AI-Based Poultry Monitoring
This project focuses on optimizing poultry feed usage while monitoring individual chicken health and growth using computer vision. It combines automated identification, egg counting, and health analysis to improve farm efficiency and reduce operational costs.
Efficient poultry management requires tracking individual birds, but traditional identification methods (tags, colors, markers) are unreliable, non-scalable, and raise ethical concerns. Recent advances in computer vision, especially visual re-identification (Re-ID), provide a promising alternative by identifying animals based on natural features without physical tagging.
To develop a scalable, non-invasive, vision-based re-identification system for poultry that enables individual tracking, early disease detection, and automated feed optimization.
Achievements / Outcomes
- Explored multiple identification methods and identified their limitations
- Built a custom dataset for chicken re-identification
- Implemented preprocessing pipeline (segmentation, background removal, augmentation)
- Tested transformer-based models (ViT, Swin) for feature extraction
- Established evaluation metrics (Top-1, Top-5, mAP)