EASY: Exponential Attention with State-Space-Based Yearning Network for Ship AIS Abnomal Detection

Accurate detection of anomalous ship behavior in AIS data remains challenged by signal noise and dynamic maritime patterns. While the Space-air-ground-sea integrated network (SAGSIN) technology has significantly reduced transmission delays and interruptions in maritime AIS communications that previously relied on remote satellite connections, anomalies still occur due to sensor errors and other factors. Identifying these anomalies is critical for maritime safety, efficient transportation management, and security surveillance, but faces challenges in signal quality, pattern complexity, and real-time processing requirements. We propose EASY, a novel framework integrating exponential attention with state-space models for fine-grained anomaly detection. By reformulating the task as token classification, EASY enables precise identification of abnormal trajectory segments while modeling multi-scale temporal dependencies. Evaluations on real-world port data demonstrate SOTA performance of 76.10% in F1, surpassing other models by 11.32%.