Clustering Algorithm for Vessel Trajectory Anomaly Detection Based on Binary Hashing Method

Timely identification of abnormal ships through trajectories make daily commercial trade transportation safer. However, current methods for identifying abnormal ships through vessel trajectories have issues such as high time cost and high sensitivity to parameters. To address such issues efficiently, a clustering method based on hashing algorithm applying Long Short-Term Memory neural networks is proposed to capture abnormal local and temporal information in trajectories. The trajectories are converted into hash codes that present thier characteristics succinctly. Then, the dominant set clustering method that does not require setting hyperparameters is used to cluster and find abnomal trajectories according to similarity relationship computed by corresponding hash codes. The proposed method is validated on a dataset of real vessel trajectories, which basically achieves the identification of abnormal vessel trajectories with lower time cost compared to other methods.