GSS612

REMOTE SENSING & IMAGE PROCESSING

Lab Practical Title:

ACCURACY ASSESSMENT
Student name : PUTERA ZIRFAN SYAMIR BIN MOHD ZAKI
Student ID : 2022825002
Lecturer Name : ASSOC. PROF. DR NOR AIZAM BINTI ADNAN
Date : 12th JULY 2023

Marks Distribution:

No. Lab Title Mark Mark each report

5
1 Introduction
5
2 Objectives and study area
15
3 Methods & Procedures
20
4 Results /Questions answer
Conclusion
5 5
50
Total mark
 TABLE OF CONTENT

1.0 INTRODUCTION .............................................................................................................. 1

2.0 OBJECTIVES AND STUDY AREA ...................................................................................... 2

2.1 Objectives .................................................................................................................... 2

2.2 Study Area ................................................................................................................... 2

3.0 METHODOLOGY.............................................................................................................. 3

3.1 Sample Points .............................................................................................................. 4

3.2 Accuracy Assessment .................................................................................................. 7

4.0 RESULT/ANSWERED QUESTION ..................................................................................... 9

4.1 Result of Accuracy Assessment Report ....................................................................... 9

4.1.1 Unsupervised Classification of Image sub040505 ............................................... 9

4.1.2 Unsupervised Classification of Image sub220208 ............................................. 11

4.1.3 Supervised Classification of Image sub040505 .................................................. 13

4.1.4 Supervised Classification of Image sub220208 .................................................. 15

4.2 Answered Question ................................................................................................... 17

5.0 CONCLUSION ................................................................................................................ 19

6.0 REFERENCES .................................................................................................................. 20

1.0 INTRODUCTION

The accuracy assessment feature in Erdas Imagine 2014 allows users to assess the
quality of their categorization findings by comparing them to ground truth data. Ground truth
data is typically obtained by field surveys or high-resolution imaging and consists of a set of
reference points or regions with known class names. Here's an overview of how Erdas Imagine
can be used to assess accuracy:

Prepare ground truth data by collecting or creating a set of ground truth data
containing the true class labels for different picture regions. This can be done by either
manually digitizing the classes of interest or by using existing reference data.

Image Classification, in Erdas Imagine, classifies the image using a classification

technique such as maximum likelihood, support vector machines (SVM), or random forests.
This technique assigns class names to distinct image pixels based on their spectral properties.

Conduct the accuracy evaluation by comparing the categorized image to the ground
truth data using the accuracy evaluation methodology. Erdas Imagine will create a report or
output comprising accuracy metrics and statistical statistics that indicate the quality of the
categorization results.

Analyze the accuracy findings and analyze the accuracy assessment results to gain a better
understanding of the classification's performance. Other measures, such as user accuracy,
producer accuracy, and kappa coefficient, provide more information about categorization
accuracy for different classes.

Consider refining the classification process by altering the classification method,

training data, or spectral signatures if the accuracy evaluation reveals low accuracy for specific
classes or overall. Users can analyze the dependability of categorization results and make
educated decisions about their application in fields such as land cover mapping,
environmental monitoring, and urban planning by using accuracy assessment.

1
2.0 OBJECTIVES AND STUDY AREA

2.1 Objectives

 To determine how correctly the classified photographs, depict parts of reality and to assess
the accuracy of the images themselves.
 To comprehend the categorized photos and perform accurate judgements using the
information provided.

2.2 Study Area

This task's study region is based on the lecturer's data, which is Landsat Images (sensor
by US): Landsat 8 TM. The information is provided in the ERDAS IMAGINE document
(sub040505.img), RRD file (sub040505.rrd), and (sub220208.img), RRD file (sub220208.rrd).
This task will also make use of prior lab 5 (Unsupervised Classification) and lab 6 (Supervised
Classification) final output.

Figure 2.0 Location of Study Area (Erdas Imagine Software)

2
3.0 METHODOLOGY

The three elements of the technique for this lab are sample points, accuracy assessment, and
examination report. Erdas Imagine 2014 makes use of all of these procedures. The flow of work
accuracy assessment is depicted in Figure 3.0 below.

