Research on Surface Defect Detection Algorithm of Strip Steel Based on Improved YOLOV3

Abstract

Abstract Aiming at the problems of slow detection speed and low detection accuracy in the surface defect detection of hot rolled strip steel, an improved YOLOV3 algorithm model was proposed in this paper. Firstly, the data sets is expanded by using the data enhancement algorithm to improve the robustness of the algorithm. Secondly, K-means++ with less randomness was used to perform clustering analysis on defect labels instead of K-means algorithm, and appropriate Anchor Boxes were selected as the initial candidate boxes for the improved YOLOV3 network. Finally, this study changed the detection network of YOLOV3, added a layer of prediction box and performed feature fusion to improve the detection ability of the network for small targets. The experimental results show that the improved YOLOV3 algorithm achieves 89.5% mean average precision on NEU-DET data set, which is 14.7% higher than the original YOLOV3 algorithm. Detection speed in 34 FPS at the same time, can meet the needs of industrial production.