SteadySeg: Improving maritime trajectory staging by steadiness recognition

Identifying the maritime activities of vessels over time is crucial for enhancing maritime traffic management and marine ecosystem conservation, by facilitating informed decision-making among maritime authorities, shipping companies, and researchers. Current methods focusing on trajectory staging for maritime activity recognition, which include both trajectory segmentation and classification, often overlook the segmentation aspect, thereby analyzing whole vessel trajectories without recognizing the diverse stages these trajectories exhibit. To address this gap, we introduce SteadySeg (Steadiness Recognition for Maritime Trajectory Segmentation), a novel approach that segments a vessel’s trajectory into discernible pattern features, identifying distinct states and sequences of state boundaries, with a design that accounts for the infrequent state changes typical in real-world maritime trajectories. We construct two maritime activity datasets, i.e., one from the sea area near Denmark and the other from waters off the United States, to evaluate SteadySeg. Experimental results show that SteadySeg improves trajectory staging accuracy by at least 6% compared to existing time-series staging models, representing a significant advancement in maritime trajectory analysis. •Introduce SteadySeg, an innovative model for maritime trajectory staging.•Predict state sequence without abrupt state transitions in short time intervals.•Create a vessel dataset with detailed navigational status and fishing gear usage.