pp. 3393-3404
S&M3399 Research Paper of Special Issue https://doi.org/10.18494/SAM4437 Published: September 29, 2023 Enhancing Vessel Trajectory Prediction via Novel Loss Function in Deep Learning Model [PDF] Seung Bae Jeon, Myeong-Hun Jeong, Tae-young Lee, and Dooyong Cho (Received April 24, 2023; Accepted August 18, 2023) Keywords: loss function, deep learning, vessel trajectory prediction, automatic identification system data
Recent developments in data collection technology and sensor precision have led to the generation of large amounts of high-quality data. The vast vessel trajectory data obtained from precise automatic identification system data facilitate the development of marine-related research fields. In particular, vessel trajectory prediction, such as preventing risks in advance or providing efficient routes by predicting the vessel location, is one of the essential parts of advanced vessel traffic service. In this study, the vessel trajectory was accurately and robustly predicted using a novel loss function. In previous studies, the loss function was designed to minimize the distance between the destination and predicted location of vessels, whereas the proposed loss function was designed to minimize the area of the triangle formed by the origin, destination, and predicted location. In experiments, the proposed approach outperformed the state-of-the-art method, reducing the mean absolute error by 12%.
Corresponding author: Myeong-Hun Jeong and Dooyong ChoThis work is licensed under a Creative Commons Attribution 4.0 International License. Cite this article Seung Bae Jeon, Myeong-Hun Jeong, Tae-young Lee, and Dooyong Cho, Enhancing Vessel Trajectory Prediction via Novel Loss Function in Deep Learning Model, Sens. Mater., Vol. 35, No. 9, 2023, p. 3393-3404. |