An Enhanced End-to-End Framework for Drone RF Signal Classification

Smart RF jamming relies on long-term spectrum prediction, which requires accurate, high-resolution RF detection and classification over extended observation periods. Detecting and classifying drone RF signals is particularly challenging due to short dwell times, high hopping rates, and narrow instantaneous bandwidths. This paper presents an enhanced end-to-end framework designed to meet these requirements for smart RF jamming, delivering high-resolution and precise detection and classification. We demonstrate that our Residual Neural Network (ResNet)-based You Only Look Once (YOLO) model effectively detects and extracts RF features from previously unseen drone signals with high accuracy, even when trained solely on a synthetic RF dataset. Furthermore, our ResNet classifier outperforms existing models, achieving 99.29% accuracy at 0 dB signal-to-noise ratio (SNR) for drone RF signals.