Design and Research of Metal Surface Defect Detection Based on Machine Vision

Abstract

To address the problems of low accuracy and inaccurate classification of surface defects detection that occur on metallic steel, this paper proposes a method to improve the accuracy of surface defect detection by adjusting the network structure of the YOLOv3 algorithm model[1]. First, the k-means++ algorithm is used for clustering to improve the matching of prior frames of different scales and feature layers by increasing the scale difference of prior frames. Secondly, the algorithm improves the recognition rate of small defective targets by adding 104×104 feature layers. Finally, the spatial pyramid pooling module is added to improve the recognition accuracy of target features after extracting feature layers of different scales from the backbone feature network. The experimental results show that the improved YOLOv3 algorithm model achieves an average accuracy of 76% on the test set and is 9% better than the original YOLOv3 algorithm than Faster-R-CNN1 in terms of detection performance.