YBEDI: A lightweight and efficient multi-scale feature fusion gender classification model for modern agriculture

Precise gender determination in day-old chicks has emerged as a critical requirement in modern poultry operations, directly impacting nutritional management, resource allocation, and overall production economics. Conventional sexing methodologies ranging from invasive reproductive tract examinations to costly genetic analyses present significant limitations due to their reliance on specialized expertise, expensive equipment, and vulnerability to operator fatigue. Inspired by advancements in neural network technologies, this research developed YBEDI (YOLOv11-based Basic Efficient Dynamic Inverted-residual system), a novel deep learning framework that automates the chick sexing process through advanced computer vision techniques. YBEDI integrates YOLOv11-n detection module with an innovative BEDI (Basic Efficient Dynamic Inverted-residual classifier) for gender identification. BEDI extracts hierarchical features through multiple BasicBlock modules and processes them via Efficient Channel Attention, Dynamic Convolution, and Inverted Residual MLP through an integrated multi-scale fusion strategy. The lightweight architecture totaling only 6.24M parameters enables deployment on standard computers with limited computational resources while maintaining real-time processing capabilities. Experimental validation demonstrates 92.5% classification accuracy across diverse operating conditions, offering a practical solution that significantly reduces dependency on specialized expertise while enhancing reliability in commercial poultry operations. Code is available at https://github.com/Keypants/YBEDI.git