Sample
Points

Accuracy
Assessment

Examine
Report
Figure 3.0 Flow of Work

3
3.1 Sample Points

Launch Erdas Imagine 2014. Start by loading the image sub040505.img and
unsupervised sub040505(isodata).

Figure 3.1 Images of sub04050(right) and unsupervised 040505” isodata” (left)

To access the "Accuracy Assessment" tool, select the Raster tab, then select
Supervised, and then select Accuracy Assessment. Launch the "Accuracy Assessment"
application and load the image unsupervised sub040505(isodata).

4
 Figure 3.2 Accuracy Assessment Tools

Go to the Edit menu and select the Create/Add Random Points option. Insert "Number
of Points" and choose “Distribution Parameters" from the drop-down menu.

Figure 3.3 Adding Random Points

5
Choose the viewer before the points are displayed. Click inside the viewer that
contains the image unsupervised sub040505(isodata) to evaluate after going to the
"Select Viewer" menu option and selecting it. Display each point in the image. Click
the View menu, then Show All. On the image that been “select viewer”, the points will
be presented.

Figure 3.4 ID Points of Random Points (left)

Click the View menu and select Change Colors. Launch the viewer and open the
reference image. In the "Reference" column, write the class code that corresponds to
each point using the reference image as a guide. For instance, the graphic below shows
that point ID#1 is in a wooded region. Therefore, "2" should be used as the class code
for the reference.

6
 Figure 3.5 Accuracy Assessment Table

3.2 Accuracy Assessment

To produce the report, select Report > Accuracy Report from the menu bar.

Figure 3.6 Produce report

7
Look further down in the report. The overall correctness of your categorization will be
presented as a percentage.

Figure 3.7 Accuracy Assessment Report

8
4.0 RESULT/ANSWERED QUESTION

4.1 Result of Accuracy Assessment Report

4.1.1 Unsupervised Classification of Image sub040505

9
Analysis:

The accuracy assessment report's final product is displayed in the result above. The outcome is based
on the image sub040505's unsupervised classification (isodata). The accuracy assessment is referred
the accuracy of five different layers: bare soil, vegetation, mangrove forests, and water bodies. The
classification accuracy across the board is 76.67%. It fell short of the required standard for
classification accuracy, which is 85% or higher. The final output is low as a result of inaccurate digitizing
and labeling performed during the process of unsupervised image.

10
4.1.2 Unsupervised Classification of Image sub220208

11
Analysis:

The accuracy assessment report's final product is displayed in the result above. The outcome is based
on image sub0220208's unsupervised classification (isodata). The accuracy assessment is based on the
accuracy of five different layers: bare soil, vegetation, mangrove forests, urban and water bodies. The
classification accuracy across the board is 86.67%. It satisfies the required standard of classification
accuracy, which is 85% or higher. The reason why the final output accuracy is high during unsupervised
image processing is due to precise accurate digitizing and labelling.

12
4.1.3 Supervised Classification of Image sub040505

13
Analysis:

The output of the accuracy assessment report is displayed in the result above. Based on the supervised
classification (maximum likelihood) of image sub0040505, the outcome was obtained. The accuracy
assessment is based on the accuracy of five different layers: bare soil, vegetation, mangrove forests,
urban, and water bodies. The overall accuracy of classification is 83.33%. It fell short of the required
standard for classification accuracy, which is 85% or higher. The reason why the output is low in the
final output is because of inaccurate digitizing and labelling during supervised image processing.

14
4.1.4 Supervised Classification of Image sub220208

15
Analysis:

The accuracy assessment report's final product is displayed in the result above. The outcome is based
on the maximum likelihood supervised classification of image sub220208. The accuracy assessment is
based on the accuracy of five different layers: bare soil, vegetation, mangrove forests, urban, and
water bodies. The classification accuracy across the board is 93.33%. The output satisfies the required
standard for classification accuracy, which is 85% or higher. The reason the final output is high is
because of the high accuracy of the digitizing and labeling performed under the supervised
classification of the image.

16
4.2 Answered Question

1. From the previous laboratory sessions (supervised & unsupervised image

classification), assess the accuracy assessment for each result. (You should have four
images from two techniques for both years).

Using unsupervised isodata and supervised maximum likelihood (for images sub040505 and
sub220208), the results are presented in chapter 4.1 (Accuracy Assessment Report).

