12.114 ENVI exercise
September 30, 2011
Daniel Sheehan
Introduction
You will be working with ENVI and Arcgis to process a Landsat 5 image so that you can geologically interpret the image. You will use the several tools of the many that are available in ENVI:
- Principal Components Analysis
- Decorrelation Stretching
- Band Ratios
- Unsupervised classification
- Supervised classification
Opening the software and adding data
Start by opening the enviExercise.mxd Arcmap project in the 12.114ENVI folder on your desktop. You should see the Landsat image and the outlines of the units mapped by the 2011 Field Camp class:
Next, start ENVI, which can be found under the Start menu under ENVI 4.8. It should look like this:
Add data from the File menu. Look for Open External File. You should see IP Software (for Image Processing). Look for Erdas Imagine and click on this. Navigate to the enviExercise folder on your desktop and add image.img. It opens a window like this:
You can see the basic data by using 3 of the bands as the Red, Green, and Blue bands. To do this, click on RGB color. Now select, in this order, the following bands (bands are called Layer_1, etc):
7, 4, 2
Then click on Load RGB. You should see 3 new windows. One is the #1 Scroll window, the second is the #1 Window and the third is the #1 Zoom x4 window. Drag the red box in #1 Scroll over the image and you should see a bigger version in #1 Window and a bigger, but pixilated version in the #7 Window:
Classifying data
Let’s start with an unsupervised classification. Look for this on the ENVI 4.8 window under Classification. You will find two types of unsupervised classification. You should chose one different from your neighbor’s choice. The window that opens should look like this:
Select your image then select Spectral Subset and click on band 6 while you have the Control key depressed. This will eliminate the thermal band from the classification:
Click on OK to complete this part of the process. Then click on OK on the Classification window. You now get a Parameter window. Just enter the name of the output image. Save this as unsuper.img in the 12.114ENVI folder on your desktop:
Add the resulting image to Arcmap. Does classification resemble the mapping?
Now try a supervised classification. To do this, you will need to create regions of interest (ROI) and to that, you will need to add the unit polygons from Arcmap. To add the polygons, look at the ENVI 4.8 window and click on Vector and then Open Vector file. You will need to change the type to shapefile in the lower right of the Open form. You should see this window:
The only things you need to change on this page are the DATUM (change to WGS 84) and the Zone (change to 11). Once you make the changes and click on OK, you will see the Available Vectors list. Click on units.shp and the Load Selected. You will see another small window:
Click on Display #1 and the OK. A new window will appear:
Change the color to brown, click on the Off radio button at the topof the window, and then click Apply. The units should now appear in your windows, assuming that you are zoomed in to that part of the windows.
Now make your region of interest using the unit boundaries. Look at the ENVI 4.8 window and select Basic Tooks then Region of Interest the ROI Tool. You should see this window:
Change the top radio button to Zoom so you can work in the Zoome x4 window. Now starting working on the #1 Map Zoom x4 window. Click in homogenous areas within a unit. Right click twice to close the area. Click New Region on the ROI tool to start a new region. Do this for as many areas as you can clearly see. Your window should look like this:
Now start the classification. On the ENVI 4.8 window, click on Classification, then Supervised Classification, then select one of the choices. You will eventually want to look at the Help pages to understand the different classification methods available. In the Classification Input file, select the original image, image.img. In the parameters window:
Select the ROIs that you created, set the probability threshold to None, select an output file called super.img, and set the Output Rule Images to No. Click on OK to start the classification. Once it is done, add it to the Arcmap window. Did this do any better at classifying?
Decorrelation Stretch
Let’s try the decorrelation stretch next. To do this, click on the ENVI 4.8 window, select Transform then Decorrelation Stretch. You should see this window:
Select Available Bands List then click OK. In the Decorrelation Stretch Input Bands window, select Band 7 for R, Band 4 for Green, and Band 2 for Blue then click OK.
On the Decorrelation Stretch Parameters window, set your output file to uncorrelated.img in the 12.114ENVI folder on your desktop then OK:
Once this completes, add to the Arcmap window. It may not display in a meaningful way so double click on the name in the table of contents or right click on the name and then click Properties. On the symbol tab, change the stretch type to Histogram Equalize and then hit apply:
Does this look like a better classification?
Principal Components
Next try Principal Components. To do this, on the ENVI 4.8 window, click on Transform the Principal Components then Forward PC Rotations then Computer New Statistics and Rotate. You will see this form:
Select the original image, image.img, then click on Spectral Subset and eliminate Band 6. Click on OK to close the subset form. Click ok on the Input File form. On the next form, the Forward PC Parameters form, set the stats filename and the output file name then click on OK to start the processing. Add the result to Arcmap and modify the Symbology properties as you did for the Supervised Classification result (Stretch Type = Histogram Equalize).
Now the hard part
Without instruction, I want you to use the Band Ratio tool on the ENVI 4.8 window under Transform. Do this for bands 7, 4, and 2. Divide each band by band 1 and set the output images to b7b1.img, b4b1.img, and b2b1.img. Then take these as the Red (b7b1.img), Green (b4b2.img), and Blue (b2b1.img) as the inputs to the Decorrelation Stretch routine. Be careful since your available bands list is now quite long.
Add the new decorrelation result to Arcmap. You will have to change the Stretch Type to Histogram Equalize in the Properties Symbology tab as you did before. How does this fit your contact data?
Add each of the band ratios to the map. To make sense of these, you will also need to work on the Symbology. Change Stretched in the Show part of the form to Classified. You will be prompted to tell you that a histogram has to be made. Click on OK. Do this for each of the band ratios. Do any of these fit the unit data?
You can also make a composite of the bands you created with the Band Ratio tool. You do this in Arcmap. Find the tool in the toolbox under Data Management Tool->Raster->Raster Processing. The tool is called Composite Bands. Fill it out like this:
Add this output raster image to ENVI and try the supervised classification using the composite image. You could use your original ROIs to do this. Refer to the instructions earlier in this document.
1