UST-YOLO11Pose-TRM: An Attention-Enhanced Keypoint Detection and Transformer Regression Framework for Yak Body...

Area of Science:

• Livestock Science
• Computer Vision
• Artificial Intelligence

Background:

• Yak (Bos grunniens) body measurements are crucial for livestock assessment.
• Traditional manual measurements are inefficient, inaccurate, and stressful for animals.

Purpose of the Study:

• To develop an intelligent, non-contact method for yak body measurement.
• To improve the efficiency and accuracy of yak growth and health monitoring.

Main Methods:

• Integrated keypoint detection (UST-YOLO11Pose) with regression modeling (Transformer-based).
• Incorporated attention mechanisms (UIB, SENetV2, TripleAttention) for enhanced keypoint detection.
• Utilized multi-head self-attention for robust regression of body parameters.

Main Results:

• UST-YOLO11Pose achieved high accuracy (mAP 0.958) with a lightweight model (10.06 MB).
• The Transformer regression model demonstrated excellent predictive accuracy (RMSE 0.185, R² 0.962).
• The system provides accurate, efficient, and non-contact yak body measurements.

Conclusions:

• UST-YOLO11Pose-TRM offers a significant advancement over traditional methods.
• The technology has strong potential for smart pasture development and precision livestock management.
• This intelligent system supports improved yak breeding and health monitoring.