Research on image classification algorithm based on depth residuals shrinkage network in Commercial Image Library

Abstract

With the advent of the age of multimedia technology and image data, commercial image library services have emerged. We need to use image classification to parse into content that the computer can understand. But when recognizing image features for classification, image data will inevitably produce noise, so in order to make image classification more accurate, we need to remove information that is irrelevant to noise. The Deep Residual Shrinkage Network (DRSN) proposed in this paper can effectively denoise images, and then learn features to classify images. The Deep Residual Shrinkage Network (DRSN) combines the Deep Residual Network (ResNet) and the SENet commonly used in the attention mechanism, and adds a soft threshold function to it. The image data set used in this experiment was downloaded from Panorama Network and Baidu Pictures and then processed. In the experiment, noise was first manually added to the data set, and the deep residual network and the deep residual shrinkage network were used for comparison experiments. It is concluded through experiments that the improved DRSN network has higher accuracy when applied to image classification.