Lab exercise 9, due right before the class on Apr. 1, 2005
ES6973: Remote Sensing Image Processing and Analysis, UTSA
http://www.utsa.edu/eps/programs/EnvSci/HXie.htm

Student name: ______________

Classification accuracy assessment of and post-classification

Purpose

            This lab is to (1) use the error matrix (confusion matrix) to assess classification accuracy and (2) do necessary post-classification and make a GIS map.

Step 1. Preparation

    The data you will use is your ETM+ reflectivity image (6 bands, UTM and WGS84) and classified results.

 Step 2. Accuracy assessment

      In lecture 8, we talked about the selection of training sites and reference sites. it usually requires field works and/or high spatial resolution images (such as aerial photos) and extensive image experiences. the accuracy assessment can be based on training sites (pixels) or reference sites. The accuracy based on training sites is usually higher than the one based on reference site. if assessment is based on reference site, we need also to calculate the referencing sample size and do sampling design (as randomly as possible).

    In this lab as a practice, it is impossible for you to follow all the steps (though you should be aware the steps and resources you need if you do a real classification project). what you will do is using the classified results from maximum likelihood and spectral angle mapper you did in last lab or do it again in this lab, using the same exact training sites (for example 3 to 6 classes) for two methods. then select the reference sites (3 to 6 classes), as random as possible, that you know (in your experiences) they are the same classes as the training sites.

    Click Classification -> Post-Classification -> Confusion Matrix -> Using Ground Truth ROIs. You will be asked to select your classified image (e.g. the maximum likelihood result). you will select the match classes (see an example):

Click OK, then click OK in the Confusion Matrix Parameters window. Then you will get Class Confusion Matrix window, which is your error matrix.

    Do the same steps for spectral angle mapper results.

    Compare the two error matrixes and do analyses and evaluation based on the error matrixes to see which classification method is better.

Step 3. Post-classification and GIS map

    Do necessary post-classification to clean up your classified results (the best one you find from the Step 2), you should include clump, sieve, combine if you need to combine two classes into one class. then covert your classification image to GIS vector format, and export as .shp file.

    Open ArcGIS, add your image and your shape file(s) to overlay the vector to raster image. make a GIS map including necessary legends and other map parameters such as scale, N arrow, name, projection, datum, classification method used, accuracy assessment information, post-classification methods used. If you are not familiar with ArcGIS, you can use ArcView, though ArcGIS will make a better map than ArcView does.

Write a report includes the procedures, analysis results, and the final GIS map.