基于语谱图特征量的肺音分类方法研究

发布时间：2018-02-23 12:11

本文关键词： 肺音种类识别语谱图灰度共生矩阵(GLCM) 支持向量机　出处：《东北师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：肺音是人体的呼吸系统与外界环境在换气过程中产生的一种生理声音信号,这些声音信号中蕴含了大量丰富的的生理病理信息。近年来随着大气污染的日益加重,各种肺部疾病的发病率也日益增加,尤其是哮喘,支气管炎等呼吸系统疾病,使得肺音的研究越来越受到学术界的重视。自从世界上最早的听诊器由法国医生雷奈克在1861年发明以来,直至今日,医学上对人体肺音的听诊一直沿用最基本的听诊器方式,医生通过听诊器中听到肺部的声音,结合自己的临床经验对病人进行诊断,存在很大的主观性,并且肺部杂音过大,传统的听诊方法极易漏听大量有用信息,从而影响医生的准确诊断。直至目前,在临床医学上,肺音听诊的结果还不能被定量的表示。但在肺部的诸多部位上检测到的声音信号中蕴含着大量可被利用的信息,因此,利用肺部声音信号诊断疾病具有巨大的潜力。人类需要一种新的诊断方式,对肺音中包含的丰富的信息进行分析,当肺音被更深层次的发掘后,医生就能更有效地利用简单的听诊技术,高效的分析和诊断出各类病症。医疗费也会大幅度下降,这一前景是十分诱人的。当前,高速发展的智能信息处理技术为这一课题的研究提供了强大的工具,将数字信号处理等技术运用到肺音研究中,已经提上了日程。本文提出了一种基于语谱图特征量提取方法,可实现对人体肺音的种类的分类识别,应用MATLAB软件进行编程仿真。将提取的9种肺部声音信号进行预处理,转换成语谱图。并运用图像处理技术及数学形态学图像处理技术对语谱图进行预处理操作,利用灰度共生矩阵(GLCM)对特征量进行提取,分别得到Contrast(对比度),Correlation(相关性),Energy(能量),Homogeneity(熵)四类特征值,将以上四个特征集构造为特定肺音种类识别的特征向量,作为支持向量机的输入向量,运用四类不同核函数对样本进行分类,实现种类识别。实验表明:本方法对特定肺音种类的识别率达到83.3%,基本实现了特定肺音种类的识别,可为医学上的肺音识别问题提供了一种新的辅助方法。
[Abstract]:Lung sound is a kind of physiological sound signal produced by the respiratory system of human body and the external environment in the process of air exchange. These sound signals contain a lot of physiological and pathological information. The incidence of various lung diseases is also increasing, especially respiratory diseases such as asthma, bronchitis, etc. Since the earliest stethoscope in the world was invented by French doctor Renaike in 1861, the medical auscultation of human lung sounds has been based on the most basic stethoscope method. Doctors hear lung sounds in stethoscopes and diagnose patients with their own clinical experience. There is a lot of subjectivity, and the lung murmur is too loud. Traditional auscultation methods are easy to hear a lot of useful information. So as to affect the accurate diagnosis of doctors. Until now, in clinical medicine, the results of lung sound auscultation can not be quantitatively expressed. But the sound signals detected in many parts of the lungs contain a lot of information that can be used, so, There is great potential for diagnosing diseases by using pulmonary sound signals. Humans need a new way to diagnose diseases by analyzing the rich information contained in lung sounds, and when lung sounds are further explored, Doctors will be able to use simple auscultation techniques more effectively to analyze and diagnose all kinds of diseases efficiently. The prospect of a sharp drop in medical costs is very attractive. The rapid development of intelligent information processing technology provides a powerful tool for the research of this subject. The application of digital signal processing and other techniques to lung sound research has been put on the agenda. The classification and recognition of human lung sounds can be realized, and the MATLAB software is used to program and simulate. The 9 kinds of lung sound signals extracted are preprocessed. Using image processing technology and mathematical morphology image processing technology to preprocess the spectral map, and using the gray level co-occurrence matrix (GLCM) to extract the feature quantity. Four kinds of eigenvalues of energy source energy (entropy) are obtained respectively. The above four feature sets are constructed as the feature vectors for the recognition of specific lung sound types, which are used as input vectors of support vector machines. Four kinds of kernel functions are used to classify the samples and realize the category recognition. The experiment shows that the recognition rate of the specific lung sounds is 83.3%, and the recognition rate of the specific lung sounds is basically realized. It can provide a new assistant method for lung sound recognition in medicine.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R443;TN912.3

【参考文献】