非线性鼾音信号时频特征分析研究

【作者】 张引红；

【导师】 李全禄；

【作者基本信息】 陕西师范大学 ， 声学， 2014， 博士

【摘要】 本文从正常弱生理医学呼吸音信号时频基本特征入手,探讨了鼾音信号的发生机理、传播特性与临床病理生理的相关性,同时考虑病理医学信号发病机理、非线性特性以及瞬时特征频率对正常医学信号的影响等,以期对不同鼾音时频特征进行探索,建立鼾音信号识别模型。主要做了以下几个方面的工作：(1)对正常鼾音信号和病理性鼾症信号的产生机理进行了分析研究。首先,从医学角度分析了两者的产生机理是不同的。正常的鼾音信号产生是由于人在睡眠时全身肌肉放松,如果鼻、咽、喉这三个部位有阻塞,气流冲击狭窄的部位,引起共鸣腔的振动而发出不同程度的响声。鼾症是鼻呼吸的气流吹动咽部的软腭和悬壅垂使之振颤所发生的声音,除响度超出60db外,往往还伴有不同程度的阻塞性睡眠呼吸暂停综合症。其次,从解剖学角度建立模型分析系统,结果表明,阻塞部位的不同,引起鼾音的响度不同,对身体各个部位引起的损害程度也不同。气道阻塞越重,响声就越大,即呼噜声就越响。(2)对鼾音信号检测的分析方法进行了研究。在信号处理方面主要采用的方法有时域分析、谱分析、时频分析等。结合生物医学信号非线性的特点及信号处理时频方法,探讨了小波变换在鼾音信号分析中的应用。由于鼾音属于弱生理信号,采集的信号包含有很多的背景噪声,因此,信号的分析处理较难。首先,利用小波熵去噪原理对信号进行了去噪分析,其次,小波变换的原理,对信号的奇异点进行了检测。第三,主要采用Hilbert-Huang谱分析方法,对实测鼾音进行分析,提取鼾音特征参数。通过比较分析,小波变换作为一种新的多分辨分析方法,可同时进行时域和频域分析,因此特别适合处理鼾音信号。(3)对不同鼾音信号的时频域特征进行了比较分析。这对信号的模式识别具有重要的意义。首先,根据医学诊断标准,甄别正常鼾音与病理鼾症信号的异同。其次,通过提取呼吸音的特征值(如平均能量、倒谱、AR模型系数以及特征参数等),利用模式识别的方法对鼾音信号进行分类。第三,考虑到阻塞性病理鼾音与一般正常鼾音信号的基频、响度、共振峰频率、时间间隔等声学特性参数的不同,研究分析病理信号信息的细微特征、时频谱特征及鼾音信号模型分析系统。结果表明,不同病因引起的异常鼾音信号,它的时域信号特征和频谱特征都是不同的,建立信号仿真分析模型也不同。(4)对甄别不同鼾音信号做了模拟仿真研究。根据鼾音信号时域分布的特点对其分类,建立数据库。采用了Matlab7.x编程语言,同时利用Matlab与Excel动态链接,实现对鼾音信号分析数据的存储与统计。针对呼吸音的非平稳性,研究呼吸音信号动态谱,采用了声谱图、模型分析、软件分析特征参数,从信号中得到病理信息,为医学鼾症的早期诊断提供依据。本论文的主要创新点：(1)提出了应用小波变换进行生理鼾音信号检测的方法。由于生理声信号的复杂性、微弱性及非线性特点,应用小波熵的去除干扰噪声方法取得了很好的效果。(2)采用正交的小波变换时频分析方法,精确分析不同鼾音信号的时频特征,提取谱特征参数能够表征生理信号的病态特征。(3)通过分析信号声学特性,对不同鼾音信号和病理鼾症信号的细微特征进行仿真研究,得到准确的检测与甄别,鼾音信号的突变点得到精确定位。(4)设计鼾音信号检测分析系统。通过拉普拉斯算子和高斯分布建立鼾音信号分析模型,实现鼾音的自动检测及分类。更多还原

