Unsupervised Anomaly Detection of Unmanned Aerial Vehicles based on Spatial-Temporal Model with Flight Data

Anomaly detection of Unmanned Aerial Vehicles (UAV) has gained great importance to guarantee safety. In this paper, we propose an unsupervised anomaly detection framework based on spatial-temporal feature learning of available multivariate flight data. First, a spatial-temporal prediction model is designed based on convolutional neural network (CNN) and long short-term memory (LSTM), which can extract the spatial-temporal features from multivariate time series sensor variables in a sequential manner. Then, a low-pass filter is utilized to smooth the residuals to decrease the impact of noise on anomaly detection performance. Finally, anomaly detection is achieved by comparing the smoothed residuals with a statistical threshold. A real flight dataset is used to verify its validation and experimental results demonstrated that our proposed method outperforms several state-of-the-art methods.