A SAR target classifier using radon transforms and hidden Markov models

摘要

This paper describes an alternative to template matching for SAR automatic target recognition. The approach uses a Radon transform to map each target chip to a vector series. It then models each series as a sample of a hidden Markov process for which identification can be performed by using a modified Viterbi recognition engine. By using appropriate preprocessing and feature selection, performance for aerial SAR can be made invariant to the orientation of the target using only a single model for each target class. Performance is comparable to, or better than, that obtainable using template matching with only a fraction of the required storage. (C) 2002 Elsevier Science (USA). [References: 8]