Compact CNNs for Real-Time Drone Detection via RF Emissions.

The proliferation of first-person-view (FPV) drones in modern conflicts has created an urgent need for compact and reliable detection systems, especially those capable of operating in complex electromagnetic environments. These drones continuously emit video signals, which generate distinct radio-frequency (RF) emissions that can be exploited for early detection. This research investigates the application of lightweight convolutional neural networks (CNNs) for automated drone detection. The system processes drone signals captured by a software-defined radio (SDR) framework, converting them into rasterized time-domain images. This input representation is computationally efficient, making it suitable for deployment on embedded systems with limited resources. Custom CNN architectures were designed and benchmarked against a dataset of approximately 40,000 labeled images, evaluating accuracy, model size, and inference performance. The models were also integrated into a GNU Radio signal processing chain for real-time evaluation. The results confirm that these compact CNNs can achieve high detection accuracy while maintaining low computational requirements, making them ideal for embedded RF monitoring applications. This approach offers comparable accuracy to existing spectrogram-based methods but with significantly reduced computational overhead by eliminating frequency-domain preprocessing.