Bridging Optical, Infrared, and Satellite Insights
Yichen Feng,Hui Yin,Hao Zhang,Langtao Wu,Haihui Dong,Jiawen Li
DOI: https://doi.org/10.3390/jmse12122203
IF: 2.744
2024-12-03
Journal of Marine Science and Engineering
Abstract:Image-based ship monitoring technology has extensive applications, and is widely used in various aspects of port management, including illegal activity surveillance, vessel identification at entry and exit points, channel and berth management, unmanned vessel control, and incident warning and emergency response. However, most current ship identification technologies rely on a single information source, reducing detection accuracy in the complex and variable marine environment. To address this issue, this paper proposes a knowledge transfer-based ship identification system integrating three modules. The system enables synchronized monitoring of visible light coastal images, satellite cloud images, and infrared spectrum images, thereby mitigating problems such as low detection accuracy and poor adaptability of image recognition. Additionally, extensive supplementary experiments were conducted to evaluate the effectiveness of the preprocessing and data augmentation modules as well as the transfer learning module. The study also discusses the limitations of current deep learning-based ship monitoring models, particularly their poor adaptability to image recognition and inability to achieve all-weather, round-the-clock monitoring. Experimental results based on three ship monitoring datasets demonstrate that the proposed system, by integrating three distinct detection conditions, outperforms other models with an F1-score of 98.74%, approximately 10% higher than most existing ship detection systems.
oceanography,engineering, marine, ocean