Research on Ship Track Clustering Method Based on Optimized Spectral Clustering Algorithm

Abstract

Maritime traffic monitoring is of great significance to the navigation safety of ships, but the main method of supervision by maritime supervision departments is still human monitoring. In order to improve the efficiency of supervision, this paper studies and analyzes the ship trajectory clustering algorithm, which can intelligently classify the unlabeled trajectory data in AIS. Aiming at the problems of low accuracy in detecting abnormal ship trajectory behavior and sensitivity to outliers and noise points in track clustering in existing clustering algorithms, this paper proposes an improved spectral clustering algorithm for ship trajectory clustering. On the one hand, the algorithm improves the affinity distance function to make the clustering more stable and reduce the problem of sensitivity to outliers, on the other hand, it also improves the K-nearest neighbor part in the spectral clustering, the trajectory is mapped to the nodes in the weight graph, and then the distance distribution is calculated by setting a threshold. Finally, based on the data of navigable merchant ships at the Port of Dover in the United Kingdom and the Port of Calais in France, it is verified that the optimized spectral clustering algorithm can improve the computational efficiency and accuracy for ship trajectory clustering, and maintain clustering consistency, the better visual clustering results can be obtained.