Robust Automated Monitoring of Dairy Cow Rumination via Improved YOLOv11 and BoT-SORT in Complex Environments

Accurate, non-contact monitoring of rumination behavior is essential for assessing dairy cow health and welfare, as well as for optimizing feeding strategies and herd management in modern precision livestock farming. However, practical deployment in commercial barns faces challenges such as occlusions, variable lighting, and dynamic cow movements. To address this, we developed a robust, automated vision-based framework for continuous rumination monitoring. The core of our system integrates an enhanced object detection algorithm with a robust tracking module, specifically improved to capture subtle behavioral features and maintain identity under complex conditions. Evaluated on a comprehensive dataset collected from commercial settings under various lighting and occlusion scenarios, our framework achieved high detection accuracy (mAP of 96.26%) and reliable tracking performance (multi-object tracking accuracy of 99.2%). This demonstrates its suitability for real-time, on-farm deployment. The study provides a practical, end-to-end solution for fine-grained behavioral analysis in complex environments, offering a tool that can enhance welfare assessment and support decision-making in dairy farm management. The methodological approach is also adaptable to other precision livestock monitoring tasks.