基于支持向量机的痤疮患者舌色苔色识别算法研究  

An algorithm study on tongue color recognition of patients with acne based on support vector machine

作　　者：张艺凡[1] 胡广芹[1] 张新峰[1] 

机构地区：[1]北京工业大学,北京100124

出　　处：《北京生物医学工程》2016年第1期7-11,共5页Beijing Biomedical Engineering

基　　金：973计划(2011CB505406)资助

摘　　要：目的针对中医痤疮患者的舌象特征,基于支持向量机(support vector machine,SVM)对舌色苔色的识别准确率及识别速度进行研究,以提高中医舌诊客观化研究的准确度及速度。方法针对专家标定的1500个典型舌样本,在苔质识别阶段将亮斑、舌苔、舌质同层识别,以避免亮斑被误识别为白苔。使用交叉验证核函数、一对一(one against one,OAO)和投票法训练多类SVM分类器,网格法、遗传法、粒子群法结合交叉验证法对RBF核函数参数寻优,得到最佳分类结果。最后与阈值法、聚类法及DAG方法进行了识别准确率及识别速度的比较。结果与其他方法比较,本文方法在舌色苔色识别准确率及识别速度上均有提高,本文识别准确率为8473%,高于DAG方法(8104%);识别速度为05599s,优于DAG方法(16394s)。结论本文方法对舌色苔色的分类准确率和识别速度都有一定的提高,对辅助医生临床诊疗及临床研究具有现实意义。ObjectiveAccording to the acne tongue features of traditional Chinese medicine,a schemebased on support vector machine（ SVM） for recognition accuracy and recognition speed of tongue color andmoss color were studied,in order to improve the accuracy and speed in objectification of tongue diagnosis intraditional Chinese medicine research.MethodsIn the first stage,the paper proposed tongue color,moss colorand highlight area peer recognition avoiding highlight area being mistakenly identified as white moss. Crossvalidation kernel function,one against one（ OAO） and voting method were used to train multi-class SVMclassifier. Respectively, GridSearch, GA and PSO algorithm combined with cross validation for parameteroptimization of SVM method,the best classification results were obtained. Finally,the proposed method withthreshold method, clustering method and the DAG method was compared the recognition accuracy andrecognition speed.ResultsThis method in recognition accuracy and recognition speed of was higher than othermethods. In this paper, the identification accuracy was 84 73%, higher than the DAG method which was81 04%. Recognition speed was 0 5599 s, better than that of DAG method as 1 6394 s.ConclusionsTherecognition accuracy and recognition speed for the tongue color and moss color of this method increased greatly,which was important to assist clinical diagnosis,treatment and research.

关 键 词：痤疮 舌诊 支撑向量机 参数寻优 交叉验证 

分 类 号：R318.04[医药卫生—生物医学工程]