Automatic Target Discrimination Using Neural Networks

In automatic target recognition systems, there are two key requirements. First, the processing of scene information must be in real-time. However, in the case of IR or missile sensor systems, the systems generate several million samples per second which is very difficult to process. Second, the results must be reliable and accurate. Correct identification of targets of aircraft or missiles is not an easy task since the shapes are similar from a distance 10 or 20 miles away. We propose a practical and efficient paradigm to achieve reliable and real-time digital scene processing. It consists of three major components: reprocessing (including template matching), real-time dynamic PCA, and Fuzzy CMAC. IAI has developed a real-time dynamic PCA (Principal Component Analysis) procedure for data compression, which will significantly increase the processing speed of the overall target discrimination system. We have already applied this novel method to traffic incident detection. Since real-time recognition of targets is most critical for military applications, Fuzzy CMAC is an ideal candidate for this purpose for the following reasons. First, the Fuzzy CMAC learning structure is a novel innovation and a creative use of fuzzy logic and CMAC based neural networks. It greatly speeds up the learning process. Second, the learning process is well suited for real-time application since the training process is based on real-time input and output data. Third, the Fuzzy CMAC is relatively straightforward to implement in high speed, highly parallel hardware. Fourth, expert knowledge can be incorporated into the system very easily. The proposed architecture can be applied to classification of signals, pattern recognition, and voice recognition. The real-time dynamic PCA can be used for data compression for transmitting information over a bandlimited channel. The Fuzzy CMAC can be used for feedforward control of chatter in machine tools, fault diagnosis in aircrafts and machine tools.