Remote Sensing and Image Classification Methods

Prashant Kumar Mali

3rd year Civil Engineering student
191010012016
Institute of Infrastructure Technology, Research
And Management Ahmedabad, India
[email protected]

I. O BJECTIVES
The objectives of this assignment are
• To provide an overview of Remote Sensing and GIS,
• To learn about various steps of Image Processing,
• To learn about Image Restoration and Enhancement,
• To understand various Methods of Image classification
and improving overall accuracy using these classification
methods.
II. I NTRODUCTION
Remote sensing refers to obtaining information about an
object, area, or phenomenon through the analysis of data
acquired by a device that is not in contact with the object,
area, or phenomenon under investigation Fig. 1. Bhuvan, is an Indian web based utility provides a set of map based
Remote sensing systems are a very important source of in- content prepared by ISRO
formation for GIS, as they provide access to spatio-temporal
information on surface processes on scales ranging from
regional to global.
The spatial data generally is in the form of maps, which
could be showing topography, geology, soil types, forest and
vegetation, land use, water resource availability etc., stored as
layers in a digital form. Integrating many layers of data in a
computer can easily generate new thematic maps.
III. A N OVERVIEW OF GIS
The Geographic Information System (GIS) based on remote
sensing is a computer-assisted system for acquisition, storage,
analysis and display of geographic data. GISs can be built
using open source software, proprietary (license required) soft-
ware, or a combination of the two. Open source GIS software Fig. 2. Land Use Land Cover data of the Ahmedabad city
is rapidly improving and can provide a robust alternative to
proprietary software Quantum GIS (QGIS) is an open source
Geographic Information System that supports most geospatial
vector and raster file types and database formats. The program
offers standard GIS functionality, with a variety of mapping
features and data editing.
IV. I MAGE P ROCESSING
Image processing is the process of transforming an image
into a digital form and performing certain operations to get
some useful information from it. Digital Image Processing is
largely concerned with four basic operations: Image Restora- Fig. 3. Processing toolbar and multiple functions of QGIS
tion, Image Enhancement, Image Classification, and Image
Transformation.
 Fig. 6. Components of a digital image

B. Image Restoration
Fig. 4. Covid Health Care Indicator Maps created using QGIS. Remotely sensed images of the environment are typically
taken at a great distance from the earth’s surface. As a result,
there is a substantial atmospheric path that electromagnetic
energy must pass through before it reaches the sensor. During
the time the image is being scanned, the satellite is following
a path that is subject to minor variations at the same time that
the earth is moving underneath as shown in figure.
A variety of systematic and apparently random disturbances
can combine to degrade the quality of the image we finally
receive. Image restoration seeks to remove these degradation
effects.
Broadly, image restoration can be broken down into the two
sub-areas radiometric and geometric restoration.
1. Radiometric restoration: removal or diminishment of
Fig. 5. Remote Sensing Process distortions in the degree of electromagnetic energy registered
by each detector. A variety of agents can cause distortion in
the values recorded for image cells such as
A. Digital Image • Uniformly elevated values, due to atmospheric haze,
which preferentially scatters short wavelength bands (par-
An image is represented by its dimensions (height and ticularly the blue wavelengths);
width) based on the number of pixels. This pixel is a point on • Striping, due to detectors going out of calibration;
the image that takes on a specific shade, opacity or color. It • Random noise, due to unpredictable and unsystematic
usually represented in Grayscale, RGB. performance of the sensor or transmission of the data;
Digital Image may consists of and
• Spatial component describes the quality of an image and • Scan line drop out, due to signal loss from specific
how detailed objects are in an image. If the grid cells are detectors.
smaller, this means the spatial resolution has more detail 2. Geometric restoration: For mapping purposes, it is es-
with more pixels. sential that any form of remotely sensed imagery be accurately
• Spectral Component describes the amount of spectral registered to the proposed map base.
detail in a band. High spectral resolution means its bands
• There are not only systematic distortions related to tilt
are more narrow.
and varying altitude, but variable topographic relief leads
• Radiometric Component corresponds to the sensitivity
to very irregular distortions (differential parallax) that
of a sensor, i.e. its ability to measure and to enable
cannot be removed easily.
distinction within the same spectral band of differences
• With satellite imagery, the very high altitude of the
in the electromagnetic energy reflected by the elementary
sensing platform results in minimal image displacements
ground surfaces.
due to relief.
• Temporal Component corresponds a series of images
• Relief displacement is the radial distance between where
taken at different time to monitor the dynamic changes
an object appears in an image to where it actually should
of the objects
be according to a Planimetric coordinate system.
 • In these instances, it is necessary to use photogrammet-
ric rectification to remove these distortions and provide
accurate map.

