Chinese Painting Image Classification Based On Deep Learning

Painting is an important art form in the course of human history.China's long history has produced a large number of paintings,which have important significance for people to understand China's history and culture.With the development of information technology,image classification has become one of the hot research topics in the image field,and the classification of Chinese painting images is also emerging.Traditional image classification methods are mostly based on shallow structure learning algorithms.Although various image features can be extracted,some features may be lost in some feature extraction processes,and feature selection requires certain painting knowledge,and feature extraction method is poorly generalized.This determines that the traditional method of classification of Chinese painting is less versatile,and the deep learning method that has arisen in recent years does not require manual interference of image feature selection.Based on this,the main researches on the classification of Chinese paintings are asfollows:In this paper,the superiority of convolutional neural network in the classification of painting images is firstly proposed.Based on the convolutional neural network,the effect of activation function on image classification accuracy is further studied.A linear and nonlinear combination function is proposed.The activation function,which combines the advantages of SoftSign and ReLU activation functions,constructs a new activation function SReLU,which is applied to the convolutional neural network and experiments on Chinese painting datasets and Dunhuang mural data.The experimental results show that the activation function proposed in this paper has improved the classification effect.Secondly,it expounds the concept of small sample learning and its main methods,and then proposes a relational network model based on convolutional neural network.This model simulates the process of human recognition of objects,trying to make convolutional neural networks learn how to learn and compare,mainly image classification tasks are achieved by comparing feature differences between images and images.Although there are many kinds of Chinese paintings,there is no professional data set available for Chinese paintings,which means that Chinese paintings do not have a large number of marked data sets.Only the convolutional neural network training to achieve the classification of new tasks requires manual marking of a large amount of sample data,so small sample of learning is necessary to realize the recognition of new painting image classes.In addition,Dunhuang murals are a wonderful work in Chinese painting.Therefore,in addition to verifying the feasibility of the network model for the classification of Chinese paintings in the new category,this paper also validates the Dunhuang mural dataset.The experimental results demonstrate the feasibility of small sample learning.