2. Which technique produces higher accuracy classified images? Specify the reasons.

The Maximum Likelihood supervised classification approach outperforms the Isodata

unsupervised classification method in terms of picture accuracy. There are various causes for
this, including:

Maximum Likelihood takes into account the statistical distribution of pixel values within
each class and determines the probability of a pixel belonging to a specific class based on its
spectral signature. If classes in a picture exhibit distinct spectral reparability, the Maximum
Likelihood technique can more successfully discern between classes, resulting in higher
accuracy. The Isodata unsupervised classification approach, on the other hand, assigns classes
entirely based on spectral similarity and clustering methods, which may not always catch tiny
spectral differences between classes.

Class ambiguity and error correction: Maximum Likelihood and other supervised
classification approaches provide a framework for error correction. The classification
accuracy can be assessed, misclassifications found, and the classification refined by altering
the training samples or utilizing post-classification modification. This repetitive method
enhances the final image's categorization accuracy. Because it allocates classes based on
statistical clustering without specific reference to ground truth data, the Isodata unsupervised
classification method is incapable of doing direct error correction as an unsupervised
approach.

17
 3. In your opinion are there any differences if you were only using 100 samples with
stratified random?

Using only 100 samples for accuracy testing in Erdas Imagine with stratified random
sampling may yield different findings than using a larger sample size. With a smaller sample
size, all classes may not be effectively represented in the accuracy evaluation. The goal of
stratified random sampling is to ensure that different classes are represented proportionally
within the sample. It may be challenging to adequately capture the entire range of variability
within each class with only 100 samples.

This can lead to biased or less trustworthy accuracy estimations for specific classes,
particularly those with a lower frequency of occurrence or more suitable spectral features.
Following that, Application to the entire image It is likely that the accuracy evaluation findings
from a smaller sample do not precisely reflect the classification performance throughout the
entire image.

A greater sample size enhances the possibility of collecting spatial and spectral
heterogeneity in the image, allowing for a more robust evaluation of classification accuracy.
With only 100 samples, there is a risk of simplifying or disregarding spatial heterogeneity,
which could lead to an incomplete knowledge of categorization performance.

Using a sample size of 100 with stratified random sampling for accuracy assessment
in Erdas Imagine may bring various restrictions and uncertainties in estimating accuracy's
representation, precision, generalizability, and confidence. Larger sample sizes are
recommended wherever possible to provide more reliable and robust accurate evaluation
outcomes.

18
5.0 CONCLUSION

In the end, accuracy evaluation in Erdas Imagine is an important stage in remote

sensing analysis for assessing the quality and dependability of classification results. Users can
measure the accuracy of their categorization and base judgements on it by comparing the
categorized image to ground truth data.

Erdas Imagine 2014 includes tools and capabilities for completing accuracy
assessments, such as the ability to build workflows, specify accuracy assessment parameters,
and generate reports with accuracy metrics and statistical measures.

These metrics, which include overall accuracy, the kappa coefficient, user accuracy,
and producer accuracy, provide information about the accuracy of classification results for
different classes as well as the overall effectiveness of the classification algorithm.

The accuracy assessment technique in Erdas Imagine reveals any discrepancies or

inaccuracies in the classification results, allowing users to fine-tune the classification
algorithm, training data, or spectral signatures as needed.

This iterative procedure enhances the classification's accuracy and dependability,

making it more useful to land cover mapping, environmental monitoring, and urban planning
applications. Users may confidently analyze and understand remote sensing data by including
accuracy evaluation into Erdas Imagine 2014 methods, ensuring the correctness and integrity
of their results and supporting evidence-based decision-making.

19
6.0 REFERENCES

Hexagon Geospatial. (2014). ERDAS IMAGINE 2014.

https://www.hexagongeospatial.com/products/erdas-imagine/erdas-imagine-2014

Gonzalez, R. C., & Woods, R. E. (2018). Digital image processing. Pearson.

Richards, J. A., & Jia, X. (2006). Remote sensing digital image analysis: an introduction.
Springer.

Jordan, C. F. (1969). Derivation of leaf-area index from quality of light on the forest floor.
Ecology, 50(4), 663-666.

20