pages
Pengenalan Pci Geomatic
No ratings yet
Pengenalan Pci Geomatic
6 pages
03 Sistem Penentuan Posisi Laut I
No ratings yet
03 Sistem Penentuan Posisi Laut I
37 pages
inSAR Processing For DEM Generation PDF
No ratings yet
inSAR Processing For DEM Generation PDF
8 pages
Tugas 01 KL 3103 Pengumpulan Dan Data Analisis Lapangan: Irsan Soemantri B. PH.D
No ratings yet
Tugas 01 KL 3103 Pengumpulan Dan Data Analisis Lapangan: Irsan Soemantri B. PH.D
8 pages
Tmatch Compressed
No ratings yet
Tmatch Compressed
122 pages
Wingtra Technical Specifications PDF
No ratings yet
Wingtra Technical Specifications PDF
20 pages
1 - ERDAS IMAGINE Viewer PDF
No ratings yet
1 - ERDAS IMAGINE Viewer PDF
20 pages
Automatic Mosaicking PC Geomatics
No ratings yet
Automatic Mosaicking PC Geomatics
13 pages
Report Practical 3 Remote Sensing
No ratings yet
Report Practical 3 Remote Sensing
27 pages
6 Configuring F180
No ratings yet
6 Configuring F180
7 pages
Sistem Referensi Geodetik Dan Penentuan Posisi Di Laut Marine Kadaster
100% (1)
Sistem Referensi Geodetik Dan Penentuan Posisi Di Laut Marine Kadaster
49 pages
Modul-1:: Introduction To Satellite Geodesy
No ratings yet
Modul-1:: Introduction To Satellite Geodesy
79 pages
KULIAH 2 MRT 10-Perpetaan
No ratings yet
KULIAH 2 MRT 10-Perpetaan
85 pages
Metashape Presentation
100% (1)
Metashape Presentation
18 pages
Structure From Motion Photogrammetry
No ratings yet
Structure From Motion Photogrammetry
12 pages
Gps 3 Upd PDF
No ratings yet
Gps 3 Upd PDF
94 pages
MagArrow Spec Sheet
No ratings yet
MagArrow Spec Sheet
2 pages
CE-321 Sensors & Platform
No ratings yet
CE-321 Sensors & Platform
87 pages
Image Enhancement: Dr. Abhishek Rawat
No ratings yet
Image Enhancement: Dr. Abhishek Rawat
41 pages
Pix4D White Paper How Accurate Are UAV Surveying Methods
No ratings yet
Pix4D White Paper How Accurate Are UAV Surveying Methods
8 pages
Gps Satellite Surveying
No ratings yet
Gps Satellite Surveying
24 pages
2006 (Luhmann) Close Range Photogrammetry
No ratings yet
2006 (Luhmann) Close Range Photogrammetry
57 pages
Bathy 500 MF
No ratings yet
Bathy 500 MF
58 pages
Download
No ratings yet
Download
48 pages
Tmatch
No ratings yet
Tmatch
44 pages
Theory of Regionalized Variable PDF
No ratings yet
Theory of Regionalized Variable PDF
18 pages
Advanced Network Adjustment - Leica Infinity
No ratings yet
Advanced Network Adjustment - Leica Infinity
18 pages
Jo Wood (1996), The Geomorphological Characterisation of Digital Elevation Models (1-3)
No ratings yet
Jo Wood (1996), The Geomorphological Characterisation of Digital Elevation Models (1-3)
81 pages
Hemisphere GNSS Technical Reference Guide v1.09
No ratings yet
Hemisphere GNSS Technical Reference Guide v1.09
586 pages
3DSurvey Brosura WEB
No ratings yet
3DSurvey Brosura WEB
16 pages
Change Detection Analysis
No ratings yet
Change Detection Analysis
4 pages
ASPRS Positional Accuracy Standards Edition1 Version100 November2014 PDF
No ratings yet
ASPRS Positional Accuracy Standards Edition1 Version100 November2014 PDF
26 pages
Lab 1 DCG40163 Remote F2019
No ratings yet
Lab 1 DCG40163 Remote F2019
23 pages
How To Install Gamit
No ratings yet
How To Install Gamit
5 pages
Intro To ArcGISPro 2019
No ratings yet
Intro To ArcGISPro 2019
8 pages
OSI-QTN - 000332 - 04 - 2024 - PT OGS Indonesia (Norbit IWMBS Multibeam Sonar)
No ratings yet
OSI-QTN - 000332 - 04 - 2024 - PT OGS Indonesia (Norbit IWMBS Multibeam Sonar)
