A framework for automatic analysis of sheep ruminating behavior based on keypoint detection

Ruminating behaviour is a critical indicator of sheep health and welfare, with abnormal rumination strongly associated with gastric ailments in sheep. Timely monitoring and analysing ruminating behaviour are imperative for assessing sheep health status and optimizing feed composition. However, deploying existing ruminant monitoring algorithms on sheep farms presents challenges due to the hypermobility of the sheep’s head and the complexity of farm environments. This study proposes a real-time analysis framework based on the RTMPose keypoint detection method for sheep rumination behaviour. The framework employs keypoint detection to precisely analyze various aspects of sheep rumination, including the number of ruminating chews, ruminating time, chewing rate, ruminating rate, and ruminating time rate. Additionally, we constructed a dataset for sheep rumination behaviour to train the framework’s object detection and keypoint detection algorithms. Experiments were conducted using fifteen challenging videos.Results demonstrate that the ruminating behaviour analysis framework based on key points achieves acceptable with 94.56% accuracy in the number of ruminating chews, 97.49% accuracy in ruminating time, and 97.99% accuracy in chewing rate, with 72.9M parameters. The proposed method can be flexibly deployed on handheld monitoring devices, presenting a viable approach for automatic identification and analysis of rumination behaviour using computer vision technology.