Multispectral image analysis using artificial neural network system on a CRAY X-MP

The authors have developed software simulation for ANN systems with two learning paradigms back-propagation and competitive learning on a CRAY X-MP computer system. They use the back-propagation network with three layers: the input layer, hidden layer and the output layer. They use seven, sixteen and four neurodes in input, hidden and output layers respectively. The seven units in the input layer represent seven spectral band values for a given pixel. The four neurodes in the output layer represent four categories of objects. In the training phase they use four training areas of size 10X10 each. The model was trained with the training set data. The authors had to iterate the training set data 900 times to ensure that weights were stabilized. The learning process consumed 35.6 seconds on the CRAY X-MP machine. A similar ANN model was also developed on an IBM PS/2 system. The learning process with the same training set data consumed 88155 seconds (244875 hours) of processing time