Drones identification and classification using fingerprints in spectrograms

Abstract

The rapid proliferation of drones and Wi-Fienabled devices has revolutionized various sectors, including agriculture, entertainment, security, and surveillance. However, this also has magnified the threat space in terms of security, privacy, and efficient spectrum management. Detecting and classifying these devices accurately is crucial to address potential threats to public safety. To alleviate this issue, this paper proposes an advanced signal classification framework to identify drones base on their unique fingerprint. This is done by using spectrogram images of different drones and Wi-Fi devices operating within the 2.4 GHz spectrum which give unique patterns to identify drones fingerprint. The approach combines the features generated by Principal Component Analysis (PCA) with a modulation index to enhance classification accuracy and robustness of different machine learning classifiers. Two tasks are considered in this paper: i) multi-class classification of different drone models and ii) binary classification of drones and Wi-Fi signals. The proposed framework is rigorously tested and challenged using different hyperparameters configurations and ablation studies. The results demonstrate the robustness of the proposed approach in identifying drones accurately.