Lab exercise 8, due right before the next class
GEO5083: Remote Sensing Image Processing and Analysis, UTSA
Student name: ______________
Image classification (per-pixel based)
This lab is to practice four types of pixel-based image classification techniques: ISODATA, Parallepiped, Maximum likelihood, and Spectral angle mapper.
Step 1. Preparation
The only data you will use is your ETM+ reflectivity image (6 bands, UTM and WGS84) or your MODIS reflectivity data (7 bands) you got in previous labs. You can do both and make some comparisons see how they relatively match or not (this will be extra 20 points).
Step 2. ISODATA
I hope you can find where the ISODATA is in ENVI. Open it and set the parameters (you should be very careful for those numbers: Maximum Class Stdev, Minimum Class Distance, Maximum Distance Error, since your input images are in reflectance from 0-1. to get a sense of what numbers you might put, you may need to run a statistic of the original image to see what you should put). If you do not know what these parameters mean, please click the Help to find out. Click OK to run it. change the parameters to see the results. Open and link your classified images with the original image to see if there is any difference. and see if you can label or link your clusters into real world classes? also compare the statistics to see the difference of the classified images. Realizing the Maximum Stdev From Mean and Maximum Distance Error are empty in the above figure. If you input numbers there, run the classification again to see what happen? explain it.
Step 3. Select training site via ROI
Find the Region of Interest tool from your Display window, select 3 or 4 ROIs that you would like to be classified. these ROIs will be used for all supervised classifications you will perform in the lab.
Step 4. Parallepiped
Open the Parallepiped tool, select the ROIs you made in step 3. In the Set stdev from Mean, you can either select single value (all bands use the same value) or multiple values (each band uses different value). in the figure below, I use single value 3 (s to mean), you can click the Preview to see the classification result, use the Change View to select the sub area you had training sites selected. You can change the value to 1 or 2 (s to mean). click the Preview to see the difference. report the difference regarding to the classified areas (pixels) as the value increasing from 1 to 3.
Step 5. Maximum Likelihood
The operation of the maximum likelihood method is very similar to the Step 4. the difference is you set up a probability threshold (0-1), 0 is totally not like, 1 is 100% like. you should select different value to test which one did the best for your classification purposes. Remember for this method, you need to check if the histogram of your ROIs is normal distribution or not. if not, you can not use this method.
Step 6. Spectral Angle Mapper
This is the most useful method was developed for hyperspectral image, but good for multiple-spectral image as well. You will import the ROIs from Import -> from ROI/EVF from input file (see left of the figure below). Select All and Apply, you will see a window (right of the figure below). As the same as step 4 and 5, you can select either single threshold value for all or select multiple values. the values are from 0 (100% the same) to 1 (totally different). Using different values to check the results.
Write a report about the procedure, answer some questions brought out in the lab, and do necessary analysis to your results.