文章摘要

文章信息/Info

Title:: Research of feature extraction and diagnosis algorithm using principal component analysis and linear discriminant analysis of electrocardiogram signal

作者:: 李鸿强1; 2△; 魏小清1; 2; 王有玺1; 2; 张振3; 宫正1; 2; 吴非凡1; 2; 1. 天津市光电检测技术与系统重点实验室，天津 300387; 2. 天津工业大学电子与信息工程学院，天津 300387； 3. 天津工业大学计算机科学与软件学院，天津 300387

Author(s):: LI Hongqiang1; 2; WEI Xiaoqing1; 2; WANG Youxi1; 2; ZHANG Zhen3; GONG Zheng1; 2; WU Feifan1; 2; 1. Tianjin Key Laboratory of Optoelectronic Detection Technology and Systems, Tianjin 300387, China; 2. School of Electronics and Information Engineering, Tianjin Polytechnic University, Tianjin 300387; 3. School of Computer Science and Software, Tianjin Polytechnic University, Tianjin 300387

Keywords:: Electrocardiogram ; Feature extraction; Classification diagnosis; Principal component analysis; Linear discriminant analysis; Genetic algorithm

摘要:: 针对心脏疾病发病率高且不易自主检测的问题，提出了一种心电信号特征提取和分类诊断算法。首先对心电信号进行提升小波变换和改进半软阈值相结合的预处理变换，在去除心电信号的噪声后，利用主成分分析（principal component analysis，PCA）对心电信号进行降维，并利用核独立成分提取心电信号的非线性特征；同时离散小波变换提取去噪后心电信号的频域特征，基于线性判别分析（linear discriminant analysis, LDA）对频域统计特征进行降维处理。将两种不同的特征向量组成多域特征空间，最后利用支持向量机对多域特征空间分类，遗传算法对其参数进行寻优，从而实现心电信号特征的分类。实验结果表明，所提出的算法能够对5类心电节拍进行准确分类，分类效率达99.11%。

Abstract:: To solve the problem of high incidence of heart disease and difficulty in autonomic detection, an algorithm for feature extraction and classification diagnosis of electrocardiogram (ECG) signal was proposed. Firstly, the ECG signal was preprocessed by lifting wavelet transform and improving semi-soft threshold. After removing the noise of ECG signal, the principal component analysis was used to reduce the dimension of ECG signal features and kernel independent component analysis was used to extract the nonlinear features of ECG signals. At the same time, discrete wavelet transform was used to extract the frequency domain characteristics of the denoised ECG signal and the frequency domain features were reduced in dimension based on linear discriminant analysis. In the end, support vector machine (SVM) was used to classify the multi-domain feature space and genetic algorithm was used to optimize the parameters of SVM to realize the classification of ECG features. Experiment results show that the proposed algorithm can classify five kinds of ECG beats accurately and the classification efficiency is 99.11%.