GEOG474/666
 Classification

Exercise 8

    Image classification is one of the most often used methods of information extraction.  A variety of algorithms exists including (1) hard classification using supervised and unsupervised approaches, (2) classification using fuzzy logic, and (3) hybrid approaches often involving the use of ancillary information.  For this exercise, we'll examine the  multispectral classification techniques available in ArcView's Image Analysis.   Remember that Image Analysis extension is only available on the PCs in Pearson Hall Training Center.

Imagery for this exercise is found under ~tracyd/Geog474/images directory and the file is called:

spot_sdel_sub.img



Start ArcView.

Examine the Online Help to see where the image classification techniques are found.

Under the Help menu.  Select Help topics and click on the Contents tab.  Under the Image Analysis extension, select "Extracting features and categorizing images" and you'll see:
"How to outline a feature"
"Changing Seed tool properties"
:"Saving a feature in a shapefile"
"Finding like areas in an image"
"Automatic land cover classification"
"Converting results to a shapefile"
"Smoothing the results"
"Advanced categorization"

Questions:

  • The "Automatic land cover classification" process categorizes multiband imagery using unsupervised or supervised approach?
  • This approach is based on using the ISODATA technique.  Describe this technique.
Perform an unsupervised classification on the Southern Delaware Spot image defining 7 classes.  Once the categorization theme is created, define each of the 7 classes into its appropriate land cover type.  Which classes do you feel fairly confident about the assigned categorization?  Which classes are you less confident?  Why?
Next, let's examine ArcView's "Advanced categorization".
QUESTIONS:
  • What type of classification algorithm is this type of categorization?
  • The seed tool uses what type of classifier?
Steps to perform "Advanced categorization":
(1) Control the Seed tool
You can access the Seed Tool Properties dialog under Image Analysis menu and change the seed radius default of 5 pixels and whether island polygons are included, default enables inclusion of island polygons.
(2) Create a shapefile
Create a feature theme in which to save a polygon created by the Seed tool by (a) add the image, (b) create a new feature theme (polygon feature type), (c) change seed tool properties (seed radius, include island polygons), (d) use see tool to populate the new shapefile, (e) stop editing theme
(3) Use Seed tool to identify land cover type.

(4) Use the shapefile to find similar areas in the image.

Find Like Areas is designed to find areas in the image with similar characteristics to an area you originally select with the Seed tool.  Under the Image Analysis menu select "Find Like Areas" option.
(5) Set Analysis Mask to limit Categorize to just a certain portion of the image.
Under the Image Analysis menu, choose Properties.  Set Analysis Mask under Analysis Extent.
(6) Apply post categorization techniques
Apply smooth filter to thematic image.

Convert theme to a shapefile.

Perform image categorization on the Southern Delaware Spot scene by defining a few land cover types (e.g., forest, water, agriculture).

QUESTIONS:  Discuss the results of the image categorization?  How well did the cover types classify the image.
 

That's it.  Please turn in your brief answers to the questions listed above.
Return to Geog474 Homepage
Last revised on December 7, 1999 by Tracy DeLiberty.