Optimization And Improvement Of Tangut Character Images Recognition Model

Character images recognition is an important research direction of artificial intelligence and a hot field.In recent years,scholars at home and abroad have done a lot of research on Tangut character recognition technology and achieved many results.However,due to the wide variety,diverse styles and complex structure of Tangut character,and the uneven distribution of existing Tangut character script samples,the performance of the Tangut character images recognition model has been greatly affected,so the task of Tangut character recognition is still a very challenging task.Based on the above background,this paper takes improving the accuracy of Tangut character recognition model as the starting point to study the algorithm to improve the accuracy of Tangut character recognition.The main research contents are as follows:In this paper,the experimental dataset is enhanced,and the data set is expanded by using rotation,horizontal translation,vertical translation,random tangent transformation,random zoom and other methods to improve the diversity of the data set.Subsequently,the Tangut character image recognition based on classical neural network was studied.Improvements were made to the network infrastructure of AlexNet and ResNet,respectively,to enable the recognition of Tangut character script.However,the AlexNet network is not deep,the ability to extract features is limited,and the application of ResNet34 network to the Tangut script dataset in this paper is too complex.Moreover,the number of Tangut text samples in this paper is insufficient,which is easy to reduce the performance of the recognition model.so transfer learning is used to avoid this problem,and a transfer learning network model based on VGG16 is proposed,according to the feature map of different shallow layers of the Tangut character image in the VGG16 network,we try to freeze fine-tune different network layers and train the specific features of the remaining layer for training.Finally,the influence of feature fusion on the recognition rate of VGG16 model is discussed.Experimental results show that the accuracy of the improved AlexNet and ResNet models on the test set is increased by 7%and 2%,respectively,and the accuracy is effectively improved,and the number of parameters is greatly reduced.However,the recognition accuracy of the improved network model is about 94%,and there is still a lot of room for improvement.However,when the transfer learning network model based on VGG16 freezes the first 7 layers of the network,the recognition accuracy of Tangut character images has been significantly improved compared with other groups,reaching 97%accuracy,and the training loss is controlled at about 0.31,and the recognition effect is better than that of other networks on the whole.This shows that the new structure proposed in this paper has certain feasibility and superiority in the task of Tangut character recognition.