Designing an attention-based approach for RF fingerprinting in drone detection and classification

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• The proliferation of unmanned aerial vehicles (UAVs) necessitates advanced radio frequency (RF) fingerprinting for detection and classification.
• Existing methods often lack the efficiency and accuracy required for real-world UAV monitoring.

Purpose of the Study:

• To introduce a computationally efficient transformer-based attention architecture for analyzing RF signals from UAVs.
• To enable intelligent and accurate detection, identification, and classification of UAVs and their operational modes.

Main Methods:

• A two-stage feature extraction pipeline involving correlation filtering and ANOVA feature selection.
• A multi-headed self-attention encoder to process spectral sequences of RF signals.
• Evaluation on the DroneRF and VTI_DroneSET_FFT datasets, including single- and multi-drone scenarios across different frequencies.

Main Results:

• Achieved high cross-validation accuracy: 100.00% for detection, 99.80% for classification, and 99.23% for flight mode classification on DroneRF.
• Demonstrated strong performance on the challenging VTI_DroneSET_FFT dataset, reaching 88.10% accuracy in a complex multi-drone scenario.
• The model's low parameter count and FLOPs make it suitable for deployment on edge devices.

Conclusions:

• The proposed transformer architecture offers a computationally efficient and highly accurate solution for UAV RF fingerprinting.
• The system demonstrates robust performance in diverse real-world scenarios, including multi-drone environments.
• Attention analysis confirms the effectiveness of the model in identifying and characterizing drones across various RFs.