We also applied this method to the same image of Palni hills (India). We combined

• certainty factors obtained by a neural network classification, using spectral data solely, with
• certainty factors obtained by neural networks representing all units of knowledge, as described in the preceding production rule system.

The combination uses the Dempster's orthogonal rule, and NN classification has been applied to image radiometries .

The performance of the result 54.13% (see line A in table below), is quite similar to maximum likelihood classification.

The resulting image (image 42) shows the representation of maximum certainty factors obtained by the neural networks which model expert knowledge units.

The result obtained by combining the two sources, according to the Dempster's orthogonal sum (see resulting image 43) gives an average performance which has reached 74.47%, ( see line B in table below). The result is 20.34% better than the initial NN classification and 15.87% better then the maximum likelihood classification.

We can note that the result is 2.59% better than ICARE classification.