Vessel trajectory classification from a graph network perspective

Maritime transportation is the cornerstone of the global economy, making understanding vessel activities and their impacts vital. This paper presents a graph-based approach for classifying navigational statuses of vessel trajectories, employing a hierarchical multi-classification deep learning framework using Automatic Identification System (AIS) data. AIS messages are first segmented into continuous tracks, then divided into fixed-size mini-trajectories called a ‘trajectlet’. Labels are assigned using domain knowledge and rule-based methods. Each trajectlet is modelled as a directed acyclic graph, capturing sequential and spatio-temporal movement dynamics. A Graph Neural Network (GNN) first classifies the AIS messages within each trajectlet as ‘stationary’ or ‘underway’. These node-level predictions are then pooled and combined with global features of each trajectlet, and then passed through a multi-layer perceptron to classify the entire trajectlet into one of four navigational statuses. The framework is validated on AIS data from vessels in UK waters, achieving 98% accuracy and F-score, and a 99% AUC–ROC, demonstrating strong predictive performance. This method enhances understanding of vessel operations and supports applications in emissions modelling, economic forecasting, autonomous navigation, and maritime situational awareness. • Models AIS messages as graphs with spatio-temporal and sequential links. • Classifies vessel status using feature-level attention in a GNN framework. • Combines node and graph-level outputs in a hierarchical multi-level classification. • Achieves 98% accuracy and F1-score, 99% AUC–ROC on UK AIS data. • Supports autonomy, port monitoring, emissions and economic modelling.