Cascaded learning empowered classification of UAVs using radio frequency under wireless interference

Autonomous UAVs are increasingly valuable in disaster response, imaging, agriculture, defense, and public services. However, they pose security risks if misused near sensitive areas like airports and power plants. Rapid UAV identification is essential for safety, and deep learning (DL) algorithms offer effective solutions for automatic drone detection and classification across diverse scenarios. To leverage recent advances in DL technology, this paper proposes a novel DL-based cascaded model that combines convolutional neural networks (CNN) with bidirectional long short-term memory (BiLSTM) networks for the precise classification of UAVs. The DroneRC and CARDRF datasets are used in this simulation study. Before training the ML models, raw RF data is preprocessed using the short-time Fourier transform, and the power spectral density approach is employed to extract the most relevant features. The feature extraction of RF signals from various drones is performed using grayscale values instead of RGB channels. Additionally, to evaluate the model’s effectiveness in robustly classifying drones, it is trained in the presence of wireless interference, such as WiFi and Bluetooth signals. To assess efficacy, various DL algorithms (CNN, LSTM, CNN-LSTM, KNN, SVM) were configured identically for comparison. According to the results, the suggested model has a low error rate and good accuracy. • Proposes a novel DL model combining CNN and BiLSTM networks for accurate UAV classification. • Raw RF data is processed using the STFT, and key features are extracted through the PSD method before model training. • The model’s robustness is assessed by training it in the presence of wireless interference from sources like WiFi and Bluetooth signals. • For performance evaluation, the proposed model is compared with CNN, LSTM, CNN-LSTM, KNN, and SVM, with identical configurations.