Neural network for radar terrain image recognition

This paper will describe a method for recognizing radar image features using a neural network computation architecture. SEASAT SAR source data is used to demonstrate the concept. The evaluation of pre-processing techniques such as line detection, extraction of textural information, and region edge detection will be discussed. The procedure used for defining a training set and the multilayer perceptron architecture using the back-propagation training algorithm will also be described. Performance results of the classifier illustrating the effect of varying node densities in the architecture will be presented.