Improved YOLOv8 metal surface defect detection model

In order to solve the problems of low detection accuracy, high miss detection rate and high false detection rate of metal surface defect detection in industrial manufacturing, an improved YOLOv8 metal surface defect detection model is proposed. Firstly, the CG module is used to replace Backbone's downsampling convolution module to enhance the model’s contextual information understanding ability; secondly, the RepGFPN module is used to replace the original feature pyramid network to improve the model's multi-scale feature extraction ability; finally, by redesigning the detection head, The GDetect module is proposed to improve the overall performance of the model. Experimental results show that the accuracy, recall and mAP@0.5 of the improved model reached 71.2%, 72.4% and 74.5% in the GC10-DET data set, which increased by 2.8%, 8.1% and 6.0% respectively. The number of parameters and calculation The volumes were reduced by 6% and 22% respectively. At the same time, the robustness and generalization ability of the model under different data sets were verified in the PASCAL VOC and NEU-DET data sets, and the detection accuracy of the target was improved. The proposed improved model has made significant progress in the field of metal defect detection, improved detection accuracy, solved common problems, and achieved high performance while maintaining lightweight, providing an efficient and feasible solution for metal surface defect detection.