Lab exercise 9, due right before the next class
GEO5083: Remote Sensing Image Processing and Analysis, UTSA
Student name: ______________
Assessment of image classification accuracy
This lab is to help you understand how ENVI can be used to do accuracy assessment based on confusion matrix or error matrix.
Step 1. Select ground truth sites via ROI
Based on the last lab, you used the same ROIs (training sites) for three supervised classification methods to classify your image. now you will select ground truth sites via ROI tool for each class you classified. You should not selected the same pixels you used as for training sites. The best way to do so is that, during the time when you select your training ROIs, you also select your ground truth ROIs. so for each class, you have pairs of ROIs: training ROIs and ground truth ROIs. the training ROIs will be used for classification while ground truth ROIs will be used for accuracy assessment.
Step 2. Confusion matrix
go to ENVI menu, Classification -> Post classification -> Confusion Matrix -> Using Ground Truth ROIs. A window called Classification Input File is popup for you to select one class map (for example the SAM classification map). Once you select the map, a new window called Match Classes Parameters is popup as the window below:
then you need to select the real match. for example, in the test, my ground truth Region #4 is the same class of Region #1 (from the training class). So the Region #4 and #1 should be a pair or Matched Class. The same reason, ground truth Region #5 and training site Region #2 are a pair, and ground truth Region #6 and training site Region #3 are a pair. Then we do the following match:
Click OK, you will see the following window:
Click OK, You see the error matrix of accuracy assessment:
the overall accuracy is 57%, and Kappa is 0.485. the classification agreement for region 2 and 3 are both 100%, but for region 1 is 0. Omission error for region 1 is 100%, for regions 2 and 3 is 0%.
Using the same ground truth ROIs for the two other classification maps to product error matrix. then compare them to see which method has the best performance among the three methods. write a report.