Signal-duration-based drone classification using YOLO detection on time-frequency spectrograms

This study proposes a method for classifying drone signals based on signal duration by detecting signal regions in a spectrogram. The in-phase and quadrature (I/Q) data of the received radio frequency (RF) signals are transformed into a time–frequency spectrogram representation of drone communication signals, and You Only Look Once (YOLO)–based object detection is applied to extract signal segments. The width of the detected region in the time domain is then measured to estimate signal duration, which is used as a feature for classification using a K-Nearest Neighbors (KNN) classifier. By combining YOLO-based signal detection with KNN-based classification, the proposed method effectively identifies drone communication signals. The classification accuracy of the proposed method is evaluated for three drone protocols, including DJI OcuSync, DJI Lightbridge, and Autel SkyLink. Experimental results show that the proposed YOLO-based detection method provides more accurate and robust classification performance than the previous energy-based method, reducing the required SNR by approximately 3 dB to achieve 90% classification accuracy.