Steel Strip Defect Classification Model Based on Convolutional Neural Network Composites

Abstract

In actual industrial scenes, the complex and diverse textures of surface flaw of steel strips lead to poor performance and generalization of the classification and recognition tasks of surface flaw of steel strips. In this paper, an end-to-end classification model of surface flaw of steel strips based on convolution neural network-EDESPNet is proposed. Improve the accuracy of steel belt defect classification. The model inputs defect samples into different branch networks for simultaneous feature extraction, and adds a weight distribution network to enhance category-related feature information, promotes model efficiency and feature spread, and rise the model's portrayal ability. The results indicate that the EDESPNet classification model is surpass VGG19, DenseNet, ResNet50, Xception and other models. The classification precision in the NEU-CLS data set is 94.17%, and the classification accuracy rate in the BS4-CLS data set is 72.52%. The EDESPNet classification model proposed in the article has higher experimental evaluation standards on the BS4-CLS data set than other classification models, and it has the highest correct rate in the NEU-CLS public datasets. The results indicate that the EDESPNet classification model has better recognition task effect and good robustness.