An Integrated Goat Head Detection and Automatic Counting Method Based on Deep Learning.

Area of Science:

• Agricultural Technology
• Computer Vision
• Artificial Intelligence

Background:

• Goat farming is vital for sustainable economies and social development.
• Precision and intelligence are needed for modern goat breeding.
• Current methods lack efficiency in automated goat management.

Purpose of the Study:

• To develop an integrated deep learning-based method for accurate goat detection and counting.
• To enhance object detection and tracking algorithms for livestock monitoring.
• To provide a practical solution for intelligent goat breeding and management.

Main Methods:

• Constructed a novel dataset of goat video images for object tracking.
• Improved the YOLOv5 object detector with RandAugment, AF-FPN, and Dynamic Head.
• Integrated the enhanced detector with DeepSORT for goat tracking and counting.
• Implemented single-line counting based on goat head tracking to prevent double counting.

Main Results:

• The improved YOLOv5 detector achieved 92.19% mAP, a significant increase from the original 84.26%.
• The integrated detection and tracking system demonstrated an average overlap rate of 89.69%, surpassing the original 82.78%.
• The single-line counting method effectively minimized double counting in practical applications.

Conclusions:

• The proposed deep learning method offers a highly accurate and efficient solution for goat detection and counting.
• The enhancements to YOLOv5 and integration with DeepSORT significantly improve livestock monitoring capabilities.
• This technology supports the advancement of precision agriculture and intelligent goat farming.