Lab exercise 7, due right before the class on October 12, 2005
EES5053: Remote Sensing. Earth and Environmental Science, UTSA
http://www.utsa.edu/LRSG/

Student Name: ___________________

 

Hyperspectral analysis and image processing

 Purpose

      In this lab, you will learn a few procedures of hyperspectral image analysis and processing. You will learn new display tools in ENVI, you will convert raw digital number to radiance using bandmath, and then you'll run an image classification algorithm to locate a specific feature in the hyperspectral image.

1. Preparation 

Create a subdirectory Lab7 under your EES5053, and make a directory MyWork under the Lab7 to save your work. The Hyperion image you will use is EO1H_p27r39_subset. Download this image from here to your directory.

2. Introduction

In this section you will get familiar with a couple of useful ENVI display tools. Answer the questions as they present themselves.

1)      load and view image file EO1H_p27r39_subset

2)      display band 50 using gray scale

3)      use pixel locator (Tools: pixel locator) to find pixel at 134, 2070

4)      use z-profile (Tools: Profiles: Z-profile)  to examine raw DN spectra at the point above

5)      How is this spectra different from the one on page 43 of your text (compare to the “solar radiation at sea level” plot)? Describe the differences?

Hyperion imagery is composed of 242 channels or wavebands.  There are actually two different sensors, a VNIR sensor (visible and near infrared), composed of bands 1 - 70, and a SWIR sensor ( short wave infrared), composed of bands 71 - 242. Each sensor has a different gain applied to it. Also, there is considerable overlap in the sensor channels in the region from 851 to 1057 nm.  The image referenced above has already been subsetted to remove the overlap channels and some noise channels. This image has been reduced to 221 channels.  So in order to convert the raw DN to radiance, you will have to use bandmath to remove the gain or scaling factor.

3. Converting raw DN to radiance

Steps to follow

(For the VNIR Bands 1-57)

1) From main toolbar go to  Basic Tools : Band Math

2) From the Band Math window, in Enter an expression, type b1/40.0

3) Click OK

4) From Variables to Band Pairings window, Click on “Map Variable to Input File”

5) Select the "EO1H_p27r39_subset" file, select Spectral Subset, highlight bands 1 to 57 (clear all first, then use Add Range 1 to 57)

6) Click OK, Click OK

7) Choose output file name, such as:    EO1Hbands1-57

           This is the Hyperion VNIR converted to absolute radiance.

(For the SWIR Bands 58-221)

1) From main toolbar go to Basic Tools : Band Math

2) From Band Math Window, in Enter an expression, type b2/80.0

3) Click OK

4) From Variables to Band Pairings window, Click on “Map Variable to Input File”

5) Select the "EO1H_p27r39_subset" file, select Spectral Subset, highlight bands 58 to 221.

6) Click OK, Click OK

7) Choose the output file name such as:    EO1H_bands 58-221

            This is the Hyperion SWIR converted to absolute radiance.

Now combine the two data files

1) From main toolbar go to Save file as : envi standard

2) In New File Builder select import the VNIR radiance file, then the SWIR radiance file. Reorder option will allow the reordering of

     the files by dragging a name above another (if necessary).

3) Give this file the name:    EO1H_bands1-221_radiance

4) Click OK

4. Preliminary image analysis and interpretation

1)     Return to the same band 50 in the new converted radiance image (the last file you saved above) and examine the pixel at 134, 2070.

2)      Now compare the same pixel using z-profile with the solar radiance spectra on p. 43. How do they compare now?  Where are they similar and where do they differ?

3)    In the z-profile plot window go to File: Save plot as: spectral library

a.       select the plot in the “Output plots to spectral library” window. Click OK

b.      in the Output spectral library window, leave all fields as is and select “Output result to memory”. Click OK

4)      Find pixel 116, 2043

5)      Examine its spectra using z-profile (save this plot as a spectral library using the same steps listed above)

6)      Go to File: Input data from: Spectral library, find the file you saved to memory 1 (the one associated with pixel 134,2070), and load it.  Examine both spectra of the two pixels in the plot window.

7)      Why are the two spectra so radically different?  What might account for the large differences in radiance across all the bands? Use your image interpretation skills to describe what the materials on the surface might be? To facilitate interpretation load a new RGB display using bands 45, 32, and 20 as RGB.  Does this false-color composite image help?

5. Image classification using Spectral Angle Mapper algorithm

1)      Leave the RGB display of the image area (from step 7 above) on the screen.

2)      From the ENVI menubar, go to Spectral: Mapping methods: Spectral Angle Mapper

3)      Select “EO1H_bands1-221_radiance” as the input file. Click OK

4)      From the Endmember Collection dialog, go to Import: from spectral library file

5)      Select the library file that should have been saved previously in Memory2 (the one from pixel 116, 2043).

6)      In the next dialog, you should see X:116 Y:2043 in the Available spectra window. Click on the Select All Items button.  Click OK.

7)      In the Endmember Collection dialog, click on Apply

8)      The Spectral Angle Mapper Parameters dialog will now appear.

a.       Set Output to Result to Memory

b.      Toggle Output Rule Images? to No

c.       Click on Preview

9)      In the Preview image that appears, notice what was classified with a max. spectral angle of 0.10. Now that you have probably already deduced that the image feature in question is a water body, do you think that the SAM algorithm correctly classified all water bodies in the image? (be careful here because the Preview window will not show the entire extent of the RGB image).  Look around on the RGB display for more water features. Did it miss any?

10)  Adjust the spectral angle to 0.15 and click on Preview again. Repeat up to a 0.20 angle. Now click on OK to run the final SAM classification.

11)  It will appear in the Available Bands window as Memory 3. Load it in a new display.

12)  Now link the two images. In either display, go to Tools: Link: Link Displays. Click OK.

13)  If you left click anywhere on either display, the other image will appear as a background. Cool, huh?

14)  Did all (in your estimation) water features get classified as water?  What did it miss?  Why didn’t the streams associated with the pond in the lower part of the image get classified? What about the dark feature at approx. 239, 2107? What do you suppose that feature is, if it’s not water?