Study of a QueryPNet Model for Accurate Detection and Segmentation of Goose Body Edge Contours

Area of Science:

• Computer Vision
• Precision Animal Husbandry
• Agricultural Technology

Background:

• Expanding scale of goose farming necessitates higher efficiency and reduced human influence.
• Real-time automated monitoring is crucial for precision animal husbandry.
• Instance segmentation enables accurate detection, counting, and analysis of individual geese.

Purpose of the Study:

• To develop an automated system for real-time monitoring of goose flocks.
• To improve the efficiency and accuracy of goose detection and counting using computer vision.
• To reduce breeding costs through enhanced traceability and management.

Main Methods:

• Training of the QueryPNet model for goose flock segmentation and extraction.
• Proposal of a novel neck module to enhance feature pyramid structure for effective feature fusion.
• Optimization of model parameters for reduced complexity and improved performance.

Main Results:

• The QueryPNet model achieved high accuracy in target detection and instance segmentation.
• Accuracies of 0.963 (mAP@0.5) for both target detection and instance segmentation were recorded.
• The system demonstrated effectiveness even with challenges like vegetation and litter occlusion.

Conclusions:

• The developed computer vision system, QueryPNet with its novel neck module, significantly enhances precision goose farming.
• The model provides accurate real-time monitoring, detection, and counting of individual geese.
• This technology offers a viable solution for improving efficiency, traceability, and cost-effectiveness in large-scale goose farming.