YOLO-BCD: A Lightweight Multi-Module Fusion Network for Real-Time Sheep Pose Estimation.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Animal Science

Background:

• Precision livestock management requires real-time animal behavior analysis.
• Current systems face challenges balancing computational efficiency and detection accuracy.
• Automated ovine posture recognition is crucial for health monitoring.

Purpose of the Study:

• To develop an optimized deep learning framework (YOLOv8-BCD) for efficient and accurate ovine posture recognition.
• To improve detection performance in complex farm environments.
• To provide a computationally efficient solution for resource-constrained devices.

Main Methods:

• Developed YOLOv8-BCD, a lightweight deep learning architecture for ovine posture recognition.
• Incorporated enhanced feature fusion and spatial-channel attention modules.
• Introduced adaptive multi-scale feature aggregation, context-aware attention weighting, and streamlined detection head optimization.

Main Results:

• Achieved 91.7% recognition accuracy with 389 FPS processing speed.
• Reduced model parameters by 19.2% and computational load by 32.1% compared to standard YOLOv8.
• Demonstrated significant improvements over baseline models in practical farming conditions.

Conclusions:

• YOLOv8-BCD offers an efficient and accurate solution for real-time ovine posture recognition.
• The framework provides technical support for automated health monitoring in intensive livestock production.
• Shows practical potential for large-scale agricultural applications requiring behavioral analysis.