A new method of radar target recognition based on neural network

This paper provides, for the first time, a new method for recognizing the number of aircraft in radar targets based on a neural network. The method converts many duplicate cycle intermediate frequency (IF) narrow bandwidth signals received by the radar which scans the targets into a complex envelope signal, calculates its autocorrelation matrix and autocorrelation matrix characteristic values, and uses these as the primary characteristic of radar target recognition. This is done in order to decrease the coherency between the characteristic components, highlight the divergence, reduce the characteristic space dimensions, compress its primary characteristic data through the APEX (adaptive principal component extraction) algorithm, and uses the BP (backpropagation) algorithm to recognize the compressed data as the quadratic characteristic. Recognition trial shows the aircraft number information extracted from the IF narrow bandwidth signal using this method, is influenced little by the targets space position or posture and it has stronger curb noise ability. Applying the APEX algorithm for compression of the primary characteristic data makes the targets characteristic stable and efficient, and results in a relatively good target recognition.