In the Realm of Aerial Deception: UAV Classification via ISAR Images and Radar Digital Twins for Enhanced Security
Ahmed N. Sayed, Omar M. Ramahi, George Shaker
Abstract
Unmanned Aerial Vehicles (UAVs) pose significant security challenges due to their widespread use in various malicious activities, including terrorist attacks and wartime operations where explosives are attached to them. Conventional radar-based detection and classification methods often rely on range-Doppler signatures, which may lead to misclassification, especially in identifying UAVs carrying explosive payloads. To address this challenge, this letter proposes an inverse synthetic aperture radar (ISAR)-based classification algorithm. To develop and validate the proposed algorithm, a quadcopter is modeled using radar digital twins to generate comprehensive datasets. Initially, a convolutional neural network (CNN) classifier is trained on a dataset comprising range-Doppler information, aiming to distinguish between a commercial quadcopter and the same quadcopter when it is modified to carry explosives. However, the model fails to accurately classify instances where the quadcopter is carrying explosives-based solely on range-Doppler data. Subsequently, the CNN model is retrained using a dataset containing ISAR images for both scenarios. When applied to a separate dataset featuring ISAR images of a quadcopter carrying explosives, the model demonstrates improved accuracy in classification. Real-measurements further validate these findings, confirming the effectiveness of the proposed ISAR-based classification approach in enhancing radar security against UAV-borne threats. This research presents valuable insights for the future development of robust countermeasures to address the evolving challenges posed by UAVs in security-sensitive environments.