An Yuning, Song Yifeng, Jiang Hehao, Wang Yuan, Liu Na, Li Xia, Zhang Zhalaga, An Xiaoping
College of Animal Science, Inner Mongolia Agricultural University, Hohhot 010018, China.
National Center of Technology Innovation for Dairy-Breeding and Production Research Subcenter, Hohhot 010018, China.
Animals (Basel). 2025 Apr 23;15(9):1201. doi: 10.3390/ani15091201.
This study introduces a non-contact, single-target method for real-time monitoring of dairy cow calving posture and behavior using the YOLOv8 model. In total, 600 videos were collected, from which 10,544 image samples were extracted through frame-by-frame processing. Complete video recordings of 86 cows (30 primiparous and 56 multiparous) were utilized to investigate changes in calving behavior. The YOLOv8 model achieved excellent performance with precision (), recall (), and mean average precision (m) of 96.72%, 96.53%, and 97.41%, respectively, and recognition of 89.19% for lying postures and 82.61% for standing postures. Behavioral analysis revealed that lying postures were more frequent than standing, and primiparous cows had more frequent posture transitions (9.07 changes) than multiparous cows (5.29 changes), particularly during early parturition. Primiparous cows also showed significantly higher average times for parturition and lying as well ashigher frequency of behavioral changes compared to multiparous cows. Additionally, calf birth weight was positively correlated with maternal behaviors, especially in primiparous cows. Our proposed model effectively and accurately recognizes calving postures in dairy cows, enabling the early detection of abnormal calving events. This provides a scientific basis and technical support for intelligent farm management.
本研究介绍了一种使用YOLOv8模型对奶牛产犊姿势和行为进行实时监测的非接触式单目标方法。总共收集了600个视频，通过逐帧处理从中提取了10544个图像样本。利用86头奶牛（30头初产奶牛和56头经产奶牛）的完整视频记录来研究产犊行为的变化。YOLOv8模型表现出色，精确率（）、召回率（）和平均精度均值（mAP）分别为96.72%、96.53%和97.41%，对躺卧姿势的识别率为89.19%，对站立姿势的识别率为82.61%。行为分析表明，躺卧姿势比站立姿势更频繁，初产奶牛的姿势转换（9.07次变化）比经产奶牛（5.29次变化）更频繁，尤其是在分娩早期。与经产奶牛相比，初产奶牛的平均分娩时间和躺卧时间也显著更长，行为变化频率更高。此外，犊牛出生体重与母体行为呈正相关，尤其是在初产奶牛中。我们提出的模型能够有效且准确地识别奶牛的产犊姿势，从而能够早期发现异常产犊事件。这为智能农场管理提供了科学依据和技术支持。
