Accurate Recognition Of Oracle Based On Deep Convolutional Neural Network

It is of great significance to identify and interpret the inscriptions on bones and tortoise shells for the inheritance of Chinese culture,because oracle records the earliest history and culture of Chinese civilization.In the age of big data in the digitization of ancient Chinese characters,the digitization of oracle materials provides a guarantee for the analysis and research of big data of oracle.The application of deep learning method in computer vision has occupied an absolutely dominant position in recent years,and the convolutional neural network which open up new research ideas for the analyzing oracle data has made great achievement in image recognition,target detection and other tasks.Therefore,starting from the recognition of oracle bones,this paper employs convolutional neural network to study the key issues of accurate detection as well as recognition of oracle,and data mining and semantic analysis are used for obtaining new progress in the research work of oracle bones.Aiming at the problems of difficulty in extracting the target features of oracle characters,small sample and diversity,the deep convolutional neural network and data enhancement method are used to accurately identify the oracle in this thesis.The research contents are as follows:(1)It's the first time to adopt the deep convolutional neural network framework which has a very powerful image feature expression and learning ability for oracle recognition.Deep convolutional neural network can learn to fit the objective function model by a large number of data training so as to solve the problem of image classification.Application of oracle data to realize the training of convolutional neural network and the character representation of oracle is studied,which based on the network model of different depth and architecture.At the same time,training and learning to achieve oracle high precision recognition by the way of sufficiently abstracting the hierarchical features of oracle through the depth feature network progressive structure.(2)To improve the recognition performance of the network model,the oracle feature graph which has specific attributes is extracted as prior knowledge by combining with the traditionalfeature extraction method.Gabor feature,gradient feature and Hog feature are extracted as the prior knowledge in this paper,and the corresponding features are mapped to the input layer to further enhance the representation ability of the network model and improve the recognition accuracy.(3)Data enhancement technology is used to solve the problem of Small sample of oracle.In order to solve the problem of the scarcity of oracle samples,data enhancement methods such as Image mirror flip and rotation are used to expand the data sample.Also,oracle are accurately identified based on the convolutional neural network model in this paper,and the traditional image feature extraction and data expansion method are combined to improve the robustness and efficiency of the network model.