Lightweight Crown Cap Defect Detection Algorithm Based on improved YOLOv10n

To address the problem of high miss detection rate and large computational volume of tiny defects in crown caps, this paper proposes an improved YOLOv10n lightweight crown cap tiny defect detection algorithm named SCPS-YOLOv10n, which realizes a marked reduction of the parameter count while improving the accuracy. The improved algorithm adopts the StarNet backbone network to replace the YOLOv10n-backbone, which significantly cuts down the parameter count, renders the backbone network lightweight and boosts the model's portability; The C2f module is replaced with the C2f_WTConv module, expanding the model's receptive field for capturing input features and improving its ability to perceive low-frequency information. Applies the PUPC module instead of the SPPF, and uses a shared convolutional layer to improve the computational efficiency and reduce redundant parameters. The lightweight detection header SSP-Detect is added to the model, and the parameter sharing method facilitates the model in reducing the parameter count without degrading the accuracy. On a crown cap dataset, the improved model is tested and attains an [email protected] of 95.4%, marking a 2% improvement over the baseline YOLOv10n model. The parameter count and computational effort have been cut down by 40% and 45%, respectively, compared with YOLOv10n.