Simulation-based Approach to Classification of Airborne Drones

Recognition of drone type provides valuable information to assess the capability of drones, which is essential to airspace monitoring. Classification of drones on the basis of radar data is dominated by the use of supervised learning, which exploits different and often combined representations of the micro-Doppler signatures of the target. However, it is expensive and cumbersome building a catalogue of several drone micro-Doppler signatures using real data. We introduce a simulation-frame-work to generate radar data from point-scatterer targets, with associated radar cross section evaluated using physical optics. Small scale lab tests validate the fidelity of the simulated radar data, while the utility of the synthetic data for classification is tested using established methodology for classification.