GO-CFAR Trained Neural Network Target Detectors

Abstract

Detecting targets embedded in noise and clutter is an essential task for many radar systems. A competent system must additionally offer high probability of detection with a low false alarm rate and a standard practice is to employ constant false alarm rate (CFAR) detectors. In this article, we develop and expand the use of neural networks to accomplish this objective. The neural networks are trained to recognize targets in a specified environment subject to the proposed conditions ascribed by a traditional CFAR detector. We show that after an initial learning process, a trained neural network can offer improved detectional performance. The improvement is related to either a lower false alarm rate or a slightly greater probability of detection.