Prediction-based trajectory anomaly detection in UAV system with GPS spoofing attack

The Global Positioning System (GPS) plays an indispensable role in the control of Unmanned Aerial Vehicle (UAV). However, the civilian GPS signals, transmitted over the air without any encryption, are vulnerable to spoofing attacks, which further guides the UAV on deviated positions or trajectories. To counter the GPS spoofing on UAV system and to detect the position/trajectory anomaly in real time, a motion state vector based stack long short-term memory trajectory prediction scheme is firstly proposed, leveraging the temporal and spatial features of UAV kinematics. Based on the predicted results, an ensemble voting-based trajectory anomaly detection scheme is proposed to detect the position anomalies in real time with the information of motion state sequences. The proposed prediction-based trajectory anomaly detection scheme outperforms the existing offline detection schemes designed for fixed trajectories. Software In The Loop (SITL) based online prediction and online anomaly detection are demonstrated with random 3D flight trajectories. Results show that the coefficient of determination ( R 2 ) and Root Mean Square Error (RMSE) of the prediction scheme can reach 0.996 and 3.467, respectively. The accuracy, recall, and F1-score of the proposed anomaly detection scheme can reach 0.984, 0.988, and 0.983, respectively, which outperform deep ensemble learning, LSTM-based classifier, machine learning classifier and GA-XGBoost based schemes. Moreover, results show that compared with LSTM-based classifier, the average duration (from the moment starting an attack to the moment the attack being detected) and distance of the proposed scheme are reduced by 24.4% and 19.5%, respectively.