Neural network adaptive wavelet classification of radar targets

The impulse responses of ultra wideband radar (UWB) targets are being approximated as a weighted linear combination of dilated and translated replica of a mother wavelet. The weights, dilation, and translation parameters are being adaptively estimated to minimize an approximation error in a least mean square (LMS) sense. Morlet wavelet is used in this study, but the algorithms applied can be modified to include other types of wavelets. The adaptively estimated target representation parameters are then being used as features for target classification using a neural network. No assumptions regarding noise statistics, apriori probability of radar targets, or stationarity are required. Radar cross section (RCS) measurements of five commercial aircraft are used in the experimental phase of the study. The performance of the wavelet based neural network classifier is compared with that of parametric recognition techniques under different noise scenarios and assuming target azimuth uncertainty