Neural Bezier Interpolation with Manifold Learning for Reliable Vessel Trajectory Prediction

With advancements in technology, large-scale ship operation data become more accessible, leading to growing interest in Automatic Identification System (AIS) research related to maritime transportation. However, the AIS requires long-distance data transmission between sea and land and is more susceptible to environmental conditions, resulting in frequent data loss and corruption. Low-frequency AIS data, caused by persistent transmission data losses, can compromise not only the data quality itself but also the quality of AIS data mining research, especially the reliability of vessel trajectory identification. To resolve this issue, this paper presents Neural Bezier Interpolation (NBI), an interpolation method that combines the predictive capabilities of deep learning model with Bezier Interpolation, aiming to capture and reflect the turning radius of a ship through control points. This hybrid approach mitigates the overfitting problem of deep learning models and overcomes the limitations of traditional interpolation models that cannot independently account for a vessel's speed and heading by integrating Bezier interpolation with deep learning. To verify the effectiveness of the proposed methodology to properly restore low-frequency data, we conduct performance comparison of the proposed approach with existing representative AIS methods. The results confirm that the proposed approach can effectively extract the characteristics of the sea area and improve restoration performance.