As international trade and global security become more reliant on marine resources, the demand for advanced maritime surveillance and port management has never been greater. One of the big challenges in this area is the detection of ships in complex environments, a task that has traditionally relied on manual techniques. These methods, while functional, are often inadequate in dynamic, cluttered marine conditions, where varying sea states, weather patterns, and ship sizes can easily confound detection efforts.
Research in the International Journal of Information and Communication Technology has introduced a new approach to ship target detection. The research combines several cutting-edge deep learning techniques, “You Only Look Once” version 4 (YOLOv4), the Convolutional Block Attention Module (CBAM), and the transformer mechanism. The team of Weiping Zhou, Shuai Huang, and Qinjun Luo of Jiangxi Polytechnic University in JiuJiang, and Lisha Yu of Shanghai Cric information Technology Co. Ltd. In Shanghai, China, have combined these into a single algorithmic program that is both accurate and reliable in the identification of vessels in challenging conditions.
Modern, fast deep-learning models such as YOLOv4 out-class traditional methods by cutting out the multiple steps needed to process an image. YOLOv4 can scan and classify objects in a single pass, making it ideal for real-time surveillance over large expanses.
CBAM is a feature-enhancing technique that works by focusing the model’s attention on the most important elements within a given image. This allows the hybrid system to identify ships even if they are surrounded by other vessels, docks, flotsam, and even rough seas. Conventional techniques often failed in distinguishing vessel from background in such images. The transformer mechanism is a powerful system that further improves the capacity of the model to process features at different levels, ensuring that important detail are not missed.
The team explains that this combined effort allows their system to outperform earlier models, particularly in the detection of smaller vessels and ships in complex maritime environments. They tested the approach on the Ship Sea Detection Dataset (SSDD), which includes remote sensing images of various marine conditions. Their results demonstrated superior speed and precision, especially when identifying minor or obscured targets. Given the critical importance of timely and accurate detection in maritime security, the implications of this improvement are significant.
Zhou, W., Huang, S., Luo, Q. and Yu, L. (2024) ‘Research on a ship target detection method in remote sensing images at sea’, Int. J. Information and Communication Technology, Vol. 25, No. 12, pp.29–45.