2 pages
Building Web Based Gis Applications With GeoExt
No ratings yet
Building Web Based Gis Applications With GeoExt
10 pages
GRS 80 Moritz PDF
No ratings yet
GRS 80 Moritz PDF
35 pages
Tutorial - Workshop Introduction To Digital Mapping Using QGIS
No ratings yet
Tutorial - Workshop Introduction To Digital Mapping Using QGIS
15 pages
Economical All-Weather Total Station
No ratings yet
Economical All-Weather Total Station
4 pages
GLS 611 / GSS611 Advanced Geodesy Physical and Satellite Geodesy
No ratings yet
GLS 611 / GSS611 Advanced Geodesy Physical and Satellite Geodesy
27 pages
Strataboxmanual8 25 06b Syq PDF
No ratings yet
Strataboxmanual8 25 06b Syq PDF
54 pages
Viewer Tutorial
No ratings yet
Viewer Tutorial
107 pages
LGO Quick Start Guide - Format Manager
0% (1)
LGO Quick Start Guide - Format Manager
5 pages
Hi Survey User Manual en 20230428
100% (1)
Hi Survey User Manual en 20230428
314 pages
TrainingGuide SAR Processing With Geomatica 3
No ratings yet
TrainingGuide SAR Processing With Geomatica 3
210 pages
RS2 Tutorial
No ratings yet
RS2 Tutorial
14 pages
PCI Geomatica - Palsar Tutorial
No ratings yet
PCI Geomatica - Palsar Tutorial
12 pages
TrainingGuide Geomatica 2
No ratings yet
TrainingGuide Geomatica 2
169 pages
HM-12/HM-13 Firmware Upgrade Instructions: Firmware Upgrade May Damage The Module Boot Loader System, Please Use Caution
No ratings yet
HM-12/HM-13 Firmware Upgrade Instructions: Firmware Upgrade May Damage The Module Boot Loader System, Please Use Caution
5 pages
Report On Building A Custom Payment Gateway System 2023
No ratings yet
Report On Building A Custom Payment Gateway System 2023
13 pages
Cs-Xi Aps
No ratings yet
Cs-Xi Aps
5 pages
Html5 & CSS: 8 Edition
No ratings yet
Html5 & CSS: 8 Edition
48 pages
Ec 411
No ratings yet
Ec 411
2 pages
Update Instruction - UV88
No ratings yet
Update Instruction - UV88
6 pages
App Develop
No ratings yet
App Develop
1 page
Cense Cakibration Manul
100% (1)
Cense Cakibration Manul
9 pages
Notula Rapat Bulan Desember 2021
No ratings yet
Notula Rapat Bulan Desember 2021
40 pages
Telefire ADR-3000 Brochure PDF
No ratings yet
Telefire ADR-3000 Brochure PDF
2 pages
PXC 3871183
No ratings yet
PXC 3871183
8 pages
The Integration of A Unified Permit License System For Engineering Office
No ratings yet
The Integration of A Unified Permit License System For Engineering Office
8 pages
Constructors 1
No ratings yet
Constructors 1
12 pages
Cubic Sequences Worksheet
100% (1)
Cubic Sequences Worksheet
3 pages
PYTHON
No ratings yet
PYTHON
18 pages
03a EasyIO FG Series User Reference v2.0
No ratings yet
03a EasyIO FG Series User Reference v2.0
20 pages
T Sanchez Resume Weebly 2
No ratings yet
T Sanchez Resume Weebly 2
1 page
Introduction
No ratings yet
Introduction
51 pages
Shubham Prashar (SDE1)
No ratings yet
Shubham Prashar (SDE1)
1 page
Steel Import License Application
No ratings yet
Steel Import License Application
6 pages
SPF Admin Course - Modeling and Mapping PDF
100% (1)
SPF Admin Course - Modeling and Mapping PDF
778 pages
Narrative Report
No ratings yet
Narrative Report
12 pages
Deployment Options
No ratings yet
Deployment Options
31 pages
Manual Parameter 828D
No ratings yet
Manual Parameter 828D
898 pages
Inferences Worksheet 3
No ratings yet
Inferences Worksheet 3
15 pages
Installation Instructions: Hotwire 6205 ADSL Modem
No ratings yet
Installation Instructions: Hotwire 6205 ADSL Modem
15 pages

