ABSTRACT The fast emergence of unmanned aerial vehicles (UAVs) has increased significant security and privacy concerns. Radio frequency (RF)‐based drone detection has emerged as an emerging solution because of its potential in operating independently of lighting conditions and line‐of‐sight constraints. However, RF signals captured in real‐time environments are generally affected by interference, noise, and complex temporal dynamics. This makes accurate drone detection a challenging task. To tackle these problems, this work presents an optimization‐based hybrid deep learning (DL) model, Transformer‐Bidirectional Long Short‐Term Memory (BiLSTM) model for robust RF‐based drone detection and classification. In the suggested work, RF signals are first transformed into time‐frequency representations by short‐time Fourier transform (STFT). A Transformer encoder with multi‐head self‐attention is presented for capturing long‐range temporal dependencies and the BiLSTM captures sequential characteristics of RF signals. To further improve detection performance, the Enhanced Nutcracker Optimization Algorithm (ENOA) is integrated for optimizing major hyperparameters of the model. Experimental analysis on noisy drone RF signal datasets shows that the suggested model outperformed existing DL on the basis of different measures. The outcome confirms the effectiveness of the suggested model for reliable RF‐based drone detection in noisy electromagnetic environments.
Kingsley et al. (Fri,) studied this question.