A Deep Learning Model for Detecting Cage-Free Hens on the Litter Floor

Area of Science:

• Computer Vision
• Artificial Intelligence
• Animal Science

Background:

• Precision poultry farming relies on accurate hen detection for management and welfare.
• Monitoring chickens in cage-free systems presents unique challenges compared to caged environments.
• Existing methods may lack the real-time accuracy needed for dynamic cage-free settings.

Purpose of the Study:

• To develop and evaluate a deep learning model for real-time hen detection in cage-free facilities.
• To assess the model's performance across various conditions including age, lighting, and observation angles.
• To establish a foundation for machine vision systems in commercial cage-free poultry houses.

Main Methods:

• A deep learning model, YOLOv5x-hens, based on the YOLOv5 convolutional neural network (CNN) was developed.
• The model was trained using over 1000 images and tested on 200 images of hens in cage-free settings.
• Statistical analyses (One-way ANOVA, Tukey HSD) were used to compare predicted and actual hen counts (p < 0.05).

Main Results:

• The YOLOv5x-hens model achieved high evaluation metrics: Precision (0.96), Recall (0.96), F1 (0.96), and mAP@0.5 (0.95) for detecting hens on litter floors.
• Stable detection performance (over 95% accuracy) was observed for hens aged 8-16 weeks under varying light and angles.
• Detection accuracy was lower (25%) for younger chicks due to equipment interference, and limitations included occlusion and uneven lighting.

Conclusions:

• The YOLOv5x-hens model demonstrates effective real-time hen detection in cage-free environments, exceeding 95% overall accuracy.
• This AI-driven approach is crucial for tracking individual birds, enabling better evaluation of production and welfare.
• Future research should address challenges like overlapping birds and occlusion to further enhance detection capabilities.