基于魏格纳分布的心杂音信号时频能量谱分析及分类研究

发布时间：2018-04-04 21:41

本文选题：心杂音　切入点：魏格纳变换　出处：《重庆大学》2012年硕士论文

【摘要】：心音信号是人体重要的生理信号之一，心音听诊不仅简单无创且能较早的发现异常。由于心杂音信号的复杂性和非平稳性，，采用现代数字信号处理方法对心杂音信号进行分析和识别成为了解心血管状态不可或缺的手段。本文以临床辅助诊断需求为目的对心杂音信号进行分析研究，内容涉及心杂音形成机制分析、心杂音信号特征向量的提取、正常心音和4类病理性杂音信号（主动脉瓣狭窄、主动脉瓣反流、二尖瓣反流、肺动脉瓣狭窄）分类识别研究。本课题的研究工作主要包括以下几个方面： 1）正常心音和杂音信号的特征分析，介绍了正常心音和杂音形成的生理机制和时域波形特点，并对心音和杂音信号进行AR模型功率谱估计，为特征向量的选取提供一定的理论依据和参考。 2）心杂音信号多成分分离算法的实现。采用基于主成分分析的奇异谱方法，对心杂音信号进行奇异值分解和重构，达到正常心音成分和杂音成分分离的效果，结果表明该方法能有效的抑制信号魏格纳变换产生的交叉项干扰。 3）心杂音信号时频能量谱方法研究。文中对比分析了常见的几种信号时频方法：快速傅里叶变换，小波变换和魏格纳变换。快速傅里叶变换得到的时频能量谱图，分辨率受窗函数宽度影响很大，小波变换可以得到尺度-能量谱，但基小波和尺度的选取不易。因此，本文最后采用魏格纳变换对正常心音和杂音信号进行时频分析，得到的二维时频能量谱分辨率高，能较好的反映正常心音和杂音信号时频域和能量方面的特征。 4）心杂音信号特征向量提取。对3M Littmann Stethoscopes数据库中正常心音和四种类型的杂音信号进行魏格纳变换得到联合时频能量谱，从中分析时域，频域和能量三方面的特征值，比如杂音持续时间，杂音峰值频率，杂音能量分数等。以此作为特征向量，为心音和杂音的分类识别提供依据。 5）心杂音信号分类分析方法。由于病理心杂音样本数量有限，因此选择支持向量机。课题研究了支持向量机核函数，多分类支持向量机的选取方法，选用以分类精度最大为判断准则网格优化方法来确定核函数参数和松弛变量最优值的选取，建立了适合心杂音分类的支持向量机模型。实验中选取正常心音和四类常见的病理心杂音样本：主动脉瓣狭窄，主动脉瓣反流，二尖瓣反流和肺动脉瓣狭窄心杂音每种类型40例进行测试，按照训练集样本数和测试集样本数之比为3:1的模式进行学习和测试。实验结果表明分类精度较高，均达到了90%以上，验证了算法的有效性。
[Abstract]:Heart sound signal is one of the important physiological signals in human body. Heart-sound auscultation is not only simple and noninvasive, but also can detect abnormality earlier.Because of the complexity and nonstationarity of the cardiac murmur signal, it is an indispensable means to analyze and recognize the cardiac murmur signal by using the modern digital signal processing method.The purpose of this paper is to analyze and study cardiac murmur signals for the purpose of clinical assistant diagnosis, including the analysis of the mechanism of cardiac murmur formation, the extraction of characteristic vectors of cardiac murmur signals, normal heart sounds and 4 kinds of pathological murmur signals (aortic stenosis).Classification and recognition of aortic regurgitation, mitral regurgitation and pulmonary valve stenosis.The research work of this topic mainly includes the following aspects:1) analyzing the characteristics of normal heart sounds and murmur signals, introducing the physiological mechanism and time domain waveform characteristics of normal heart sounds and murmur signals, and estimating the AR model power spectrum of heart sounds and murmur signals.It provides some theoretical basis and reference for the selection of feature vectors.2) the realization of multi-component separation algorithm for cardiac murmur signal.The singular spectrum method based on principal component analysis (PCA) is used to decompose and reconstruct the singularity value of the heart murmur signal, so as to achieve the effect of separating the normal heart sound component from the murmur component.The results show that this method can effectively suppress the crossover interference generated by signal Wigner transform.3) study on time-frequency energy spectrum of cardiac murmur signal.In this paper, several signal time-frequency methods are compared and analyzed: fast Fourier transform (FFT), wavelet transform (WT) and Wigner transform (Wigner transform).The resolution of the time-frequency energy spectrum obtained by the fast Fourier transform is greatly influenced by the width of the window function. The wavelet transform can obtain the scale-energy spectrum, but it is difficult to select the basis wavelet and the scale.Therefore, in the end, we use Wigner transform to analyze the signal of normal heart sound and murmur. The result shows that the two dimensional time-frequency energy spectrum has high resolution and can reflect the characteristics of normal heart sound and murmur signal in time-frequency domain and energy domain.4) feature vector extraction of cardiac murmur signal.Using Wigner transform to obtain the joint time-frequency energy spectrum of normal heart sounds and four types of murmur signals in the 3M Littmann Stethoscopes database, the time domain, frequency domain and energy characteristic values are analyzed, such as the duration of the murmur, the peak frequency of the murmur.Noise energy fraction, etc.It is used as the feature vector to provide the basis for the classification and recognition of heart sounds and murmurs.5) Classification and analysis method of cardiac murmur signal.Because of the limited number of pathological heart murmur samples, support vector machine was chosen.In this paper, the kernel function of support vector machine and the selection method of multi-classification support vector machine are studied. The optimal value of kernel function parameter and relaxation variable is determined by mesh optimization method with the maximum classification accuracy as the criterion.A support vector machine model suitable for classification of heart murmur is established.Normal heart sounds and four common pathological heart murmur samples were tested in 40 patients with aortic stenosis, aortic regurgitation, mitral regurgitation and pulmonary stenosis.Study and test according to the ratio of training set sample number to test set sample number is 3:1.The experimental results show that the classification accuracy is higher than 90%, and the validity of the algorithm is verified.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R318.0;TN911.7

【相似文献】