Automatic Target Recognition of Aircraft using Inverse Synthetic Aperture Radar

Following the recent advancements in radar technologies, research on automatic target recognition using Inverse Synthetic Aperture Radar (ISAR) has correspondingly seen more attention and activity. ISAR automatic target recognition researchers aim to fully automate recognition and classification of military vehicles, but because radar images often do not present a clear image of what they detect, it is considered a challenging process to do this. Here we present a novel approach to fully automate a system with Convolutional Neural Networks (CNNs) that results in better target recognition and requires less training time. Specifically, we developed a simulator to generate images with complex values to train our CNN. The simulator is capable of accurately replicating real ISAR configurations and thus can be used to determine the optimal number of radars needed to detect and classify targets. Testing with seven distinct targets, we achieve higher recognition accuracy while reducing the time constraints that the training and testing processes traditionally entail.