Improved glove defect detection algorithm based on YOLOv5 framework

Aiming at the problems of low efficiency, poor accuracy and narrow application area of manual detection, this paper proposed a glove defect detection algorithm YOLO-G with high accuracy, low number of parameters and easy deployment. In the preparation phase, data augmentation is performed to expand the dataset, reduce the image size, and increase the data richness. The K-means algorithm is used to calculate anchors. Then Ghostnet is used to replace some structures in the YOLOv5 backbone and Neck, which effectively reduces the size of the model. The attention mechanism is added to make the model more sensitive to key information and improve the detection effect. The experimental results show that the mean average precision of YOLO-G reaches 90.4%, the number of parameters decreases by 47%, and the amount of calculation decreases by 49.4%, which is more suitable for the deployment of embedded scenarios.