基于贝叶斯分类器的脉象自动识别方法_王慧燕.pdfVIP

28 5 Vol. 28 No. 5 2009 10 Chinese Journal of Biomedical Engineering October 2009 1* 2 王慧燕 徐 珊 1 ( , 310018) 2 ( , 310053) : , , , , , , , , SMOTE Tomek links , , ; , , , : , 90% , : ; ; Automatic Pulse Recognition Method Based on Bayesian Classifier WANG H iYan1* XU Shan2 1 ( College of Comp uter Science Inf ormation Engineering , Zhej iang Gongshang Univ ersity , H ang hou 310018) 2 ( College of Basic Medicine , Zhej iang Chinese Medicine University , H ang hou 310053) Abstract : P lse diagnosis is one of the most important examinations in traditional Chinese medicine (TCM) . D e to the s bjectivity and f zziness of p lse diagnosis in TCM, q antitative methods are demanded. In view of the complexity of p lse signals and the nonlinear relationships between p lse parameters and p lse types, a new p lse signal recognition method was proposed and p lse q antitative diagnostic models were b ilt. First, the characteristic parameters of p lse signals were comp ted and a database containing characteristic parameters of p lse signals and p lse types was b ilt. Second, the dataset was balanced sing synthetic minority oversampling techniq e ( SMOTE) and Tomek links method, which made different p lse types have approximately eq al sample n mber. Third, a Bayesian network str ct re was learned from the balanced database and the Markov blanket of p lse type in the str ct re was selected as feat re set and taken as inp t of a Bayesian classifier. Fo rth, the mapping relationships between p lse signals and p lse