Monitoring Cattle Ruminating Behavior Based on an Improved Keypoint Detection Model

Area of Science:

• Animal Science
• Computer Vision
• Agricultural Technology

Background:

• Cattle rumination is a key health indicator.
• Current monitoring methods (manual observation, wearables) have limitations like labor intensity and animal harm.
• A non-contact, automated solution is needed for efficient cattle health monitoring.

Purpose of the Study:

• To develop and validate a non-contact method for monitoring cattle rumination behavior.
• To automatically identify chewing counts, duration, and frequency using computer vision.
• To provide a technical framework for precision management in animal husbandry.

Main Methods:

• Utilized an improved YOLOv8-pose keypoint detection algorithm for cattle pose estimation.
• Constructed rumination motion curves from keypoint data.
• Applied multi-condition threshold peak detection to count chewing instances.
• Developed a comprehensive framework for analyzing rumination indicators.

Main Results:

• The modified YOLOv8-pose achieved 96% mAP, with improved precision (4.5%) and recall (4.2%).
• The system demonstrated an average chewing count error of 5.6% and a standard error of 2.23% compared to actual data.
• The method accurately captured keypoint information and analyzed rumination behavior.

Conclusions:

• The proposed non-contact keypoint detection method is feasible and effective for analyzing cattle rumination.
• This technology enables quicker detection of rumination abnormalities, supporting informed management decisions.
• The framework offers technical support for precision livestock management and modern farming practices.