MoCo-AIS Enhances Vessel Trajectory Similarity Learning

Analyzing mobility patterns, such as extracting route patterns, predicting movement, and detecting anomalies, fundamentally relies on accurately computing trajectory similarity. Traditional methods for this, based on distance measures, are computationally intensive. This has led to a push towards more efficient, learning-based approaches. While supervised learning methods can be used, they require extensive labeled data and often merely replicate existing distance metrics, limiting their ability to generalize. Self-supervised learning, particularly through contrastive learning, offers a solution by learning from unlabeled data. However, a standardized framework for comparing different deep learning models in this context has been lacking, making it difficult to assess consistent trajectory representations. To address this, researchers introduce MoCo-AIS, a unified framework for learning vessel trajectory embeddings. It is built upon the Momentum Contrast (MoCo) paradigm, which structures similarity learning by creating positive and negative pairs of trajectories. Using this framework, a variety of leading deep learning models were evaluated on large, real-world vessel-tracking AIS datasets, encompassing diverse navigation behaviors and operating conditions. The results indicate that MoCo-AIS substantially improves similarity learning compared to current baseline methods and also provides a valuable platform for benchmarking future trajectory representation models.