基于改进OpenPose算法的猪只行为识别方法  

Pig behavior recognition method based on improved OpenPose algorithm

作　　者：李光昌 刘飞飞 李嘉豪 LI Guangchang;LIU Feifei;LI Jiahao(School of Mechanical and Electrical Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China;School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou 341000, China)

机构地区：[1]江西理工大学机电工程学院,江西赣州341000 [2]江西理工大学电气工程与自动化学院,江西赣州341000

出　　处：《河南农业大学学报》2022年第3期460-470,共11页Journal of Henan Agricultural University

基　　金：国家自然科学基金项目(61364014);江西省重点创新研发平台计划(20181BCD40009)。

摘　　要：【目的】提高猪只行为识别效率,提供猪只健康状态的快速判别依据,推进中国生猪养殖智能化和规模化。【方法】借鉴OpenPose人体姿态估计算法并对其改进,构建猪只姿态估计模型,在视频流图像中提取猪只骨骼关节点,通过计算关节点间距与骨骼关节角度描述猪只行为特征,利用K-最近邻算法(K-nearest neighbor,Knn)对猪只行为进行分类。【结果】采用基于改进OpenPose算法对猪只行为的识别准确率达到94%以上,优于采用YOLO v4算法的识别结果;与采用DeepCut、Associative Embedding和DeeperCut姿态估计算法相比,采用改进OpenPose算法构建的猪只姿态估计模型对猪只各类行为识别准确率提高了4%以上。【结论】本识别方法能够满足生猪规模化养殖中猪只行为自动化监控的需求,有效降低工人劳动强度,为猪只健康异常状态的判别提供辅助决策信息,在养殖业智能化和信息化领域具有巨大应用潜力。【Objective】In order to improve the efficiency of pig behavior recognition and provide the basis for rapid identification of pig health status,and promote the intelligent and large-scale pig breeding in China.【Method】A pig pose estimation model was constructed by referring to and improving OpenPose human pose estimation algorithm,and the pig bone joints were extracted from video stream images,and the behavior characteristics of pigs were described by calculating the distance between joints and bone joint angles.K-nearest Neighbor(Knn)algorithm was used to classify pig behaviors.【Result】The research results showed that the recognition accuracy of pig behavior based on improved OpenPose algorithm was more than 94%,which was better than that of YOLO V4 algorithm.Compared with the attitude estimation algorithms of DeepCut,Associative Embedding and DeeperCut,the pose estimation model constructed by the improved OpenPose algorithm improved the recognition accuracy of various behavior of pigs by more than 4%.【Conclusion】This method can meet the needs of automatic monitoring of pig behaviors in large-scale pig breeding,effectively reduce the labor intensity of workers,and provide auxiliary decision-making information for the identification of pig abnormal health status.It has great application potential in the field of intelligent and information breeding in the future.

关 键 词：猪只行为 姿态估计 OpenPose算法 KNN算法 计算机视觉 

分 类 号：S828[农业科学—畜牧学] TP391.41[农业科学—畜牧兽医]