PUBLISHED PAPERS
2025 |
Drone Classification from RF Signals: A Comparative Study of Convolutional Networks and Attention Mechanisms Proceedings Article Luis Paulo Albuquerque Guedes; Pedro Henrique Monteiro Guedes; Remulo Caminha Resumo | Links | BibTeX | Tags: CNN, RF-based UAV Classification, Vision Transformers @inproceedings{Guedes2025DroneClassification,The classification of unmanned aerial vehicles (UAVs) via radio-frequency (RF) signals employs advanced signal-processing and machine-learning techniques to identify and categorize emissions, playing a fundamental role in security and surveillance applications in sensitive environments. The present study conducts a comparative analysis between the VGG-16 and Transformer architectures, aiming to identify preprocessing and model configurations that maximize classification accuracy for drone RF signals without compromising computational feasibility in defense-embedded systems. Applying the VGG-16 model with 20 ms time blocks resulted in approximately 97% accuracy and F1-score, outperforming classical methods (linear regression and k-NN) by up to 17 percentage points. Furthermore, it was found that all deep models exhibited significant gains when operating on spectrogram inputs, substantially surpassing traditional approaches. |