GEOM30009 IMAGING THE

ENVIRONMENT

Assignment 3

Information Extraction from Images

Due for submission at 11:55 pm on Friday of Week 9

Value: 15% of Subject Mark

Objective

The purpose of this exercise is to learn preliminary image processing and information extraction

from multi-spectral images. To do this assignment, a Landsat image set over Melbourne and the

ENVI software will be used. The assignment involves working with images captured at different

wavelength bands and combining images to analyse vegetation.

Background

Radiometric enhancement is often a preliminary step in image interpretation and information

extraction from aerial and satellite images. An example of radiometric enhancement techniques

is histogram stretching. This technique is used to increase the image contrast and improve the

visual quality of the image. Another useful technique is creating band ratios for combining

multispectral images to highlight various features. Briefly, this technique involves an arithmetic

operation on multiple bands resulting in a new image. Band ratios are used to highlight spectral

signatures of different objects. For example, healthy vegetation has low reflectance in the red

wavelengths of the electromagnetic spectrum and high reflectance in the near-infrared

wavelengths. Therefore, by dividing a near infrared band by a red band we can create a new image

in which healthy vegetation is highlighted by large values whereas everything else has low values.

Vegetation indices like Normalized Difference Vegetation Index (NDVI) can further help

distinguish healthy vegetation from stressed vegetation.

Data

A Landsat-8 dataset of Melbourne acquired on 11 February 2024 will be used for this assignment.

Information about the resolution and wavelength bands of Landsat-8 images can be found on the

Landsat-8 website:

https://landsat.gsfc.nasa.gov/satellites/landsat-8/landsat-8-bands/

Software

ENVI will be used for reading the dataset and processing the images. Information about the different

processes can be obtained from the software Help documentation (Menu bar > Help > Contents).

Tasks

The assignment consists of three main tasks:

1. Applying radiometric enhancement

2. Combining the images to create band ratio images including NDVI

3. Performing image classification.

You should be able to complete each task in one lab session. The whole assignment should be

completed within three weeks.

Task 1: Applying radiometric enhancement techniques

In this task, you will be able to visualize different bands of the image and use the histogram to adjust

the illumination, contrast and brightness.

Steps:

1. Unzip the dataset file into a folder in your local disc.

2. Start ENVI and open the dataset (open the …_MTL.txt).

3. Right click on the image layer in the Layer Manager panel (left side of the software) and

choose Zoom to Layer Extent. Now, you can see the whole image.

4. Like the previous assignment, go to Data Manager (under “File” menu) and expand the

“Band Selection” menu … to select different bands for different colour channels. Try different

band composition such as true colour (Default), the false colour (B3 to Blue, B4 to Green and

B5 to Red), and a full IR false colour image (B5 to Blue, B6 to Green and B7 to Red). To ensure

you locate the correct band, it’s important to familiarise yourself with the names and

numbers of the Landsat 8 bands beforehand.

5. Stretch the image histogram for each band using the Histogram Stretch button

which is

in the main tab. By dragging vertical lines in Histogram Stretch window, histogram for each

band is manipulated and you can see the change on the image. Also, you can choose various

stretching method from the drop-down Stretch Type button. A good stretching can be done

by Linear type.

6. Explore the image using the zoom and pan tool in image display.

7. Head back to Data Manager – this time try selecting its icon

in the toolboxes above instead

of going through the “File” menu. For this step you can select various individual bands from

the dataset. For instance, select a Near Infrared band and press Load Data. Remember to

stretch the image once you have displayed it.

8. The image is now displayed as a grey tone image (gray colour). You can change the colour by

right clicking on the band and choosing Change Colour Table.

Now we will create simple ratio images and combine them to make useful images like NDVI.

Task 2: Creation of ratio images and combining them together

We will create the following ratio images:

 B2/B5 to highlight water.

 B5/B4 to highlight vegetation.

 B7/B2 to highlight soil/clay.

Steps:

1. Start ENVI and open the Landsat-8 dataset.

2. On the Toolbox pane, find Band Algebra Key and double click on Band Ratios.

3. In the prompted window, select the bands of a particular ratio for the numerator as well as

the denominator. Next, press Enter Pair and then OK.

4. In the following window, enter output file name as well as the saving directory. If you are

creating the band ratio image highlighting water, name the image Water, for instance.

5. Having done that, the grey scale image will be shown on the display window.

6. Repeat the process for Vegetation and Soil band ratio. Remember to take a snapshot of each

band ratio image as you need them for your report.

Now you have three ratio images each highlighting a certain feature. The next step is to combine

these in a colour composite visualization.

7. Click on Data Manager Key, from drop-down Band Selection button, select the red, green

and blue layers as band ratio image for soil, vegetation and water, respectively.

8. By pressing Load Data, you will see the band composite on the display window.

9. Save this image and take a snapshot of it.

Now, you will create an NDVI image.

