A dynamic individual method for yak heifer live body weight estimation using the YOLOv8 network and body parameter...

Area of Science:

• Livestock management
• Artificial intelligence in agriculture
• Animal science

Background:

• Live body weight (LBW) is crucial for livestock growth monitoring.
• Yaks (Bos grunniens) present unique challenges for LBW measurement due to their wild nature.
• Accurate and non-invasive LBW estimation is needed for efficient yak farming.

Purpose of the Study:

• To develop and validate an automated system for real-time yak LBW estimation.
• To utilize YOLOv8 for accurate yak detection and identification.
• To establish a correlation between image-derived body parameters and actual LBW.

Main Methods:

• Employing YOLOv8 for yak detection and identification from 2D images.
• Extracting yak body parameters for LBW estimation.
• Utilizing simple linear regression to correlate estimated and actual LBW.
• Evaluating model performance using precision, recall, mAP50, R², RMSE, and MAPE.

Main Results:

• YOLOv8 achieved high detection performance (81.8% precision, 86.0% recall, 90.6% mAP50) for yaks in normal posture.
• Yak identification accuracy was excellent (97.8% precision, 96.4% recall, 99.0% mAP50).
• The LBW estimation model showed strong results for 12-month-old yaks (R²=0.96, RMSE=2.43 kg, MAPE=1.69%).

Conclusions:

• The developed YOLOv8-based system enables real-time, non-invasive yak LBW estimation.
• This technology can significantly reduce labor and stress in large-scale yak farming.
• Future work will focus on improving model generalization across different conditions and breeds.