Neural Network-Based Moving Target Detection and False Target Discrimination for Synthetic Aperture Radar

Moving targets and spoofed RF signatures represent common signals that can affect synthetic aperture radar (SAR). We describe a machine learning (ML) solution for detecting moving targets and distinguishing digital radio frequency memory (DRFM)-generated RF signatures from ground movers and from ground clutter in synthetic aperture radar. The proposed model employs a residual network (ResNet) to classify SAR chips that contain only ground clutter, a moving target, or a DRFM-generated target. The chips with ground clutter and moving targets are taken from SAR images collected from the Airborne Radar Testbed (ARTB), an MIT Lincoln Laboratory-operated aircraft-mounted radar system. The chips with false RF signatures were generated in software to represent a DRFM-generated target. We present a ResNet-based classifier that is able to achieving high classification accuracy and improve on state of the art algorithm performance.