基于小波分析的心电信号分类研究

发布时间：2018-04-04 14:36

本文选题：心电信号　切入点：小波变换　出处：《中南大学》2014年硕士论文

【摘要】：心律失常在人们的生活中比较寻常,心律失常的发生可能会对人的生命造成影响,因此,为了预防心律失常的发生,我们必须更加准确和及时的对心律失常进行检测。近年来,利用计算机对心律失常进行自动处理已成趋势,但是由于心电信号的电流微小,受到的干扰较多,同时,由于个人的区别和心律失常分类的规则不统一等缘故,计算机对心脏的诊断至今难以满足医院的需求。本文针对这一状况,在前人的基础上对常见的六种心律失常进行识别,主要的工作如下：心电信号预处理：使用小波变换的分解重构法去除信号中的基线漂移,使用小波变换的阀值法去除高频肌电干扰和工频干扰,本文结合小波变换的分解重构法和小波变换的阀值法既可以消除信号中的主要噪声干扰,又可以避免有用成分丢失。通过美国麻省理工学院和贝丝以色列医院(Massachusettes Institute of Technology and Beth Israel Hospital, MIT-BIH)心律失常数据库中的数据仿真可知,达到了较好的效果。心电信号QRS波群的检测：针对目前的Mexican-hat小波检测法,由于心律失常的影响,容易造成低幅R波漏检,同时由于高大P波、T波和高频噪声的影响,容易造成R波误检,因此,本文采用连续小波变换和多种策略的方法来检测QRS波群,通过MIT-BIH数据库仿真,检测正确率达99.50%。心电特征参数的提取与选择：由于传统的特征提取只考虑时域特征,其具有一定片面性,不足以反映心电信号的本质,为了更加准确的反映心电信号的本质特征,本文综合采用时域特征和小波域特征作为心电信号的特征向量,时域上提取了RR1、RR2、QRS波宽和心率变异性(Heart rate variability, HRV)四个特征,小波域上提取了第四层尺度信号,第四层小波信号和第三层小波信号,然后对特征向量进行了优化,为后续分类奠定了良好的基础。心律失常的分类：设计一种支持向量机(Library for Support Vector Machines, LIB SVM)分类器,对优化后的特征向量进行训练和测试,然后对MIT-BIH心律失常数据库中的六种典型的心律失常类型进行分类,分类的整体正确率达到96.60%以上。
[Abstract]:Arrhythmia is more common in people's life, the occurrence of arrhythmia may affect human life. Therefore, in order to prevent the occurrence of arrhythmia, we must more accurately and timely detection of arrhythmia.In recent years, it has become a trend to use computer to process arrhythmia automatically, but because of the small current of ECG signal, the disturbance is more, at the same time, because of the difference of individual and the rule of arrhythmia classification, etc.The diagnosis of the heart by computer is still difficult to meet the needs of the hospital.In this paper, six common arrhythmias are identified on the basis of previous studies. The main work is as follows:ECG signal preprocessing: wavelet transform decomposition reconstruction method is used to remove baseline drift, wavelet transform threshold method is used to remove high frequency EMG interference and power frequency interference.In this paper, the decomposition reconstruction method of wavelet transform and the threshold value method of wavelet transform can not only eliminate the main noise interference in the signal, but also avoid the loss of useful components.Through the data simulation of Massachusetts Institute of Technology and Beth Israel Hospital (MIT-BIH) arrhythmia database of Massachusetts Institute of Technology and Beth Israel Hospital, the results are satisfactory.Detection of QRS wave group in ECG signals: for the current Mexican-hat wavelet detection method, it is easy to cause low amplitude R wave miss detection due to arrhythmia, at the same time, because of the influence of high P wave T wave and high frequency noise, it is easy to cause R wave false detection.In this paper, continuous wavelet transform and multiple strategies are used to detect QRS wave groups, and the correct detection rate is 99.50 by MIT-BIH database simulation.Extraction and selection of ECG feature parameters: because traditional feature extraction only considers time domain feature, it has certain one-sidedness, which is not enough to reflect the essence of ECG signal, in order to reflect the essential characteristics of ECG signal more accurately.In this paper, the time domain feature and the wavelet domain feature are used as the feature vectors of ECG signal. The RR1 / RR2 QRS wave width and heart rate variability (HRV) are extracted from the time domain, and the fourth layer scale signal is extracted in the wavelet domain.The fourth layer wavelet signal and the third layer wavelet signal are optimized, which lays a good foundation for the subsequent classification.Classification of arrhythmias: a support vector machine library for Support Vector machines (LIB SVM) classifier is designed to train and test the optimized feature vectors, and then classify six typical arrhythmia types in MIT-BIH arrhythmia database.The overall correct rate of classification is above 96.60%.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7;O174.2

【参考文献】