11. From the Toolbox pane, navigate Spectral > Vegetation, then double click on NVDI.

12. Select the multispectral image set then press OK.

13. Choose the storing directory and name the image NDVI and then press OK.

14. You will see the NDVI image on the display window in grey scale. You can try various

colour table from Change Colour Table.

15. Take a screenshot of your image and save your image. Don’t forget to add a Colour Bar

(Toolbar > Annotations > Colour Bar).

Task 3: Image classification

Classification is a process in which all the pixels in an image are assigned labels as belonging to

certain categories. It is typically used to process satellite imagery with multi or hyper spectral bands.

Basically, classification is either supervised or unsupervised. In this task, we perform a supervised

classification (Maximum likelihood). To classify the image, follow these steps:

Steps:

1. Right-click on the multi spectral image in the Layer Manager and select New Region of

Interest since supervised classifications need training samples and they can be provided in

ENVI as ROIs.

2. Change the ROI name to water. In Geometry, different options can be used to create

selections of water in the image. Select one (for example Rectangle) and draw rectangles

over the pixels visually interpreted as water (the more training samples you provide the

better is expected the classification to work) and press Enter. Once you are satisfied with the

number of selections done, click the New ROI icon. Let’s create ROIs for the most obvious

features in the image: background, urban, water, bare land, dry forest, forest and low green

vegetation.

3. The previous step was about selecting the training pixels. in order to evaluate the

classification, we need to select test pixels as well. For doing that, repeat previous step for all

classes and select test pixels in a new ROI. Please note that these ROIs will not be used in

training the classification method, so use different names, such as Water_test.

4. From the Toolbox pane. Go to Classification > Supervised Classification.

5. Select Maximum likelihood classification. Maximum likelihood classification assumes that the

statistics for each class in each band are normally distributed and calculates the probability

that a given pixel belongs to a specific class. Each pixel is assigned to the class that has the

highest probability (that is, the maximum likelihood).

6. Input raster: Select the multi-spectral image and click on Spectral subset. In the new dialog,

unselect the Coastal aerosol band and then press OK. Press OK again in the Classification

input file dialog box.

7. Input ROIs: Think about the training classes you made, and select the relevant training classes

(ROIs) in ROI selection. Make sure to select only training classes, not test classes. Leave other

options as default.

8. In Output raster, define the directory and the name of the outcome as Classified then press OK

and the classified image will be displayed in the main pane.

9. If not satisfied with the classification, try to increase the number of training samples in every

ROI or the number of classes (less or more classes) and proceed with all the consequent steps.

10. On the Layer Manager pane, open the folder Classes and change the colour of each class and

try to make the image similar to the true colour image. An example colour code: black- >background, light green -> low green vegetation, green -> forest, dark green -> dry forest,

orange -> bare land, blue -> water and white -> settlements.

11. Evaluate the classification results: From the Toolbox pane. Go to Classification > Post

Classification > Confusion Matrix Using Ground Truth ROIs. On the opened Confusion

Matrix Using Ground Truth ROIs pane select the classified image and then OK. The test ROIs

are available on the Ground Truth ROI section. On the “matched classes” section the

corresponding classes are shown by default, otherwise match them by selecting

corresponding classes from “Ground Truth ROI” and “Classification Image” sections and click

“add combination”. Then click Ok to see the Confusion Matrix Parameters pane. On this pane,

leave all options as default (checked Pixels and Percent), then select Ok. Now you will see the

confusion matrix providing accuracy values for the classification. Report the results on your

report and discuss about the accuracy results according to the shown parameters on the

confusion matrix.

Submission

Write a 1200±20% word scientific report and include the following content:

1. Provide a proper introduction. Address the purpose of radiometric enhancement, band

ratios and classification, and state the aim of this assignment.

2. In the Methods section describe briefly the process you performed to complete each of

the three tasks.

3. In the result section provide an analysis of your results.

4. In the Discussion section address the following questions:

i. What does each ratio image display? Can you relate the appearance of each particular

feature to the spectral reflectance of that feature and the bands used in the ratio?

ii. How do different features (e.g. water, soil, vegetation) appear in the ratio composite

image? Why do they appear differently?

iii. What do the values in an NDVI image represent? How does vegetation appear in the

NDVI image? Why?

iv. What is the role of histogram manipulation in visualizing ratio images and NDVI? Explain

what histogram manipulation does to your visualizations (mention the input and output

values).

v. Analyse the classified image and describe the classes. Highlight classification errors and

discuss why these occur.

5. Provide a clear and concise conclusion summarizing your findings.

6. Provide a list of references if you use external sources in your report.

7. Statement of contribution – include a statement describing the contribution of each group

member to the assignment tasks and the report.

Submit a digital version of your report via LMS and in pdf format only.

Marking rubric

Appropriate length and proper formatting 5%

Proper introduction 5%

Proper Method 5%

Three simple ratio images present and correct 15%

Ratio composite present and correct 5%

NDVI image present and correct 15%

Classification image present and correct 15%

Questions answered and properly discussed 25%

Logical conclusions 10%

51作业君版权所有