Parent Topic: Report
Supervised Report
The first list you see is a summary of the classification, listing all the classes in your training channel. The ``Pixels'' column lists the number of pixels that were classified into each class. The ``%Image'' lists the percentage of the image that that class covers.

This list is followed by the ``CONFUSION MATRIX''. The confusion matrix gives information on how much of each original training area was actually classified as being in the class that the training was meant to represent. If many pixels in the training areas were classified in different classes than intended, it is likely that the training areas were not appropriate. 

       ______Areas_________  _____Percent Pixels Classified by Code______

       Code Name     Pixels     0   10   20   30   40   50   60   70   80
       --------------------  --------------------------------------------
         10 Water1      470   0.2 96.4  0.0  2.8  0.6  0.0  0.0  0.0  0.0
         20 Water2      145   2.8  0.7 89.7  6.9  0.0  0.0  0.0  0.0  0.0
         30 Urban      3829   1.5  0.0  0.0 92.9  2.7  0.0  0.5  2.4  0.0
         40 Range      1835   0.0  1.7  0.0  7.4 79.1  1.2  0.0  3.8  6.9
         50 Crop1      1536   0.0  0.0  0.0  5.7  4.7 88.4  0.0  0.0  1.2
         60 Crop2      2057   1.7  0.0  0.0 10.2  0.7  0.0 87.5  0.0  0.0
         70 Crop3       350   0.0  0.0  0.0  2.0  2.6  0.0  0.0 95.4  0.0
         80 Forest     1973   0.0  0.5  0.0  1.3  1.6  0.4  0.0  0.0 96.2

        Average accuracy =  90.70%
        Overall accuracy =  90.05%

        Kappa Coefficient  =  0.87654    Standard Deviation =  0.00336
         Confidence Level :
         99%  0.87654 +/- 0.00867
         95%  0.87654 +/- 0.00659
         90%  0.87654 +/- 0.00553
In this example, we see that of the 470 pixels in the ``Water1'' training area, 96.4 percent were classified as ``Water1'', while 0.2 percent were not classified at all (0). Looking down the matrix we see that ``Range'' (40) suffered from the worst classification confusion, with only 79.1 percent of the training area classified as ``Range''.

The average accuracy is the average of the accuracies for each class, and the overall accuracy is a similar average with the accuracy of each class weighted by the proportion of test samples for that class in the total training or testing set. Thus, the more accurate estimates of accuracy, (i.e., those from larger test samples), are weighted more heavily in the overall accuracy.

In the above example, average and overall accuracy are calculated as follows: 

   Average accuracy = (96.4 + 89.7 + 92.9 + 79.1 + 88.4 + 87.5
                      + 95.4 + 96.2) / 8

   Overall accuracy = (0.964x470 + 0.897x145 + 0.929x3829 + 
                       0.791x1835 + 0.884x1536 + 0.875x2057 + 
                       0.954x350 + 0.962x1973) /
                       (470 + 145 + 3829 + 1835 + 1536 + 2057 + 
                               350 + 1973)
Also, the Kappa coefficient, standard deviation, and confidence levels are given in the report.

The last section, the "The Totalization Report" is a more detailed expansion of the confusion matrix.

Parent Topic: Report
About PCI Help Gateway