SAR Processing Training

Uploaded by

SAR Processing Training

Uploaded by

Processing SAR data with

PCI Geomatics Page iii

Version Released Revisions Revised by

PCI Geomatics Page v

Module 1: SAR data structure and support in Geomatica 16

Module 2: Applying radiometric enhancements 45

Module 3: Polarimetric decompositions 63

Module 4: Using the SAR Polarimetry Target Analysis tool 99

Module 5: Classifying SAR data 121

Module 6: Change detection using polarimetric data 135

PCI Geomatics Page vii

Page viii PCI Geomatics

About this guide

PCI Geomatics Page 9

Table 1. Key Geomatica programs for working with SAR data

Page 10 PCI Geomatics

Figure 1. Workflow diagram

PCI Geomatics Page 11

Table 2. Algorithms for processing SAR data

Process Algorithm Description

Speckle filtering PSBOXCAR POLSAR Boxcar filter (Module 2)

Page 12 PCI Geomatics

Process Algorithm Description

Texture analysis SARTEX SAR texture analysis (Appendix A)

PCI Geomatics Page 13

Process Algorithm Description

Page 14 PCI Geomatics

Lesson 1.1: Supported SAR sensors and data formats

PCI Geomatics Page 16

The channel type is divided in two subcategories:

Table 3. Channel types

Channel type Subtype Example

The transmit/receive configuration refers to the polarization or polarizations

PCI Geomatics Page 17

Table 4. Configuration types

Configuration Channel type Details

ex.2: ⃗𝑇𝑥 (V), 𝑅⃗𝑥(V) → VV

Module 3 and Appendix B provide more information on the concept of polarization.

Page 18 PCI Geomatics

Lesson 1.2: Data structure and matrix type

PCI Geomatics Page 19

Figure 2. Radarsat-2, Single Look Complex (SLC), Fine quad-polarization, beam 29

Figure 3. Sentinel-1A, Ground-range detected (GRD), Stripmap, (SM) dual-

Page 20 PCI Geomatics

Figure 4. UAVSAR, Calibrated multi-look complex (MLC), Polarimetric mode,

Figure 5. Viewing vendor-provided GCPs in Focus

PCI Geomatics Page 21

Figure 6. Selecting georeferencing source

Figure 7. Select georeferencing layer

Note If an image is already open in Focus, but with a different

Page 22 PCI Geomatics

Figure 8. North Up or Raster Up selection

PCI Geomatics Page 23

Table 5. Matrix types: complex data

Symbol Full name Total no. of channels

Table 6. Matrix types: detected data

Symbol Full name Total no. of Channels

Page 24 PCI Geomatics

Lesson 1.3: Complex data support in Geomatica

Figure 9. Radarsat-2, Fine quad-polarization, beam 29 (FQ29), May 7, 2010

Figure 10. Radarsat-2, Fine quad-polarization, beam 29 (FQ29), May 7, 2010

PCI Geomatics Page 25

To change the data interpretation of complex images

Figure 11. Interpretation options for complex data

Numeric Values window

To open the Numeric Values window

Page 26 PCI Geomatics

Figure 12. Numeric Values window: Raw (Intensity)

Figure 13. Numeric Values window: Complex data

PCI Geomatics Page 27

Other benefits of complex data support in Geomatica include the following:

Page 28 PCI Geomatics

Lesson 1.4: Ingesting and extracting a calibrated

Ingesting SAR data

To ingest a file using the Import to PCIDSK option

Ingesting a file with SARINGEST