【Abstract】 Based on the time-frequency properties of the breathing signal in the normal weak physiologiacal medicine, the relevance of their occurrence mechanism and propagation characteristic of clinacal physiopathology are researched, at the same time, their effects on the normal medical signal and nonlinear characteristic are analysed. In order to explore the different time-frequency feature of snoring signal, the snoring identification model are set up. In this paper, the primary works are outlined as follows:(1) The mechanism of production on the normal snore and pathological sleep apnea syndrome are studied. First, the production mechanism of both are different in the medical science. When he is sleeping, all of the muscle are relaxing. The normal snore signal is formed from the vibration of resonator if nose and pharynx or throat have been obstructed and the air go throuth the narrow parts. Sleep apnea is formed from shaking of the pharyngeal soft palate and nasal breathing air blows, in addition to the loudness beyond60db, there are often accompanied by varying degrees of obstructive sleep apnea syndrome. Second, model analysis system is established from anatomy, the result shows that the different obstructive position may cause different sound loudness and the damage to the body is different. The heavier airway obstruction, the greater the noise is, namely the purring sound is ringing.(2) The detection analysis mothods of the snoring signal are studied. It exists many analysis methods include time domain and spectral analysis and time-frequency analysis in the signal processing. It is discussed that the wavelet transform applied to the snoring signal with nonlinear characteristic of biomedicine signal and time-frequency method. It is difficult to dispose of this weak physiological signal with background noise. First, signal denoising are analysed using the wavelet entropy principle. Second, the wavelet transform is applied to detect the signal singularity. Third, it mainly uses the Hilbert Huang spectral analysis method to analyze the measured snoring and extract its characteristic parameters. Through the comparative analysis, as a new kind of multi-resolution analysis of wavelet transform method, it can be analyzed both in time and frequency domains at the same time. Therefore it is suitable for processing snore signal.(3) The time-frequency domain characteristics of different snoring signals are analysed and compared. It is important to mode recognition of the signal. First, it is to identify the differences and similarities between normal snoring and sleep apnea pathology according to the medical diagnosis standard. Second, by extracting the characteristic value of breath sounds (such as the average energy, cepstrum, AR model coefficient and characteristic parameters, etc.). It using the method of pattern recognition to snore signal classification. Third, it is considered that the many parameters about obstructive pathological snoring and normal snoring signal, such as fundamental frequency, loudness, formant frequency and time interval of acoustic. And pathological signal information and microscopic characteristics and snoring signal model analysis system are researched. The result shows that different signal’s time and frequency characteristic are different and the signal simulation models are different.(4) The analog simulation system of different signal are studied. The database is set up according to signal distributional and classify. Adopting Matlab7.X programming language and using Matlab and Excel dynamic link,it is realized the snoring signal analysis of the data storage and statistics. Adopted a sonogram analysis and non-stationary sounds signal dynamic spectrum, model analysis and software characteristic parameters, pathological information from signals, it is provided the basis for medical science in the early diagnosis of sleep apnea.The main contributions of this thesis are as follows:(1) The wavelet transform, as a new snoring signal detection method, is proposed. As complexity of the physiological acoustic signal, weak and nonlinear characteristics, the application of the wavelet entropy of noise removing interference method has achieved good results.(2) Adoped the orthogonal wavelet transform time-frequency analysis method, it is effect to accurate analysis of different time-frequency characteristics of snoring signal and extract the spectral feature parameters in the characterization of the physiological signal pathological features.(3) Snoring sound signal’s the subtle features are researched by analyzing simulated acoustic characteristics of the different pathologica apnea signal, it gets accurate detection and signal mutation.(4) Snore signal analog simulation software is designed. The result of the signal modeling through Laplacian and Gaussian distributions is a set of possibilities per sound sample for snore events.更多还原