Live body weight (LBW) is one of the most important parameters for supervising the growth and development of livestock. The yak (Bos grunniens) is a special species of cattle that lives on the Qinghai-Tibetan Plateau. Yaks are more untamed than regular cattle breeds, thus it is more challenging to measure their LBW. In this study, a YOLOv8 yak detection and LBW estimation models were used to automatically estimate yak LBW in real-time. First, the proper posture (normal posture) and individual yak identification was confirmed and then the YOLOv8 detection model was used for LBW estimation from 2-dimensional (2D) images. Yak LBW was estimated through yak body parameter extraction and a simple linear regression between the estimated yak LBW and the actual measured yak LBW. The results showed that the overall detection performance of yak normal yak posture was described by precision, recall, and mean Average Precision 50 (mAP50) indicators, reaching 81.8, 86.0, and 90.6%, respectively. The best yak identification results were represented by precision, recall, and mAP50 values of 97.8, 96.4, and 99.0%, respectively. The yak LBW estimation model achieved better results for the 12 mo old yaks with shorter hair with R², root mean square error (RMSE), mean absolute percentage error (MAPE), and Multiple R values of 0.96, 2.43 kg, 1.69%, and 0.98, respectively. The results demonstrate that yak LBW can be estimated and monitored in real-time using this approach. This study has the potential to be used for daily yak LBW monitoring in an unstressed manner and to save considerable labor resources for large-scale livestock farms. In the future, to reduce the limitations caused by the impacts of yak hair and light condition data sets of dairy cows and yaks of different ages will be used to improve and generalize the model.

Yak behavior is a valuable indicator of their welfare and health. Information about important statuses, including fattening, reproductive health, and diseases, can be reflected and monitored through several indicative behavior patterns. In this study, an improved YOLOv7-pose model was developed to detect six yak behavior patterns in real time using labeled yak key-point images. The model was trained using labeled key-point image data of six behavior patterns including walking, feeding, standing, lying, mounting, and eliminative behaviors collected from seventeen 18-month-old yaks for two weeks. There were another four YOLOv7-pose series models trained as comparison methods for yak behavior pattern detection. The improved YOLOv7-pose model achieved the best detection performance with precision, recall, mAP0.5, and mAP0.5:0.95 of 89.9%, 87.7%, 90.4%, and 76.7%, respectively. The limitation of this study is that the YOLOv7-pose model detected behaviors under complex conditions, such as scene variation, subtle leg postures, and different light conditions, with relatively lower precision, which impacts its detection performance. Future developments in yak behavior pattern detection will amplify the simple size of the dataset and will utilize data streams like optical and video streams for real-time yak monitoring. Additionally, the model will be deployed on edge computing devices for large-scale agricultural applications.

Behavior is an important indicator of yak heifer welfare and health status, with key behavioral patterns reflecting critical conditions, including fattening, reproduction, and disease. Computer vision-based pose estimation has become an essential technology for livestock behavior monitoring. This study developed a yak heifer pose estimation model named Multistage Feature Attention PVT-based Dynamic Yak Heifer Pose Estimation Model (MFPVT-YakHeifer) based on improved Pyramid Vision Transformer version 2 (PVT v2) and keypoint detection modeling. The proposed and compared models were trained with a novel YakPoseData set encompassing yak heifer images collected in different poses and environmental conditions. The results showed that the model achieved performance metrics of 90.41% mean average precision at an intersection over union (IoU) threshold of 0.5, 64.37% mean average precision at 0.75 IoU, 64.11% mean average precision, and 91.95% mean average recall at 0.5 IoU, all of which are higher than those of the other 4 benchmark models. Finally, the MFPVT-YakHeifer model has been deployed on edge computing device for livestock farm applications. The future work will focus on animal dataset expansion, real-time video analysis implementation, and computational efficiency optimization for edge deployment.