A Novel Lightweight Framework for Non-Contact Broiler Face Identification in Intensive Farming.

Area of Science:

• Agricultural Technology
• Computer Vision
• Animal Science

Background:

• Precision broiler farming requires efficient individual identification for effective management and welfare monitoring.
• Traditional and existing CNN-based methods face limitations in dense farming environments, particularly concerning occlusion and computational efficiency.

Purpose of the Study:

• To develop a high-precision, lightweight face recognition framework (YOLO-IFSC) tailored for dense broiler farming.
• To improve multi-scale feature extraction, reduce computational redundancy, enhance spatial feature aggregation, and emphasize critical facial regions.

Main Methods:

• The study builds upon the YOLOv11n architecture, integrating four specialized modules: Inception-F for multi-scale features, C2f-Faster for computational efficiency, SPPELANF for occlusion handling, and CBAM for attention.
• A self-constructed dataset was utilized for experimental evaluations.

Main Results:

• YOLO-IFSC achieved a mean Average Precision (mAP@0.5) of 91.5%.
• The framework demonstrated a 40.8% reduction in parameters and a 24.2% reduction in Floating Point Operations (FLOPs) compared to the baseline.
• Real-time inference speed was maintained at 36.6 Frames Per Second (FPS).

Conclusions:

• YOLO-IFSC provides a cost-effective, non-contact solution for broiler face recognition in precision farming.
• The proposed framework significantly advances individual tracking and welfare monitoring capabilities.