SAG-YOLO: A Lightweight Real-Time One-Day-Old Chick Gender Detection Method

Area of Science:

• Animal Science
• Computer Vision
• Machine Learning

Background:

• Feather sexing is a manual method for chick sex identification based on feather growth rate.
• Automation is needed to improve the efficiency and scalability of chick sex identification.

Purpose of the Study:

• To develop an improved SAG-YOLO method for automated chick sex detection.
• To enhance the accuracy and efficiency of chick sex identification in the poultry industry.

Main Methods:

• Utilized StarNet lightweight Backbone for reduced parameter size and computational complexity.
• Incorporated Additive Convolutional Gated Linear Unit (Additive CGLU) module in the Neck for enhanced multi-scale feature interaction.
• Implemented Group Normalization Head (GN Head) to decrease parameters and computational overhead.

Main Results:

• Achieved 90.5% precision, 90.7% recall, and 97.0% mean average precision (mAP), outperforming YOLO v10n.
• Reduced model parameters by 0.8633 M and 2.0 GFLOPs, with a 0.2 ms faster GPU inference speed.
• Demonstrated 100% accuracy for female chicks and 96.25% for male chicks in video stream detection.

Conclusions:

• The improved SAG-YOLO method offers a practical and efficient solution for intelligent sex sorting of day-old chicks.
• The model exhibits robust generalization capabilities, handling challenges like motion blur and feature fuzziness.