APPLICATION OF MULTIIAYER NEURAL NETWORKS IN THE CLASSIFICATION OF FOREST TYPES USING RADAR IMAGERY

The use of multilayer neural networks for the classification of forest types on the basis of the processing of radar images is investigated. It is demonstrated by modeling that with respect to stochastic characteristics the described classification method (using multilayer neural networks) and a training procedure designed to increase its accuracy (via the a priori specification of radar reflectance values typical of specific types of vegetation) are highly effective. The methods were tested and verified using SIR-C images recorded by the U.S. Space Shuttle.