Lab 6: Advanced Classification and Accuracy Assessment (Due by 4pm Mar 16)

Introduction

In this lab you will learn how to conduct supervised classification of satellite imagery. You will learn several different methods of creating and evaluating signatures as well as how to measure the accuracy of your classification.  

Instructions

Read pages 221-263 of the Erdas Field Guide. Open the Tour Guide and skim Chapter 17. For this lab, you will use supervised classification to extract thematic land-cover information from the same TM image (germtm.img) you used in Lab 5. The image file can be found in I:\Students\Instructors\Geoffrey_Duh\GEOG4581\Lab6. Copy the file to your working folder before you begin. Complete the supervised classification part of the tutorial exercise (pp 443-470) and produce a map composition of the classified map and its legend. Put your name on the map with a text box. Then do an accuracy assessment (pp 483-487) and print the results. Please generate 25 random points, instead of 10, for this exercise. Answer the following questions.

1. Explain each of the different ways you defined signatures.
2. What is a feature space image?
3. What do you look for on the histograms of signatures when evaluating those signatures?
4. What is separability?
5. What do the univariate signature editor statistics tell you?
6. Explain the decision rules (classifiers) that you used to produce the supervised classification.
7. Explain the accuracy assessment report.  Make sure you explain what the different numbers of the error matrix and overall accuracy values mean.