Accurate extraction of functional areas in port waters is essential for enhancing port operational oversight, optimizing vessel scheduling, and supporting maritime safety. However, existing approaches often rely on supervised learning, extensive parameter tuning, and labeled datasets, limiting their scalability, adaptability, and operational efficiency. To address these gaps, this study proposes PortMiner, a novel unsupervised data mining framework that systematically extracts functional areas from raw vessel trajectory data without requiring manual annotations. The framework introduces a Spatio-Temporal Adaptive Sliding Windows (STASW) method to detect stop behaviors dynamically, using self-adaptive parameters derived directly from the data. Trajectories are first encoded into geohash-based sequential grids, enabling efficient detection of stop and port inbound/outbound behaviors. Functional zones such as berths, anchorages, and navigational channels are then delineated through multi-level spatial aggregation and connectivity-based clustering. Experimental results on benchmark datasets show that STASW achieves 98.83% accuracy, outperforming state-of-the-art deep learning methods, while significantly reducing computational time and cost. Validation against official nautical charts confirms PortMiner’s high fidelity in identifying port-functional structures. The extracted results are also made publicly accessible via an interactive platform ( https://portminer.netlify.app/ ), offering practical insights for intelligent port operation and maritime logistics planning.
Qiang et al. (Mon,) studied this question.