关键词: 交通运输工程, 运动模式, 单向距离, 谱聚类, 相似性度量

Abstract: The vessel motion pattern recognition from marine large data can discovery high-level situational awareness, which improves the efficiency of maritime surveillance technology. Therefore, a motion pattern of vessel recognition methods based on one-way distance spectral clustering algorithm was proposed. The anti-interference of the one-way distance measurement was fully used to form the similarity value of trajectory based on one-way distance, and then the similarity matrix of the trajectory was also obtained. The regular motion patterns of vessels were extracted from the trajectories spatial distribution learnt by the spectral clustering algorithm in unsupervised learning way. Finally, the motion patterns for vessels entering into Xiuying port were analyzed by using real AIS data sample in Qiongzhou strait. And distance of inner pattern class and distance between of pattern class were calculated separately. Four obtained typical patterns were in consistence with those of the real traffic. The experiment results demonstrate that the proposed clustering methods is of good performance and well fit for identifying motion patterns of vessel in areas, such as coastal ports, narrow channels and complicated traffic zones.

Key words: transportation engineering, motion pattern, one-way distance, spectral clustering, similarity measure