A comparison of radar signal classifiers

The performances of two neural network classifiers are compared with those of two conventional information theoretic pattern classifiers for the classification of radar returns. Each of the classifiers is applied to the problem of discriminating between the radar signal returns of five commercial aircraft at various azimuth angles. The neural network classifiers are frequency-sensitive competitive learning (FSCL) and FSCL learning vector quantization (FSCL-LVQ), a variation on T. Kohonen´s LVQ classifier (1988). These are compared to the nearest-neighbor and the maximum-likelihood classifiers. It is shown that the performance of the neural classifier is close to that of the maximum-likelihood and the nearest-neighbor classifiers. The results indicate that the neural classifiers are relatively insensitive to the noise level of the training data