A new machine learning framework designed to detect malicious interference in unmanned aerial vehicles (UAVs), commonly known as drones, has shown strong performance in identifying both sudden and slow-developing sensor attacks, according to research in the International Journal of Automation and Control. The system, called GTF-MAD (Graph Time-Frequency Mixed Anomaly Detection), achieved a peak F1-score of 99.71% in detecting bias in tests on a quadrotor drone.
UAVs depend on sensors such as GPS (which provides satellite-based location data) and gyroscopes (which measure rotation and orientation). These act as the drone’s navigational senses. However, they are vulnerable to manipulation. GPS spoofing can feed false location signals to a drone, while gyroscope bias injection introduces small but persistent errors into motion readings. Both can accumulate into major navigation failures if undetected.
Traditional detection systems rely on fixed rules, physical flight models, or machine learning patterns in sensor data. However, they struggle with changing sensor relationships during flight, lack of frequency-based signal analysis, and difficulty detecting slow-burn attacks that evolve over time.
GTF-MAD addresses these issues through three components. An adaptive graph attention network models sensors as a dynamic system of relationships that change during flight. A dual time-frequency architecture analyses signals both as time sequences and as frequency patterns, capturing vibrations and periodic disturbances. A trend detection module combines statistical methods to identify slow, stealthy deviations.
Chen, J., Zhou, Y. and Xue, X. (2026) ‘Time series data-driven UAV sensor attack detection: an adaptive graphtime-frequency hybrid approach’, Int. J. Automation and Control, Vol. 20, No. 7, pp.1–25.
No comments:
Post a Comment