RF Based Enhanced Drone Detection and Classification Using Short Time Fourier Transform

The proliferation of unmanned aerial vehicles (UAVs) has led to an increased demand for effective drone detection systems. The paper proposes a novel approach for drone detection leveraging Radio Frequency (RF) signals and employing two unique feature extraction techniques based on Short-Time Fourier Transform (STFT). In Method I, we applied STFT to the time series signal segments of DroneRF dataset to generate spectrogram capturing the change in frequency signatures with time for background RF activity and different drone operations. In Method II, we analyzed the spectrograms and extracted 10 statistical features for each time column for background activity and different drone operations. We used ensemble learning classifier consisting of XGBoost and KNN for detection and classification of drones due to large similarity in samples in terms of distribution of RF signal. We compared both the methods for various evaluation metrics, number of features and processing time for classification in 2-class (drone vs no drone), 4-class (type of drone) and 10-class scenario (mode of operation). We found that Method I performed better for 10class classification while Method II used less number of features and carried out classification in less processing time as compared to Method I.