A novel hybrid framework for realistic UAV detection using a mixed RF signal database

Merabtine, Nassima; Loscri, Valeria; Djenouri, Djamel; Latif, Shahid

Abstract

Advances in Unmanned Aerial Vehicles (UAVs) empower a plethora of applications but also raise significant security and privacy challenges. Effective UAVs detection systems are crucial for mitigating these risks. This paper deals with this problem and tackles the challenges associated with real-world testing and the limitations of existing simulation methodologies for validating and evaluating UAVs detection protocols. A novel, realistic, and extensible framework is introduced , which includes a MATLAB-based surveillance system, a Python-based detection module utilizing Stacked Denoising Autoencoder (SDAE) and Local Outlier Factor (LOF) algorithms, and a hybrid database of both real and synthetic wireless RF signals. The synthetic wireless dataset is generated by the proposed surveillance system module. The alignment between the synthetic and real data is validated with an average Mean Squared Error (MSE) of less than 0.25. The detection module proves highly effective, achieving 96% accuracy in correctly classifying Wi-Fi signals and 88% accuracy in identifying UAV signals as anomalies (outliers). This innovative approach facilitates ongoing research and development in UAV detection, with the extensibility to incorporate new RF signal types and UAV models.

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: 2024 IEEE Future Networks World Forum (FNWF)
• Start Date: Oct 15, 2024
• End Date: Oct 17, 2024
• Acceptance Date: Sep 17, 2024
• Deposit Date: Sep 18, 2024
• Peer Reviewed: Peer Reviewed
• Keywords: Anomaly Detection; drone detection; UAVs; Machine Learning; Cyber Critical Infrastructures
• Public URL: https://uwe-repository.worktribe.com/output/12895965
• Additional Information: This work has been partially supported by the ASTRID-ANR DEPOSIA project and the Horizon Europe MLsysOps project. It has also been supported in part by the EU CHIST-ERA project (Grant EP/Y036301/1 from EPSRC, UK) and in part by the AGYA Academy (Grant 01DL20003 from BMBF, Germany).