Tongue Image Color Classification Based On Ensemble Learning In Traditional Chinese Medicine

The color classification of tongue images based on machine learning is an important part of the modernization of TCM tongue diagnosis.Sorting the color of the tongue is a key step to achieve the objective diagnosis of TCM tongue diagnosis,and lays a foundation for the follow-up TCM automatic diagnosis.In the current literature on automatic classification of tongue color,it is generally carried out using a single classifier.In order to further improve the accuracy of tongue color classification,this paperThe integrated learning method is used to study the classification of tongue color.The main work of this thesis is as follows.1.Determine the color characteristics of the tongue image.This paper compares the RGB,Lab,and HSV color models,and finally selects Lab and HSV as the color features of the tongue image.For the tongue image,the moss color often affects the judgment of the color of the tongue.In this paper,the k-means method is used to isolate the tongue image.In order to facilitate the extraction and judgment of the subsequent tongue color.Then,the segmented tongue image is manually selected by the sample sub-block to increase the total sample size.In terms of the selection of classification methods,integrated learning combines the advantages of multiple classifiers,which often yields superior performance over a single classifier.Therefore,this paper chooses the integrated learning method as the basic research method of this paper,and then through the experimental comparison of the representative methods in the integrated learning method,the representative algorithm Adaboost method in the boosting method is selected as the main research method.2.DataBoost-IM combined with GE-SMOTE is proposed to deal with the imbalance of tongue image data.There is a large difference in the number of samples between the tongue image samples,and the samples in each category are not balanced.In response to this problem,this paper selects the three different data set processing methods and selects the GE-SMOTE method for data set preprocessing.Then,the classification method of the unbalanced data set combined with the DataBoost-IM method and the GE-SMOTE method is designed to deal with the imbalance of the tongue image sample set.Experiments have shown that higher recognition accuracy is obtained for small sample categories compared to using the GE-SMOTE method to process data sets and then using Adaboost classification.3.The Adaboost cascading framework classifier with auxiliary decision function is used to further improve the accuracy of tongue color classification.This paper designs and implements the improved Adaboost cascading framework classifier,which is based on the Adaboost classifier,combined with the cascading framework,and then added by the auxiliary judgment function.The experimental results show that this method achieves higher recognition accuracy,and the time complexity of the algorithm is lower.