Automatic detection of bumblefoot in cage-free hens using computer vision technologies

Area of Science:

• Animal Science
• Veterinary Medicine
• Artificial Intelligence

Background:

• Cage-free (CF) housing is becoming dominant in egg production, increasing the incidence of bumblefoot in hens.
• Bumblefoot causes pain and welfare issues, hindering hen mobility and access to resources.
• Current methods for detecting bumblefoot are challenging, especially in early stages, lacking automated solutions.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) deep learning models for automated bumblefoot detection in hens within CF environments.
• To compare the performance of different YOLOv5 models (YOLOv5s-BFD, YOLOv5m-BFD, YOLOv5x-BFD) for bumblefoot detection.
• To identify optimal training parameters (epochs, batch size, camera height) for effective bumblefoot detection.

Main Methods:

• Development and testing of three deep learning models: YOLOv5s-BFD, YOLOv5m-BFD, and YOLOv5x-BFD.
• Training and validation of models using datasets from hens in CF environments under various settings.
• Performance evaluation based on precision, recall, mAP@0.50, mAP@0.50:0.95, and F1-score.

Main Results:

• The YOLOv5m-BFD model demonstrated superior performance across key metrics.
• YOLOv5m-BFD achieved the highest precision (93.7%), recall (84.6%), mAP@0.50 (90.9%), and F1-score (89.0%).
• Optimal performance for YOLOv5m-BFD was observed with 400 epochs and a batch size of 16.

Conclusions:

• The YOLOv5m-BFD model is recommended for accurate and automated bumblefoot detection in laying hens in CF systems.
• This study provides a foundation for implementing automated bumblefoot detection systems in commercial poultry operations.
• Future work will focus on adapting the model for detecting bumblefoot in broilers.