Research On The Application Of Support Vector Machine （SVM） In Ancient Ceramic Classification

The production and use of ceramic ware have demonstrated the economic and cultural progress of ancient China.Thousands of years of ceramic history have witnessed the cultural heritage of ancient ceramics and expressed the wisdom of labor and scientific principles.The appraisal of Potter and porcelain includes visual inspection and scientific and technological appraisal.With the development of machine learning and data mining technology in the work of ancient ceramic dating,researchers have gradually constructed and improved the theoretical system of appraisal of pottery and porcelain.Based on support vector machine（SVM）and transductive support vector machine（TSVM）,constructs classification models of ancient ceramics under supervised learning and semi-supervised learning.In the work of supervised SVM classification model for ancient ceramics,four SVM multi classification models were constructed with Linear,Poly,RBF and sigmoid as kernel functions respectively.Combining the square root function of arcsinh and slightly non curvilinear property,named m-arcsinh kernel function is generated.Different weights are given to arcsinh and square root functions respectively to meet the needs of SVM for different scale classification tasks.The classification performance of the five kernel functions is evaluated by eight indicators.Thai is,accuracy,average precision,average recall,average F1 score,run times and the independent precision value,recall value and F1 score value of the three types of ancient ceramic samples from the Yuan,Ming and Qing Dynasties.Then,seven classification datasets of different scales are selected from UCI database to carry out the comparative experiment of five kernel functions to verify the classification performance of m-arcsinh kernel function.The experimental results show that m-arcsinh kernel has fast computing power and high accuracy.As a kernel function of SVM,it can better solve the task of ancient ceramic classification.For the construction of semi supervised TSVM-DT optimized combination ancient ceramic classification model,the structure,weight distribution and diversity calculation method of TSVM-DT semi supervised classification model are explained.Then,the same training set and test set are used to train and test TSVM and DT models respectively to adjust the unlabeled samples.Calculate the diversity coefficient between TSVM-DT and DT semi supervised classifiers.Train TSVM-DT semi supervised classification model,and analyze the influence of edge density and unlabeled sample proportion on classification accuracy value.Finally,the feasibility and reliability of the TSVM-DT semi supervised classification model are further verified by 7 datasets in UCI database,and the accuracy value and running time of the experiment are evaluated.The experimental results show that TSVM-DT semi supervised classification model has significantly improved the classification accuracy in ancient ceramic samples and multiple classification works,and has better feasibility and adaptability...