Surface Defect Recognition Classification Based on Multi-Scale Convolutional Neural Network

Abstract

In order to improve the quality of industrial products, an image recognition model based on multi-scale convolutional neural network is proposed for different scale targets of surface defect detection. First of all, the traditional convolutional neural network structure was replaced with a three-layer full connection layer by a two-layer convolutional layer. After processing, the convolutional neural network was used as the skeleton network to deeply extract the features of defect images. Secondly, pooling is used to reduce dimension of feature images to obtain feature images of different scales. Then the defect detection of different sizes is realized by fusing feature maps of different scales. Experimental results on DAGM2007 data set show that this method can realize surface defect detection and the detection accuracy is optimized compared with other defect detection methods.