Efficient Cow Body Condition Scoring Using BCS-YOLO: A Lightweight, Knowledge Distillation-Based Method.

Area of Science:

• Animal Science
• Computer Vision
• Artificial Intelligence

Background:

• Traditional dairy cow body condition scoring (BCS) relies on subjective, labor-intensive methods unsuitable for large farms.
• Automated BCS is needed for consistent, efficient monitoring of dairy herd health and productivity.

Purpose of the Study:

• To develop BCS-YOLO, a lightweight, automated framework for accurate dairy cow BCS using YOLOv8.
• To enhance detection accuracy and reduce model complexity for resource-limited farm environments.

Main Methods:

• BCS-YOLO integrates the Star-EMA module with multi-scale attention for optimized feature representation.
• A Star Shared Lightweight Detection Head (SSLDH) simplifies the model for efficient deployment.
• Channel-based knowledge distillation focuses on key body regions for improved performance.

Main Results:

• BCS-YOLO achieved a 33% reduction in model size compared to baseline models.
• The framework demonstrated a 9.4% improvement in mean average precision (mAP).
• The system provides consistent and accurate BCS under complex farm conditions.

Conclusions:

• BCS-YOLO presents a robust, non-invasive solution for automated dairy cow BCS.
• This technology supports sustainable livestock management, reduces labor, and improves animal welfare and productivity.