UAV Detection and Classification in Complex Environments Using Radar and Combined Machine-Learning Approaches

Abstract

Detecting uncrewed aerial vehicles (UAVs) has introduced significant challenges in ensuring safe and secure airspace, particularly in urban areas with high environmental clutter or complex environments. This article proposes a novel two-stage method for UAV detection and classification using a scanning frequency-modulated continuous wave (FMCW) radar system and machine-learning (ML) techniques. In the first stage, azimuth-range scattering point data transformed from the received radar signals are clustered using hierarchical density-based spatial clustering of applications with noise (HDBSCAN), and environmental boundaries are generated with a convex-hull algorithm to represent static clutter zones. In the second stage, a long short-term memory (LSTM) network analyzes points outside these boundaries, leveraging trajectory patterns to classify UAVs. Unlike conventional Doppler-based methods, the proposed approach excels in scenarios with slow-moving UAVs exhibiting near-zero Doppler shifts. Experimental results demonstrate that the proposed method achieves a detection and classification accuracy of up to 99.83% and an F1 score of 94.69%, outperforming conventional methods in both precision and clutter handling. These findings highlight the robustness of the proposed system in complex environments and its suitability for practical UAV detection applications.