Radar-Based Digital Twins for Classification of UAVs and Avian Targets

In this study, the efficacy of range-Doppler imaging is explored for the detection and classification of Unmanned Air Vehicles (UAVs), with attention to the radar system’s operating frequency and bandwidth. The investigation employs full-wave Electromagnetic (EM) CAD software to scrutinize the influence of varied radars, spanning different frequency bands, on the precision of range-Doppler images of a rotating blade. Notably, mmWave radars, distinguished by their expansive bandwidth, demonstrate superior range-Doppler accuracy compared to other examined radar systems. Building on this, a subsequent inquiry is undertaken to evaluate the performance of Machine Learning (ML) algorithms in drone classification amid the presence of avian organisms. The mmWave radar is modeled using EM CAD tools to generate diverse datasets encompassing a quadcopter UAV and avian subjects. Employing two distinct ML algorithms, the study reveals that an increased avian presence diminishes the radar’s ability to effectively detect and classify drones. The CNN model achieves 99% classification accuracy when a single bird coexists with the drone, declining to 90% in scenarios featuring a drone amidst a swarm of ten birds. We believe that our presented workflow presents a paradigm shift in how defense scientists can validate possible counter measures against illicit uses of compact drones.