FDD-YOLO: an enhanced model for high-precision feather defect detection in badminton manufacturing

Abstract

Natural feathers are fundamental to the aerodynamic stability and structural integrity of shuttlecocks, yet their quality assessment remains a pervasive manufacturing bottleneck. Conventional manual inspection is labor-intensive and susceptible to subjective error, while existing automated systems frequently fail to resolve complex surface anomalies such as insect-induced damage, discolorations, and irregular fractures. Here, we propose FDD-YOLO, a high-precision defect detection framework utilizing a re-engineered YOLOv11 architecture optimized for industrial deployment. To enhance feature extraction and diagnostic accuracy, we integrated three pivotal refinements: Spatial and Channel Reconstruction Convolution (SCConv) to mitigate informational redundancy, Deformable Convolution v4 (DCNv4) to adaptively capture irregular geometries through variable receptive fields, and the parameter-free SimAM attention mechanism to prioritize salient feature representations without increasing the parameter count. Evaluated on a specialized feather defect dataset, FDD-YOLO achieved a mean Average Precision (mAP@0.5) of 66.4%, significantly outperforming contemporary benchmarks, including YOLOv8, YOLOv10 and YOLOv11. This framework provides a robust foundation for intelligent quality control, enabling the transition toward high-precision automated manufacturing in the sporting goods sector.