基于AR模型参数谱估计的先心病特征提取及分类识别研究
发布时间：2018-11-06 09:18
【摘要】：先天性心脏病已经给人类的生活与健康带来了很大的影响,并且其发病率一直居高不下,据不完全统计,全球每年大约新增150万先天性心脏病患者,并且有将近六成的患者得不到及时的治疗而引起病情延误甚至死亡,但是要是及时发现和诊治,多数患者都能够恢复正常。目前临床上的各种诊断方法都存在一定的缺陷,而心音信号能够表现出人体状态变化和心脏病变的病理信息,所以可以使用数字信号处理方法对信号展开分析,给临床诊断结果提供参考。 本文将以MATLAB为平台从信号预处理、特征分析、提取与识别这三个方面来对心音信号开展分析研究。 预处理包含了降噪与分段定位。本文首先采用了小波阈值消噪算法消除噪声,并且比较了这个算法中选择不同的阈值和小波基对降噪效果的影响。然后采用不借助参考信号的定位方法实现心音信号的定位分段,首先使用归一化香农能量法提取心音信号的包络,然后再结合信号自身的特性准确找出S1与S2的准确位置。 特征分析、提取包含了信号时频分析和功率谱估计。应用了STFT、Choi-Williams分布这两种时频分析法来对比正常和异常人的心音信号之间的差别,然后采用了基于AR模型的参数谱估计法从时间、频率和强度三个角度来对心音信号进行特征提取。 提取到的特征使用SVM方法进行识别诊断,对比了RBF、多项式函数和Sigmod函数构建的分类器对识别效果的影响,通过实验发现,RBF构建的分类器的分类准确率最高,可以达到76.7%。
[Abstract]:Congenital heart disease has had a great impact on human life and health, and the incidence of congenital heart disease has been high. According to incomplete statistics, there are about 1.5 million new congenital heart disease patients in the world every year. And nearly 60% of the patients can not get timely treatment, causing delay or even death, but if timely detection and treatment, most patients can return to normal. At present, all kinds of clinical diagnosis methods have certain defects, and the heart sound signal can show the pathological information of human body state change and heart disease change, so the digital signal processing method can be used to analyze the signal. To provide reference for clinical diagnosis. In this paper, we use MATLAB as the platform to analyze and study the heart sound signal from three aspects: signal preprocessing, feature analysis, extraction and recognition. The preprocessing includes noise reduction and segmental location. In this paper, the wavelet threshold denoising algorithm is first used to eliminate noise, and the effects of different thresholds and wavelet bases on the noise reduction are compared. Then the method of locating the heart sound signal without reference signal is used to realize the location segmentation of the heart sound signal. Firstly, the envelope of the heart sound signal is extracted by using the normalized Shannon energy method, and then the accurate position of S1 and S2 is found accurately according to the characteristics of the signal itself. Feature analysis, extraction includes signal time frequency analysis and power spectrum estimation. Two time-frequency analysis methods of STFT,Choi-Williams distribution are applied to compare the difference between normal and abnormal heart sound signals, and then the parameter spectrum estimation method based on AR model is used to estimate the time of heart sounds. The heart sound signal is extracted from three angles of frequency and intensity. The extracted features are identified and diagnosed by SVM method. The effect of the classifier constructed by RBF, polynomial function and Sigmod function on the recognition effect is compared. The experimental results show that the classifier constructed by RBF has the highest classification accuracy. It can reach 76. 7.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R541.1;TN911.7
本文编号：2313864
[Abstract]:Congenital heart disease has had a great impact on human life and health, and the incidence of congenital heart disease has been high. According to incomplete statistics, there are about 1.5 million new congenital heart disease patients in the world every year. And nearly 60% of the patients can not get timely treatment, causing delay or even death, but if timely detection and treatment, most patients can return to normal. At present, all kinds of clinical diagnosis methods have certain defects, and the heart sound signal can show the pathological information of human body state change and heart disease change, so the digital signal processing method can be used to analyze the signal. To provide reference for clinical diagnosis. In this paper, we use MATLAB as the platform to analyze and study the heart sound signal from three aspects: signal preprocessing, feature analysis, extraction and recognition. The preprocessing includes noise reduction and segmental location. In this paper, the wavelet threshold denoising algorithm is first used to eliminate noise, and the effects of different thresholds and wavelet bases on the noise reduction are compared. Then the method of locating the heart sound signal without reference signal is used to realize the location segmentation of the heart sound signal. Firstly, the envelope of the heart sound signal is extracted by using the normalized Shannon energy method, and then the accurate position of S1 and S2 is found accurately according to the characteristics of the signal itself. Feature analysis, extraction includes signal time frequency analysis and power spectrum estimation. Two time-frequency analysis methods of STFT,Choi-Williams distribution are applied to compare the difference between normal and abnormal heart sound signals, and then the parameter spectrum estimation method based on AR model is used to estimate the time of heart sounds. The heart sound signal is extracted from three angles of frequency and intensity. The extracted features are identified and diagnosed by SVM method. The effect of the classifier constructed by RBF, polynomial function and Sigmod function on the recognition effect is compared. The experimental results show that the classifier constructed by RBF has the highest classification accuracy. It can reach 76. 7.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R541.1;TN911.7
【参考文献】
相关期刊论文 前6条
1 朱冰莲,杨磊;心音信号的短时傅立叶变换分析[J];重庆大学学报(自然科学版);2004年08期
2 陈天华;韩力群;邢素霞;郭培源;;基于小波变换的心音信号滤波方法研究[J];计算机仿真;2010年12期
3 闵芳;孙芙蓉;;基于噪声特性阀值的语音去噪[J];现代计算机;2006年01期
4 粱天昱;;运动对心脏机能及心肌能量代谢的影响综述[J];中国校外教育;2010年10期
5 张国华;袁中凡;李彬彬;;心音信号特征提取小波包算法研究[J];振动与冲击;2008年07期
6 周静;杨永明;何为;;心音信号的分析及其特征提取方法的研究[J];中国生物医学工程学报;2005年06期
相关博士学位论文 前1条
1 宋雪雷;基于小波变换和支持向量机的电能质量扰动分析方法[D];哈尔滨工业大学;2007年
，本文编号：2313864
本文链接：https://www.wllwen.com/yixuelunwen/xxg/2313864.html
最近更新
教材专著
