A Multi-AttentionLSTM model for multi-ship encounter classification

Maritime situation awareness is critical for ensuring safe navigation and efficient operational conduct within the maritime domain. Multi-ship encounter situations at sea, compared to pairwise ship encounters, are characterized by greater complexity and uncertainty, posing significant challenges to decision-making processes. Identifying ship evasive maneuvers during these encounters is crucial for recognizing critical situations. Therefore, this paper proposes a supervised Multi- AttentionLSTM model for multi-ship situation classification. This model detects the presence of evasive maneuvers in multi-ship encounter situations. The model incorporated multiple branches by utilizing an attention mechanism-enhanced Long Short-Term Memory (LSTM) approach for classifying multivariate time series data. Experiments on expert-labeled, real-world Automatic Identification System (AIS) data demonstrate the effectiveness of our proposed method.