Research On Characteristic Analysis And Recognition Methods About Heart Sound Signal

Heart sound signal is defined by myocardial contractility, cardiac valvular heart walland blood crash, caused by sound vibrations generated by the aorta wall, etc., containimportant signal of human physiology and pathology of the information contained in thevarious parts of the heart. Currently the diagnosis of disease through the heart sound signalmethod is mainly manual auscultation, if there are any effective algorithm to automaticallycollect the identification and classification of heart sounds while finding lesions automaticallymake an assertion even make a alarm signal, there will be a great clinical significance, this iswhat the study carried out according to the demand mentioned above. In this paper, thefollowing works are finished.First, this thesis in the view of physiological point to make the heart sound signalsgenerated which are described and analyzed the main components of the heart sound signalsand time-frequency characteristics of heart sound signals common heart murmur andtime-frequency characteristics. As heart sound signal automatic identification system dataacquisition part, the paper use HKY-06B acquisition heart sound heart sound sensor signal,heart sound signal format conversion and storage through CoolEditPro2.1recording software.Second, the pretreatment heart sound signal. This process includes framing heart soundsignals, heart sound signal endpoint detection and heart sound signal de-noising. Heart soundsignal sub-frame using the sliding rectangular window on the heart sound signal processing.Endpoint detection using a dual-threshold endpoint detection method. Noising heart soundsignals using wavelet analysis. This article analyzes the selected basis function method, thedecomposition level and threshold values through discussion and analysis wavelet,combining the characteristics of heart sound signals, select the appropriate method ofpretreatment process. Experimental results show that the wavelet function by sym7heartsound signal decomposition layer5, can achieve the desired de-noising effect.Third, This paper using two methods of heart sound signals were extracted, respectively,of the heart sound signal EMD decomposition and extraction of heart sound signals MelFrequency Cepstral (MFCC parameters) lineages. By comparing the results of the two ways,MFCC parameters are found could be decomposed to obtain better recognition results thanEMD, and greatly shorten the training time model.Fourth, this paper using Gaussian mixture model to classification and identification the heart sound signals. The heart sound signal are classificated into the main signal of normaland pathological heart sound heart sound signal. Through the44cases of normal heart soundsignal Gaussian mixture model libraries were established to construct a model library of heartsounds. When recognizing, extracting characteristic parameters of the model to be identifiedwith the training heart sound signal is found, the likelihood function value between themodels it, like the model to obtain the maximum likelihood function value corresponding tothe treatment is the result of measuring the heart sound classification. Experimental resultsdemonstrate the feasibility of heart sound signal pattern recognition system, the correctidentification rate of normal heart sounds and pathological heart sound can reach90%.