智能化农业装备学报（中英文）2025，Vol.6Issue(3)：88-97,10.DOI:10.12398/j.issn.2096-7217.2025.03.009
基于YOLOX和DeepSORT的淡水鱼行为检测算法研究
Research on freshwater fish behavior detection algorithm based on YOLOX and DeepSORT
摘要
Abstract
Freshwater fish behavior recognition is an important part of ecological research,which helps to understand the ecological habits and environmental adaptability of fish.In order to recognize freshwater fish behavior,this paper proposes a freshwater fish behavior detection algorithm based on YOLOX and DeepSORT algorithms.The fish target detection dataset in Open Image Dataset V7 was selected as dataset 1,and the freshwater fish target detection model was trained by YOLOX.The manually labeled and image enhanced dataset 2 was retrained by transfer learning,and the freshwater fish target detection model was completed.Based on target detection,the appearance features were obtained,and DeepSORT was used to build a freshwater fish target tracking algorithm to obtain the fish's position information,speed and acceleration.By analyzing fish movement information,rules are set to classify fish activity,mortality,rushing,and other common behaviors.The results show that transfer learning and image enhancement reduce the sample size required for model training,and the freshwater fish target detection model has a good accuracy of 83%.DeepSORT algorithm has a good effect on freshwater fish target tracking results,and can accurately extract freshwater fish motion information with a processing speed of 10 f/s and good real-time performance.MOTA,MOTP and IDF1 reach 83.582%,96.245%and 94.105%,respectively,with good tracking performance.Based on the motion information,the freshwater fish behavior was defined according to the preset rules,and the random forest method was used to classify the freshwater fish.The accuracy of the random forest method reached 99.72%.The freshwater fish behavior detection algorithm proposed in this study based on YOLOX and DeepSORT not only ensures high detection accuracy,but also realizes good real-time performance,which can effectively identify the behavior pattern of freshwater fish,and provides strong technical support for ecological research.关键词
YOLO/DeepSORT/淡水鱼/水产养殖/行为检测Key words
YOLO/DeepSORT/freshwater fish/aquaculture/behavior detection分类
农业科技引用本文复制引用
张哲鼎,王超柱,陈骏,OKINDA Cedric,刘龙申..基于YOLOX和DeepSORT的淡水鱼行为检测算法研究[J].智能化农业装备学报（中英文）,2025,6(3):88-97,10.基金项目
江苏省现代农机装备与技术示范推广项目(NJ2022-34) Jiangsu Province Modern Agricultural Machinery Equipment and Technology Demonstration Project(NJ2022-34) （NJ2022-34）
