Detecting wing fractures in chickens using deep learning, photographs and computed tomography scanning

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Animal Welfare Science

Background:

• Current poultry slaughterhouse animal welfare monitoring relies on subjective visual inspection by officers.
• Limitations of human inspection include production line speed, subjectivity, and fatigue, necessitating objective methods.
• Wing fractures and bruises are key indicators of acute pain and suffering in poultry.

Purpose of the Study:

• To evaluate the efficacy of deep learning models for detecting poultry wing fractures and bruises.
• To assess the applicability of computed tomography (CT) scans and photographs for automated welfare assessment.

Main Methods:

• Developed and trained three deep learning models: Model_CT for fractures using CT scans, Model_Photo_Fractures for fractures using photographs, and Model_Photo_Bruises for bruises using photographs.
• Utilized 3D ResNet34 architecture for CT-based model and 2D EfficientNetV2_s for photo-based models.
• Collected 306 CT scans and 285 photographs for model training, validation, and testing.

Main Results:

• Model_CT achieved 98% accuracy in detecting fractures from CT scans.
• Model_Photo_Fractures reached 96% accuracy in identifying fractures from photographs.
• Model_Photo_Bruises demonstrated 82% accuracy in detecting bruises from photographs.

Conclusions:

• Deep learning models, utilizing CT scans and photographs, offer an objective approach to identifying wing fractures and bruises.
• This automated detection can significantly improve the accuracy and objectivity of animal welfare monitoring in poultry.
• Implementation of these technologies holds the potential to elevate overall poultry welfare standards.