Adaptive Compression for Ship Trajectory Clustering: A QB-Based Method for Multiple Navigation Scenarios

This paper addresses the issue of the difference between silhouette coefficient (SC) and Davidson–Bouldin Index (DBI) in a single water area, with a view to enhancing the generalization of quick-bundles (QBs) in ship trajectory clustering algorithms. The proposed methodology integrates an adaptive ship trajectory compression for clustering ship trajectories, with the aim of targeting both inland river and open oceanwater areas. It is evident that the considerable temporal disparity between the sampling of the Automatic Identification System (AIS) data in inland waters results in the trajectory compression and sampling process exhibiting suboptimal efficiency and elevated temporal complexity. Consequently, the model requires many iterations to reach convergence during the training process. In view of this, the present work proposes an adaptive ship trajectory compression algorithm that is adaptive in nature. This algorithm employs a sliding window model to divide and compress straight lines and curves in ship trajectories by automatically setting thresholds. This enhancement leads to a substantial improvement in compression efficiency, while retaining the feature information of the original trajectory. The model has been validated to exhibit excellent ship trajectory clustering performance in different waters such as inland rivers (i.e., the Wuhan section of the Yangtze River) and port waters (i.e., Tsawwassen Ferry Terminal area). This model is beneficial for improving the efficiency of intelligent supervision in maritime departments and ensuring the safety of ship navigation.