Classification of Geometrical Targets Using Natural Resonances and Principal Components Analysis

A classification scheme for radar target recognition based on natural-resonances and principal-component analysis named as NRPCA is described and tested. It uses a multilayer perceptron (MLP) neural network as classifier and complex natural resonances (CNRs) as inputs to the net. The benefit of using natural resonances is that they are aspect-angle independent. Nevertheless, the extraction process of natural resonances is an ill-conditioned problem and noise affects the estimation of the parameters. To reduce this noise sensitivity the NRPCA algorithm extracts the resonances from a reconstructed response generated by a principal-component analysis (PCA) stage over a set of reference targets prior to the MLP neural network classifier. Furthermore, NRPCA uses the step time response instead of impulse time response to minimize any error in CNR extraction process (matrix pencil method). To test the NRPCA algorithm several geometrical conducting bodies are used as reference targets of a radar system.