融合EM检测与VMS聚类互证的秋刀鱼捕捞行为识别OA

Pacific saury(Cololabis saira)is a typical pelagic economic species in the northwest Pacific,and fluctuations in its resources and the effectiveness of fishery regulation have attracted increasing attention.Conventional vessel monitoring system(VMS)data provide high-precision position information,but lack visual and behavioral details of on-board operations,limiting their ability to support fine-scale identification of fishing activities.In this study,we develop a fishing activity recognition method for Pacific saury stick-held dip-net fishing vessels by integrating electronic monitoring(EM)images with VMS data,aiming to achieve refined discrimination and visual reconstruction of vessel operating states.Based on EM images from the processing area and time-synchronized VMS data from a single vessel and season,we construct an automated catch event recognition framework using a YOLOv11n model for the processing-cabin scene.Catch events are identified according to a"fish-crew co-detection"criterion.In parallel,a Gaussian mixture model(GMM)is applied to cluster VMS vessel speeds,providing an auxiliary classification of operating states.The EM-based catch event time series and speed-based GMM clusters are then jointly used for dual-modal verification.Results show that the YOLOv11n model achieves classification accuracies of 99%for crew and 97%for fish(Pacific saury),while the agreement between EM-derived catch/non-catch states and VMS speed clusters reaches 94%after excluding missing VMS records.The proposed"visual-trajectory"dual-modal verification framework significantly enhances the verifiability and robustness of fishing activity identification,enabling both visual auditability via EM and quantitative support via VMS,and aligns with regulatory requirements for Pacific saury fisheries that are measurable,reportable,and verifiable.