Real-Time Beef Cattle Body Condition Scoring Using EdgeBCS-YOLO: A Lightweight Framework for Edge Deployment.

Area of Science:

• Precision Livestock Farming
• Computer Vision
• Artificial Intelligence

Background:

• Accurate body condition scoring (BCS) is crucial for cattle health and productivity in precision livestock farming.
• Manual BCS is subjective and labor-intensive, while existing automated methods are often too computationally intensive for edge deployment.
• There is a need for efficient, real-time automated BCS solutions suitable for on-farm edge devices.

Purpose of the Study:

• To develop a lightweight object detection framework for real-time beef cattle BCS in unstructured farming environments.
• To optimize the model for edge deployment, balancing accuracy with computational efficiency.
• To evaluate the performance of the proposed framework on a dynamic blurring dataset and an edge computing platform.

Main Methods:

• The study proposes EdgeBCS-YOLO, a lightweight framework built upon YOLO11n.
• Key innovations include Position-Sensitive Feature Fusion (PSFF), a Texture-Aware Star Module (TASM), an Efficient Grouped Detection Head (EGDH), and Focal and Global Knowledge Distillation (FGD).
• The model was evaluated for accuracy (precision, recall, mAP@50) and edge performance (model size, FPS, inference latency).

Main Results:

• EdgeBCS-YOLO achieved 90.8% precision, 82.7% recall, and 88.9% mAP@50 on a dynamic blurring dataset.
• On the NVIDIA Jetson Orin NX Super, the model demonstrated a small size (3.95 MB), high system FPS (33.35), and low average inference latency (13.26 ms).
• These metrics indicate high accuracy and efficiency for real-time edge deployment.

Conclusions:

• EdgeBCS-YOLO presents a practical and efficient solution for automated beef cattle body condition scoring on edge devices.
• The lightweight design and strong performance metrics make it suitable for real-world precision livestock farming applications.
• This framework addresses the limitations of manual scoring and computationally demanding automated methods.