Improved YOLOv3 Method for Detecting Surface Defects in Cold Rolled Steel Strip

Surface defects of cold rolled strip steel is an important influence on the quality of strip steel products, however, cold rolled strip steel is more prone to various surface defects due to the long production process, thin specifications and other factors, so the detection of these defects is the key to the production of high-quality products. So far, although many researches have proposed relevant algorithms, they cannot well meet the requirements of speed and accuracy, therefore, we have improved on the basis of YOLOv3, so that it can achieve the optimal effect in the detection of surface defects of cold rolled strip steel. Briefly, we introduced SENet in the feature map extraction stage to improve the feature extraction ability of the model, automatically learn the relevance and importance of the features of each layer in the channel dimension, and realize the adaptive fusion between different features; and replace the K-Means clustering algorithm with K-Means++ clustering algorithm. The experimental results show that the improved model proposed in this study has better performance.