FEATURE-EXTRACTION AND PATTERN-CLASSIFICATION OF REMOTE-SENSING DATA BY A MODULAR NEURAL SYSTEM

A modular neural network architecture has been used for the classifica tion of remote sensed data in two experiments carried out to study two different but rather usual situations in real remote sensing applicat ions. Such situations concern the availability of high-dimensional dat a in the first setting and an imperfect data set with a limited number of features in the second. The learning task of the supervised multil ayer perceptron classifier has been made more efficient by preprocessi ng the input with unsupervised neural modules for feature discovery. T he linear propagation network is introduced in the first experiment to evaluate the effectiveness of the neural data compression stage befor e classification, whereas in the second experiment data clustering bef ore labeling is evaluated by the Kohonen self-organizing feature map n etwork. The results of the two experiments confirm that modular learni ng performs better than nonmodular learning with respect to both learn ing quality and speed. (C) 1996 Society of Photo-Optical Instrumentati on Engineers.