基于YOLOv8模型的凡纳滨对虾(Litopenaeus vannamei)摄食强度量化及分类方法研究  

YOLOv8 BASED FEEDING INTENSITY OF LITOPENAEUS VANNAMEI QUANTIFICATION AND CLASSIFICATION METHODS

作　　者：王磊 赵海翔[2,3] 崔鸿武 黄桢铭 高阳 李皓 崔正国 曲克明[2] 朱建新[2] WANG Lei;ZHAO Hai-Xiang;CUI Hong-Wu;HUANG Zhen-Ming;GAO Yang;LI Hao;CUI Zheng-Guo;QU Ke-Ming;ZHU Jian-Xin(Fisheries College,Zhejiang Ocean University,Zhoushan 316022,China;Key Laboratory of Sustainable Development of Marine Fisheries,Ministry of Agriculture and Rural Affairs,Yellow Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Qingdao 266071,China;College of Fisheries and Life Science,Shanghai Ocean University,Shanghai 201306,China;Fisheries College,Ocean University of China,Qingdao 266003,China)

机构地区：[1]浙江海洋大学水产学院,浙江舟山316022 [2]中国水产科学研究院黄海水产研究所农业农村部海洋渔业与可持续发展重点实验室,山东青岛266071 [3]上海海洋大学水产与生命学院,上海201306 [4]中国海洋大学水产学院,山东青岛266003

出　　处：《海洋与湖沼》2024年第5期1282-1293,共12页Oceanologia Et Limnologia Sinica

基　　金：国家重点研发计划,2023YFD2400403号;山东省重点研发计划,2023TZXD052号;中国水产科学研究院基本科研业务费,2023TD53号。

摘　　要：为实现凡纳滨对虾(Litopenaeus vannamei)摄食强度量化及分类,克服投喂主观性,优化饲料利用率。实验采用YOLOv8模型对凡纳滨对虾进行识别和检测以及YOLOv8-segment模型对凡纳滨对虾的虾头进行分割,根据饲料区域内对虾的数量和虾头的像素面积,实现对虾摄食强度的量化,并利用Convnext模型将图像中对虾的摄食强度划分为强摄食、中摄食和低摄食3个级别。实验结果显示,迭代200次后,YOLOv8模型对饲料区域内对虾目标检测的mAP50达到了99.5%,YOLOv8-segment模型对虾头分割的mAP50达到了92.1%,展现了YOLOv8模型的强大性能。经实验验证发现,在不同摄食强度下饲料区域内对虾的数量和虾头的像素面积存在明显差异。Convnext模型对凡纳滨对虾3种摄食强度的图像分类准确率为98.8%。该方法可以客观有效地将对虾摄食强度进行量化,并将对虾的摄食强度进行分类,为实现对虾的精准投喂提供了理论基础和技术支撑。To quantify and classify the feeding intensity of Litopenaeus vannamei and overcome the subjectivity of feeding,and optimize the feed utilization rate.The experiment used the YOLOv8 model to identify and detect L.vannamei and the YOLOv8-segment model to segment the shrimp heads.Based on the number of shrimp in the feeding area and the pixel area of the shrimp head,the feeding intensity of the shrimp was quantified.The feeding intensity of the shrimp was then classified using the Convnext model,dividing the images into strong feeding,moderate feeding,and low feeding categories.The experimental results showed that after 200 iterations,the mAP50 of the YOLOv8 model for detecting shrimp targets in the feeding area reached 99.5%,and the mAP50 of the YOLOv8-segment model for segmenting the shrimp head reached 92.1%,which demonstrated the strong performance of the YOLOv8 model.The experiment verified that there were significant differences in the number of shrimp in the feeding area and the pixel area of the shrimp head under different feeding intensities.The classification accuracy of the Convnext model for the three feeding intensity categories of L.vannamei was 98.8%.This method can objectively and effectively quantify and classify the feeding intensity of L.vannamei,providing a theoretical basis and technical support for precise feeding of the shrimp.

关 键 词：凡纳滨对虾 摄食强度量化 计算机视觉 YOLOv8 

分 类 号：S968.22[农业科学—水产养殖]