A neural clustering approach for high resolution radar target classification

Abstract Neural learning techniques, the Self-Organizing Feature Map and Learning Vector Quantization, have been applied to the automatic target recognition (ATR) problem in the presence of high range resolution radar target signatures. The database is collected by placing the targets on a rotary turntable and slowly turning them over a complete 360° azimuth while the radar signatures are collected. Our pattern recognition system is composed of a feature identifier and a classifier. A simple Euclidean distance classifier using those identified features provides a baseline of 97% mean probability of correct classification.