Recognition And Spectrum Characteristic Analysis Of Heart Sound Based On Singular Spectrum And EMD

Heart sound is the state reflects of the mechanical movement of the heart and blood vessels,which contains a lot of physiological and pathological information.Due to the complexity of the cardiovascular system,the heart sounds essentially appear as a nonlinear process.How to obtain the valuable characteristic information from the nonlinear heart sound spectrum becomes an important subject in the heart sound analysis.This paper attempts to combine the singular spectrum with the empirical mode to combine the technical path to implement the characteristic analysis of the heart sound spectrum,and use the support vector machine to classify it.The work of the paper is as follows:(1)heart sound signal preprocessing: Because the heart sound signal is extremely weak,When under strong noise conditions the heart sounds are masked by noise,in order to avoid interference of noise on the heart sound,so denoising is before the analysis of heart sound.In this paper,we use the singular spectrum analysis(Singular Spectrum Analysis,SSA)and the noise characteristics to denoise the heart sound.Firstly,do the singular spectrum analysis to the signal,According to the randomness of the noise signal,the variance of the corresponding eigenvector in all directions is very small.so the noise corresponding to the singular spectrum is very flat.Finding the point where the slope begins to level off.That means from the point the change rate of the singular value is very small almost unchanged.This singular values are corresponding to noise.And then reconstruct the remaining singular values to effectively isolate the heart sound from the noise.The simulation experiments of heart sound with noise signals is carried out through the algorithm.Compared with wavelet de-noising,the result shows that the algorithm can achieve better de-noising effect.The innovation of the de-noising method is that the environmental noise is the object of analysis,combine SSA method with the characteristics of the noise to find its corresponding singular value,and then removal these singular values and their corresponding eigenvector to achieve the noise reduction of heart sound.(2)The characteristics parameters of heart sound: According to the normal heart sound main component(S1 and S2)spectrum range(10 Hz ~ 150 Hz)extract heart sound components,the percentage of the energy in the range of 10 Hz to 150 Hz accounted for in the range of 10 Hz to 1000 Hz is taken as a condition,to determine whether the singular value corresponds to the normal heart sound main component.Within the allowable range of 5% error,the spectral analysis of the reconstructed normal heart sound main component reveals that use the singular value that energy percentage is greater than 0.96 to reconstruct normal heart sound main component is more accurate.The energy percentage is greater than 0.96 corresponds to the first and second components of the heart sounds.And then reconstructs the first second heart sound according to the singular value.Excluding the first second heart sound the main component of the remaining part of the signal is heart murmur.And decompose the residual signal using the empirical mode to obtain a set of intrinsic modal functions.Analysis IMF’s power spectral density map found that there are significant differences between normal and abnormal heart sound in frequency range.The normal heart sound within 100 Hz low frequency,abnormal heart sound at 100 Hz ~ 300 Hz high frequency,and then calculate the energy from 200 Hz to 300 Hz of IMF’s and the energy from 0 to 200 Hz,then calculate the ratio of the two energy.The calculated results showed that the energy ratio of IMF1 ~ IMF3 in normal and abnormal heart sounds was statistically different(P<0.05).Finally select the above mentioned energy ratio of IMF1 ~ IMF3 as the characteristic parameters.(3)Classification of heart sounds: the Libsvm is used to design a classifier,Using the radial basis function as a kernel function,And the k-fold cross validation method is used to find the optimal parameter group.Train and test the heart sound sample data.Experiments show that the method can achieve better classification effect.the classification accuracy of normal heart sounds and abnormal heart sounds is up to 100%.