ISAR Image Drone Classification with CNN Using Millimeter-Wave Fast Chirp Modulation MIMO Radar

With the development of drone technology, concerns have been raised regarding the potential application of drones in terrorism and other crimes. Accordingly, a drone detection system that can classify incoming drones is needed to contain potential drone threats. In this study, we generate the inverse synthetic aperture radar (ISAR) imagery of various drones using millimeter-wave (mmW) fast chirp modulation (FCM) multiple-input and multiple-output (MIMO) radar and propose a drone classification method to distinguish the generated ISAR imagery using convolutional neural networks (CNNs). Two experimental cases were investigated to demonstrate the effectiveness of our proposed method. In case 1, we tested five types of drones (3DR Solo, DJI Phantom 3, DJI Mavic Pro, Parrot Anafi, and DJI Mavic Mini) moving under ideal conditions in the laboratory and generated the ISAR images of the drones. The models of five drones could be classified with high accuracy by learning the features of the ISAR images. In case 2, we classified the same models of flying drones using trained CNN models based on their ISAR images. Notably, its classification accuracy was comparable to that of other studies in drone classification. The experimental results indicated that ISAR imagery features are valid for drone classification.