Geog 430 - Remote Sensing

Lab 5: Supervised Classification

Work through the supervised classification section of the handout (and the accuracy assessment part). Yes, pages are missing. That's intentional.

Grab the kit_8_00.img from the geog430 subdirectory on the J:\ drive. It's a TM image of this area. You will need over 100mb free on your disk for this lab - IF you're careful about file management. Keep an eye on this.

Open the image and check it out.

Now - let's make a few band-band plots. Goto Classifier - feature space image. Set output to viewer. Make sure to put the root output files on your drive, not the C:\ drive (the default). They will be drawn in separate viewer windows - and can be reopened later if necessary. Check them all out. What patterns do you see in the 1-2, 3-4, 3-5, and 4-5 plots? What do the different patterns mean in the real world (linking the cursor - feature space will help).

Next, run a supervised classification. Use the following classes (unless you can - and do - justify different ones):

• Water
• Agriculture - hay
• Agriculture - fallow
• range 1
• range 2
• roads (asphalt)
• roads/buildings (concrete/aluminum)
• forest 1
• forest 2
• forest 3
• exposed basalt
• Light stuff in the hills to the north of E'brg (granite)
• Riparian Vegetation.

Remember to draw multiple training sites for each class - you will later merge them in the signature editor and delete the originals. Be VERY careful when digitizing these training sites. When you have your final training sites, number the VALUE column 1-13 (or however many classes you end up with). Otherwise, you will end up with a LOT of blank classes.

When you have finished this, do a error assessment (Confusion matrix). Start with at least 150 points - finishing with at least 100. Make sure you use the 'stratified random' method of generating points. Again - use the stratified random method.

In addition to handing in the confusion matrix - interpret it for me. What classes were created well, which weren't, where was the overlap, why? Also be sure to tell me what each class represents (ie - that, in your image, say, class 4 is the exposed basalt class). It never hurts to do a screen dump of the map and legend (table) and print it in color. Or dump things into ArcMap and work with it there (the mapping environment is much nicer in Arc - analysis is better in Erdas). Makes my interpretation much easier.

Hand in:

• interpretions of the band-band plots
• supervised classification (call it super.img) - print it out in color (see Ben) - note, the final map is easier to do in ArcMap.
• Confusion matrix (a clean one, not just what you print out of erdas by default - simplest to do a table in Word)
• Confusion matrix interpretation
• A 1 paragraph (or 2) summary regarding what you learned in this lab, liked, disliked, would do differently a second time, etc.

Worth 15 points. Due Monday, 15 May.