Development of a Tool to Detect Open-Mouthed Respiration in Caged Broilers.

Area of Science:

• Animal Science
• Computer Vision
• Agricultural Engineering

Background:

• Open-mouth panting in broiler chickens signals heat stress and welfare concerns.
• Detecting panting in crowded conditions is difficult due to small target size, occlusions, and background clutter.

Purpose of the Study:

• To develop an enhanced object detection method for accurately identifying broiler chicken panting.
• To improve the monitoring of heat stress and welfare in broiler chickens.

Main Methods:

• Utilized a lightweight YOLOv8n framework with a P2 detection head for small target recognition.
• Incorporated Space-to-Depth grouped convolution (SGConv) for fine-grained feature extraction.
• Employed a Bidirectional Feature Pyramid Network (BIFPN) and Squeeze-and-Excitation (SE) channel attention for enhanced feature representation and noise suppression.

Main Results:

• Achieved 0.92 mAP@50 on an independent test set and 0.927 mAP@50 with leakage-free retraining.
• Demonstrated significant performance improvement over baseline YOLO models.
• Confirmed strong generalizability and real-time performance for practical application.

Conclusions:

• The enhanced YOLOv8n method offers a practical solution for intelligent broiler welfare monitoring.
• This technology aids in effective heat stress management and improved animal well-being.
• The dual validation of generalizability ensures reliability in diverse farming environments.