C. Image Enhancement
Image enhancement is concerned with the modification
of images to make them more suited to the capabilities of
human vision. Typical image enhancement techniques include
following which are usually applied to the image output for
image interpretation.
• Gray scale conversion,
• Histogram equalization,
• Color composition, Fig. 7. Radiometric Restoration and Geometric Restoration
• Image smoothing
• Image sharpening,
• Noise removal,
• Contrast adjustment,
• Color conversion between RGB and HSI, etc.,

D. Image Classification
Image classification is the process of categorizing and
labeling groups of pixels or vectors within an image based
on specific rules. The categorization law can be devised using
one or more spectral or textural characteristics.
There are two types of image classification:
1. Supervised Classification
• Identify training set (information class),
• Select sample pixels in an image representing specific
class,
• Direct the image processing software to use training sets
as references for the classification of all other pixels in
the image. Fig. 8. Images at different Contrast levels
2. Unsupervised Classification
Unsupervised classification requires no advance information
about the classes of interest. Rather, it examines the data and
breaks it into the most prevalent natural spectral groupings, or
clusters, present in the data.
k-means Clustering
k-means is a commonly used partitioning based clustering
technique that tries to find a user specified number of clusters
(k), which are represented by their centroids, by minimizing
the square error function. Although k-means is simple and can
be used for a wide variety of data types.
The steps of the k-means algorithm are written below:
• Initialization: choose randomly k pixel vectors (data
points) to initialize the clusters.
• Nearest-neighbor search: for each input vector, find the
cluster center that is closest, and assign that input vector
to the corresponding cluster.
• Mean update: update the cluster centers in each cluster
using the mean (centroid) of the input vectors assigned
to that cluster.

Fig. 9. Images at different exposures and other enhancement parameters

 Fig. 12. Image Segmentation
Fig. 10. Supervised Classification

V. DATASET

Hyperspectral Image is considered in the form of a ma-

trix which comprises information in rows and columns with
number of pixels and bands.
Hyperspectral Data has been used for classification due to
following:
• Bands are continues and regularly spaced.
• Higher spectral resolution, thus giving the opportunity to
push (Spectral resolution is 10 nm).
• further the information extraction capability.
Fig. 11. Unsupervised Classification • Possibility of more accurate discrimination among mate-
rials of interest.

E. Image Segmentation Salinas scene was collected by the 204-band AVIRIS sensor
over Salinas Valley, California, and is characterized by high
Image segmentation is the process of assigning a label spatial resolution (3.7-meter pixels). The area covered com-
to every pixel in an image such that pixels with the same prises 512 lines by 217 samples. This image was available
label share certain characteristics. The result of image only as at-sensor radiance data. It includes vegetables, bare
segmentation is a set of segments that collectively cover soils, and vineyard fields. Salinas ground truth contains 16
the entire image, or a set of contours extracted from the image. classes.

Each of the pixels in a region is similar with respect to some

characteristic or computed property, such as color, intensity,
or texture. Adjacent regions are significantly different with
respect to the same characteristic.
The steps for performing Image Segmentation are as follow:
• Organize the dataset in a matrix X.
• Perform k-means clustering.
• Obtain cluster map.
• Convert the cluster map into segmentation map using
connected component.
• Apply k-means clustering.
 VI. R ESULTS
Using this dataset, k-means clustering is performed
and NMI Normalised Mutual Information is calculated,
which represent overall accuracy of image. We have started
with different value of k and found optimum value of k
corresponding to which overall accuracy is maximum.
Similarly, we have applied image segmentation on the same
data set after k-means clustering and tried to find optimumn
value of k seg for which overall accuracy is maximum.

Hence, Utilizing k-means clustering and image segmenta-

tion we have improve overall accuracy from 72.42 to 79.51
percentage.

VII. C ONCLUSION
Remotely sensed data is important to a broad range of Fig. 13. Classifiaction Using kmeans
disciplines. This will continue to be the case and will likely
grow with the greater availability of data promised by an
increasing number of operational systems.
In this seminar we have discusses several methods image
classification, including supervised and unsupervised classifi-
cation.
Although the supervised and unsupervised classification
methods have their respective advantages to varying degrees,
there are limitations in the application of various methods.
For example, supervised classification requires a certain
number of prior conditions, and human factors will affect the
classification results have an impact.
Using unsupervised classification and image segmentation
on hyperspectral images we have improved overall accuracy
of given dataset.
VIII. R EFERENCES
[1] Anand Mehta, “Segmentation based Projected Clustering Fig. 14. Classifiaction using kmeans + Image Segmentation
of Hyperspectral Images”,
[2] Dr. S. C. Liew , “Principles Of Remote Sensing”,
Centre for Remote Imaging, Sensing and Processing National
University of Singapore,
[3] Bhuvan - Indian Geo Platform of ISRO,
https://bhuvan.nrsc.gov.in/
[4] https://www.sciencedirect.com/topics/engineering/image-
processing
[5] https://www.researchgate.net/

Fig. 15. Outputs of the above algorithms

Fig. 16. Final Result