Research On Chinese Painting Classification Method Based On Deep Learning

With the country’s prosperity and development,the people’s living standards have improved significantly.While paying attention to material life,they have also begun to cultivate their inner spiritual world.As one of the treasures of traditional Chinese culture,Chinese painting is appreciated and collected by more and more people.Therefore,the trend of digitization of Chinese painting is becoming stronger and stronger,and the management needs of the Chinese painting digital art gallery make the classification and recognition technology of Chinese painting images become the key to urgent problems.After analysis,it is found that Chinese painting processing technology faces two difficulties.One is that because of the "semantic gap" in Chinese painting images,it is difficult to achieve good classification results only by extracting the global underlying features.Therefore,it is necessary to obtain significant and useful features as the classification basis.The second is that Chinese painters mostly focus on subjective artistic conception when painting,and the styles created by different authors of the same subject are quite different.Therefore,it is difficult to classify Chinese paintings based on content.The content of the article mainly explores the semantic classification algorithm of Chinese painting images based on deep learning.The main work is as follows.1.In order to capture the overall style and local brushstroke characteristics of Chinese painting images at the same time,a parallel hybrid model based on the combination of convolutional neural network(CNN)and long short-term memory network(LSTM)is proposed to realize the classification of traditional Chinese painting images.CNN network is designed to obtain the overall style features of Chinese painting images,and then the Chinese painting images are divided into blocks,and LSTM network pair is designed.These areas are processed to obtain the dependency between the areas to obtain the stroke characteristics.Finally,a feature fusion strategy is designed,that is,an adaptive weighted fusion layer is designed in the proposed hybrid model,and the above two types of features are adaptively weighted and fused through a learnable method,which is used as the final feature representation of the Chinese painting image.Enter the Softmax classification layer to achieve classification.In order to verify the effectiveness of the proposed model,comparative experiments were conducted on the self-built Chinese Painting Data Set(CP).The experimental results show that the proposed model can achieve 96% classification accuracy,which is better than other algorithms.2.In order to solve the problem of weak supervision of Chinese painting images,a multi-instance learning(MIL)algorithm based on LSTM is proposed to realize the classification of traditional Chinese painting images.In order to realize the classification of Chinese painting images under the MIL framework,first,a multi-instance modeling scheme is designed,that is,the Chinese painting image is divided into blocks by using the Pyramid Overlapping Grid method to convert it into multi-instance bags.Secondly,a sequence generator is designed,that is,using the nature of multi-instance learning to define a "specificity" criterion function,and use it to select some discriminative bags from the positive bag to construct a discriminative instance set,and then,convert multi-instance bags into equal-length and ordered signal sequences.Third,a multi-layer LSTM network model with an attention mechanism is designed,and signal sequences are input to the model,and semantic analysis is performed on it to obtain memory coding features.They are used as the final feature representation of the corresponding Chinese painting image.Finally,the Softmax classification layer in the model is used to realize the classification and prediction of the Chinese painting image.The innovations in the work are follows.The algorithm introduces the attention mechanism into the multi-layer LSTM network model,that is,by calculating the matching degree between the input sequence and the output vector of the current layer of the network,different weights are assigned to the input of the higher layer LSTM to fully Mining the degree of influence of different instances on the classification of Chinese paintings.For the basic needs of Chinese painting image classification,on the basis of in-depth understanding of the characteristics of Chinese painting image classification,LSTM is introduced into the multi-instance learning method.Compared with other MIL algorithms based on bag embedding,LSTM can better capture the interdependence between instances.Comparative experiment results show that it is feasible to introduce LSTM network into MIL,and the performance of the proposed MIL algorithm is better than other classification algorithms in the classification of traditional Chinese painting images.