Lightweight Fishing Vessel Trajectory Prediction Model based on Dual-attention

With the increasing volume of shipping, fishing vessel trajectory prediction has become particularly important in ensuring route safety and optimizing path planning. However, existing methods often have a large bias in long time prediction and have a high demand for computational resources. To solve this problem, this paper proposes a fishing vessel trajectory prediction model based on the dual attention mechanism by combining the Transformer self-attention and frequency domain enhanced channel attention (FECAM). The model firstly represents the AIS data by high-dimensional embedding vectors, which overcomes the limitation of the traditional 4-dimensional vectors, and utilizes a Transformer to capture the long-term dependencies in vessel trajectories. Then, FECAM enhances the model’s ability to identify key features by optimizing the frequency domain feature extraction, while reducing the interference of high-frequency noise. Experimental results show that the proposed model outperforms existing methods in prediction accuracy and significantly reduces the number of parameters and computational overhead of the model. Especially in long-time prediction, the model shows better accuracy and computational efficiency, which is highly practical and resource-efficient for resource-constrained devices and embedded systems.