基于改进YOLOv8n模型的南极磷虾群智能检测技术研究OA

Antarctic krill(Euphausia superba)is the most abundant biological resource in global fisheries.The advancement of its fishing technology is of great significance for improving fishing efficiency and economic benefits in polar fisheries.Intelligent trawl beams based on sonar image recognition have been gradually applied to Antarctic krill fishing.To address the low-efficiency problem caused by long-time manual sonar image recognition,the study proposes an improved lightweight model,RG-EA-YOLOv8n-seg,for intelligent trawl data preprocessing.In this model,the RepGhost module is introduced to reduce computational cost during inference;a C2f-Trans module is incorporated to enhance global context perception;BiFPN is adopted to strengthen multi-scale feature fusion;and the CoECA attention mechanism is utilized to dynamically suppress background noise.Experimental results on our self-built acoustic dataset demonstrate a precision of 90.8%,recall of 91.4%,and mAP@0.5 of 88.7%,representing improvements of 4.6%,3.9%,and 4.8%respectively over the baseline model,while achieving an inference speed of 27 FPS.Furthermore,the study develops a pixel-level vertical parameter extraction method based on segmentation masks to automatically calculate the center depth,thickness,and seabed depth of krill swarms.With this method,the relative errors of depth and thickness estimation are within±5%,and the seabed depth measurement deviation is less than 0.3 m,providing accurate quantitative support for adjusting the trawl opening depth of intelligent trawls.