Research On Recognition Of Bottom-mark Images Of Blue And White Porcelains In Ming And Qing Dynasties

For thousands of years,blue and white porcelain has always been the focus of the times.Entering modern society,this passion for the ancient blue and white ceramic becomes more intensified,but people's knowledge of ancient ceramics,traditionally,is mainly to observe and judge through the human eye,which is based on past experience.With the coming of recent large data and artificial intelligence technology era,now we get the image which is only need by ordinary fixed or mobile terminal,and can give some identification of ancient ceramics,which will greatly improve the public awareness and appreciation of the ability about ancient ceramics.Firstly,this dissertation introduces some background knowledge of ancient Ming and Qing dynasties,especially describe the basic characteristics of blue and white porcelain in the Ming and Qing dynasties.Secondly,we review the basic knowledge of convolution neural network,and then give the fundmental framework of convolution neural network about the image of blue and white porcelain.Lastly,we describe the initial set of experiments and experimental steps of the image recognition experiment.From the view of specific experimental means,this dissertation use the VS / Opencv platform to deal with a large number of blue and white porcelain image rapidly at the beginning,and establish the bottem-mark training set after the image segmentation.In order to prevent over-fitting,the experiment uses a large number of data enhancement to improve the generalization ability of the sample model,and then uses the Tensorflow / Keras GPU framework in the linux / python environment to carry out training such as the original CNN and other four programs After the training,it is found that CNN as the core of the algorithm program has provided with high robustness and good classification ability.Finally,the improved algorithm of bottom-mark image training in nonsegmentation is discussed in this paper.The experimental results show that the accuracy of the blue and white porcelain which based on convolution neural network recognition system is over 99 percent.The main innovations of this dissertation are:1.The application of depth learning theory is the first study about ancient ceramic identification analysis;2.Test can provide user-oriented multi-terminal services by Tensorflow and other tools for the ancient ceramic appreciation in the industrial applications.