ABSTRACT

In recent years, with the increasing demand for high-precision clothing image classification, people has attached great importance on the clothing image classification method based on CNN. In this paper, we discussed how to improve the accuracy of image classification by preprocessing the data set, combining the data set with CNN structure and improving the loss function. The comparative experiments between different structures of CNN and loss functions have been done. The experimental results show that the method we used can better improve the effect of classification and classification accuracy.

Keywords: - CNN, Image classification, Image feature, Loss function