Deep Learning Methods for Automatic Identification of Male and Female Chickens in a Cage-Free Flock

Area of Science:

• Animal Science
• Computer Science
• Artificial Intelligence

Background:

• Rooster selection and monitoring are crucial for egg fertility and hatchability in poultry breeding.
• Traditional methods for identifying and evaluating roosters can be labor-intensive and subjective.
• Phenotypic characteristics like comb and body size are key indicators of rooster desirability.

Purpose of the Study:

• To develop and evaluate deep learning object detection models for identifying roosters versus hens.
• To compare the performance of different You Only Look Once (YOLO) model variants for chicken sex identification.
• To assess the effectiveness of using comb size and body size for automated rooster detection.

Main Methods:

• Trained You Only Look Once (YOLO) models, including YOLOv5 and YOLOv11 variants, using a dataset of over 2500 chicken images.
• Focused on phenotypic traits such as comb size and body size for distinguishing between male and female chickens.
• Statistically compared model performance metrics (precision, recall, mAP, F1 score) using one-way ANOVA.

Main Results:

• Object detection models accurately identified roosters based on comb and body size in a cage-free setting.
• YOLOv5xu and YOLOv11m models showed superior performance for rooster detection using body size, achieving high precision and recall.
• Specific YOLO variants demonstrated strong capabilities in differentiating roosters based on comb size as well.

Conclusions:

• Deep learning models, particularly YOLO variants, provide an effective tool for automated rooster identification and monitoring.
• This technology can aid in the performance evaluation and genetic selection of roosters in poultry breeder farms.
• The study establishes a foundation for future research into tracking rooster activity and optimizing breeding programs.