基于EMD关联维数和多重分形谱的心音识别  

Heart sound recognition based on EMD correlation dimension and multi-fractals pectrum

作　　者：郭兴明[1] 张文英[1] 袁志会[1] 何彦青[1] 李传鹏[1] 

机构地区：[1]重庆大学生物工程学院生物流变科学与技术教育部重点实验室,重庆400044

出　　处：《仪器仪表学报》2014年第4期827-833,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(30770551);重庆市新型医疗器械重大专项(CSTC;2008AC5103)资助项目

摘　　要：心音是一种高度非线性、非平稳性的振动信号,传统的线性分析方法不足以揭示其内在特征,提出了一种基于EMD关联维数和多重分形谱的心音分类识别方法。首先通过EMD将心音信号分解成若干个IMF,利用互相关系数准则对其进行筛选以得到主IMF分量,并利用G-P算法求主IMF分量的关联维数;然后分析多重分形谱的4个特征参数,其中多重分形谱宽度的差异性最大,将它结合主IMF分量的关联维数,作为二叉树支持向量机的输入向量实现对临床采集的正常心音和5类异常心音信号的分类识别。结果表明,该方法能有效地提取心音特征,同时提高了心音信号的识别率。Heart sound is a kind of highly nonstationary and nonlinear vibration signal,traditional linear analysis methods can not fully reveal its internal characteristics.In this paper,a new heart sound classification recognition method based on empirical mode decomposition （EMD） correlation dimension and multi-fractal spectrum is proposed.Firstly,heart sounds are decomposed into a finite number of intrinsic mode functions （IMFs） using EMD.The main IMF components are selected using the criteria of mutual correlation coefficient,and the G-P algorithm is used to calculate the correlation dimension of the main IMF components.Secondly,four characteristic parameters of multi-fractal spectrum are analyzed,in which the multi-fractal spectrum width has the largest divergence.Finally,the correlation dimension of the main IMF components and multi-fractal spectrum width as the eigenvectors are sent into the binary tree support vector machine （BT-SVM） for the classification and recognition of normal heart sounds and five kinds of pathological heart sound signals acquired from clinic.The clinical data test results show that the proposed approach not only can effectively extract the heart sound characteristics,but also improve the recognition rate of heart sound signals.

关 键 词：心音 经验模式分解 关联维数 多重分形谱 

分 类 号：R318[医药卫生—生物医学工程] TH77[医药卫生—基础医学]