A Novel Hybrid Approach for Detecting Abnormal Vessel Behavior in Maritime Traffic

The Abnormal Behavior Detection of Vessels (ABVD) is crucial for enhancing Vessels’ situational awareness as well as achieving effective management of maritime traffic. Clustering algorithms are extensively employed for ABVD tasks, however, they are not only time-consuming but also frequently inaccurate when dealing with complex and dynamic vessel trajectory detection tasks. To address the above problems, we propose a novel hybrid approach, CSDLSTM(Course/Speed-DBSCAN and LSTM), which combines Density-based Clustering(DBSCAN) and long short-term memory networks to improve the efficiency of ABVD. Specifically, a three-dimensional CS-DBSCAN (Course/Speed-DBSCAN) algorithm is proposed, which incorporates speed and course features along with geospatial information in the clustering process to accurately describe vessel behavior. In addition, the clustering results are used as traffic patterns to train the LSTM predictor for real-time maritime anomaly detection. Experiments conducted on real AIS datasets demonstrate that the CSD-LSTM method has greater precision and efficiency in ABVD.