An anomaly detection method based on ship behavior trajectory

Abstract

An anomalous trajectory detection method based on ship trajectory clustering and prediction is proposed. The method consists of two modules, namely, trajectory clustering based on improved DBSCAN and Trajectory prediction by ProbSparse Attention-based Transformer. we propose the concept of ship behavior similarity for the first time in this paper, by decomposing the velocity according to the ship's heading to get ‘velocity coordinates' to replace latitude and longitude coordinates, on the basis of which the ship behavior trajectories are clustered and the core trajectories are extracted. By comparing the behavioral similarity between the predicted trajectory and the core trajectory with the preset threshold, it is judged whether the abnormal trajectory or the ship has abnormal operation. AF-DP is used in this method for compression of trajectory data, and Fast-DTW is used to calculate the behavioral similarity between trajectories. Notably, this is the first application of ProbSparse Attention-based Transformer in the maritime domain. The proposed method is applicable to inland river traffic near Wusongkou, Shanghai. The final experimental results show that the method we proposed can effectively identify the abnormal behaviors in the trajectory, and can provide technical support for the analysis of ship track data and the risk management of maritime transport in this region.