Recognition of vessel types based on YOLOv8

The continuous development of global international trade has posed significant challenges to maritime security and safe transportation, highlighting the importance of maritime regulation. Traditional ship inspection techniques can identify and locate vessels in images or videos, yet they often fall short of meeting the complex regulatory demands due to specific operational needs and risk assessment standards associated with different types of ships. Consequently, this paper independently constructs a ship image dataset and employs the YOLOv8 algorithm to further identify various types of vessels and their associated shipping companies. This approach provides richer data for maritime regulation, enhances monitoring capabilities, and helps prevent maritime accidents. Experimental results demonstrate that the average recognition accuracy and recall rates for different ship types are 0.902 and 0.909, respectively. Comparative experiments with four types of models, including YOLOv3, YOLOv5, YOLOv6, and YOLOv8n-Ghost,demonstrate that YOLOv8n performs well in terms of accuracy, recall, mAP50, and other metrics. It meets the refined needs for maritime vessel identification, facilitating better regulation of maritime traffic.