A Survey on Deep Learning Methods for UAV Frequency-Hopping Signal Detection

Abstract

To address the security threats posed by unauthorized UAV operations, commonly known as "black flights," traditional frequency-hopping signal detection methods often exhibit low accuracy and poor adaptability in environments characterized by low signal-to-noise ratios and complex electromagnetic conditions. This paper provides a systematic review of deep learning-based methods for detecting UAV frequency-hopping signals, followed by a detailed comparative analysis of three representative approaches. First, we introduce a detection method that synergizes compressed sensing with multi-level deep learning, which achieves the highest reported detection accuracy (≥99.3%) in controlled settings. Next, we describe an improved YOLOX-tiny model that incorporates an attention mechanism and a lightweight architecture, achieving a real-time processing rate of 57 FPS in mobile scenarios with a 40.6% reduction in parameters. Finally, a two-stage anti-interference framework named YOLOv3-CNN, which is designed for complex electromagnetic environments and demonstrates remarkable robustness (≥96% accuracy) under non-line-of-sight and strong interference, is analyzed. The results confirm that deep learning effectively breaks through the limitations of traditional methods in complex settings. Multi-scenario generalization, low-cost hardware adaptation, and integrated multi-task system development emerge as three key directions for future research to pave the way for the technology's transition from lab to field.