基于YOLO算法的牛体姿态估计研究

Cattle Posture Estimation Based on YOLO Algorithm

牛体姿态估计在分析牛行为和评价牛健康状况方面具有至关重要的作用。针对目前无法对牛进行全天候监控并及时地获得相应行为信息的问题,采用YOLO算法对牛体姿态进行研究,建立牛体姿态估计模型,用于识别牛并提取牛体骨架结构。测试结果表明,YOLOv8n-pose牛体姿态估计模型与YOLOv7-w6-pose相比,精确度、召回率、L_(oks)=0.50时的平均精确度均值(mAP_(0.50))分别提升4.7%、3.0%、2.7%,模型参数量、计算量分别降低59.2%、91.0%,单张图像平均检测时间减少5.74 ms。YOLOv8牛体姿态估计模型具有较高的检测精度和检测速度,可为大规模畜牧业中的牛体姿态估计提供可靠且有价值的参考。

Cattle posture estimation plays a crucial role in analyzing cattle behavior and evaluating cattle health.To address the problem that cattle cannot be monitored around the clock and corresponding behavioral information cannot be obtained in a timely manner,the YOLO algorithm is used to study cattle posture and establish a cattle posture estimation model for identifying cattle and extracting their skeleton structure.The test results show that compared with YOLOv7-w6-pose,the YOLOv8n-pose cattle posture estimation model has increased accuracy,recall rate,and mean average accuracy at Loks=0.50(mAP0.50)by 4.7%,3.0%,and 2.7%,respectively.The model parameter and computational complexity have decreased by 59.2%and 91.0%,respectively,and the average detection time per single image has decreased by 5.74 ms.The YOLOv8 cattle posture estimation model has high accuracy and inference speed,providing a reliable and valuable reference for cattle posture estimation in large-scale animal husbandry.