RF-Based Lightweight Machine Learning for Comprehensive Drone Activity Classification

The rapid rise of commercially available drones, driven by miniaturization and affordability, has introduced potential security risks alongside numerous societal benefits. While drones serve various industries, their unregulated use necessitates robust detection and classification methods. This paper proposes a novel, efficient model for classifying unwanted drone activity. We leverage statistical and time-domain features extracted from radio frequency (RF) signals and machine learning (ML) classifiers. This approach aims to not only detect drones but also distinguish their models and operational modes. Utilizing the DroneRF dataset, our feature selection strategy prioritizes informative signals in the low-band spectrum, achieving ${100\%}$ accuracy in both drone detection and drone model identification tasks. Furthermore, our hybrid model, combining time-domain and frequency-domain features, achieves an accuracy of ${97.66\%}$ in classifying different flight modes. This demonstrates the effectiveness of this comprehensive feature extraction approach for drone activity classification.