Contrastive Learning for Graph-Based Vessel Trajectory Similarity Computation

With the increasing popularity of automatic identification system AIS devices, mining latent vessel motion patterns from AIS data has become a hot topic in water transportation research. Trajectory similarity computation is a fundamental issue to many maritime applications such as trajectory clustering, prediction, and anomaly detection. However, current non-learning-based methods face performance and efficiency issues, while learning-based methods are limited by the lack of labeled sample and explicit spatial modeling, making it difficult to achieve optimal performance. To address the above issues, we propose CLAIS, a contrastive learning framework for graph-based vessel trajectory similarity computation. A combined parameterized trajectory augmentation scheme is proposed to generate similar trajectory sample pairs and a constructed spatial graph of the study region is pretrained to help model the input trajectory graph. A graph neural network encoder is used to extract spatial dependency from the trajectory graph to learn better trajectory representations. Finally, a contrastive loss function is used to train the model in an unsupervised manner. We also propose an improved experiment and three related metrics and conduct extensive experiments to evaluate the performance of the proposed framework. The results validate the efficacy of the proposed framework in trajectory similarity calculation.