BiAF: research on dynamic goat herd detection and tracking based on machine vision

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Rangeland management is shifting towards intelligent systems for resource optimization.
• Traditional livestock monitoring methods disrupt natural animal behavior.
• Accurate grazing pattern analysis is crucial for scientific grazing practices.

Purpose of the Study:

• To develop a non-disruptive machine vision algorithm for livestock grazing monitoring.
• To enhance target detection and tracking accuracy for goat herds.
• To provide an information-driven approach for achieving grass-animal balance.

Main Methods:

• Proposed BiAF-YOLOv7 algorithm, enhancing YOLOv7-tiny with an optimized CBAM attention mechanism and refined SPPCSPC module.
• Improved anchor boxes in YOLOv7-tiny for enhanced target detection.
• Integrated BiAF-YOLOv7 with DeepSORT for robust goat herd tracking.

Main Results:

• BiAF-YOLOv7 achieved high performance metrics: 94.5% precision, 96.7% recall, 94.8% F1 score, and 96.0% mAP on a goat herd dataset.
• The system successfully tracked goat herds in large-scale environments.
• Demonstrated the algorithm's effectiveness and practicality for livestock monitoring.

Conclusions:

• The BiAF-YOLOv7 algorithm is a practical and effective tool for livestock grazing monitoring.
• Machine vision-based monitoring offers broader applicability in large-scale rangeland management.
• This study provides innovative methods for information-driven grass-animal balance.