Radar detection of small UAVs in severe ground clutter using 2D spatial-temporal matched filtering

Abstract Based on the empirical characterization of radar signatures from three distinct Unmanned Aerial Vehicles (UAVs)—the DJI Mini 4 Pro, Mavic 3 Pro, and Phantom 4 Pro—this paper proposes a detection framework applicable to both Frequency-Modulated Continuous Wave (FMCW) and pulsed radar architectures. Unlike conventional approaches that rely on theoretical point-target models, we design a two-dimensional spatial-temporal matched filter. This approach addresses the challenge of detecting low Radar Cross-Section (RCS) targets in severe ground clutter, where standard Constant False Alarm Rate (CFAR) algorithms frequently suffer from threshold breakdown. By shifting the paradigm from amplitude thresholding to 2D pattern matching via Normalized Cross-Correlation (NCC), the proposed method integrates target energy across both fast-time (range) and slow-time dimensions. The study presents validation comprising both Monte Carlo simulations and outdoor field trials. Simulation results demonstrate sensitivity gain of 10 to 14 dB over standard CA-, GO-, SO-, and OS-CFAR methods in Rayleigh and Weibull clutter. Crucially, experimental validation on the outdoor field radar data, for the three small UAVs, confirms the robustness of the 2D matched filter, which consistently achieves a Precision-Recall Area Under the Curve (AUC) between 0.93 and 0.99, effectively suppressing false alarms even for low Signal-to-Clutter Ratio (SCR).