Neural networks for extraction of weak targets in high clutter environments

It is shown that feed-forward and graded response Hopfield neural networks can implement an optimum postdetection target track receiver. For the Hopfield net the spurious states correspond to the important case of multiple track detection. Simulations are presented that show that substantial signal-to-noise gain can be achieved. The advantage of the Hopfield network is that the desired final output cells are just time-evolved input cells, so that no additional cells are required for output display. Although a feed-forward network may relegate display to an external device and compute only cells corresponding to the presence of particular patterns, additional cells are still required, over and above the input cells, for the detection of each pattern. If a large number of patterns is required, then the advantage of the Hopfield network, which needs only the input cells, becomes evident. >