Study On Method Of Heart Sound Signal Extraction Based On EMD And SVD

Heart sound is an important physiological signal of body, that can reflect the health of heart and the cardiovascular system. Detecting and analyzing the heart sound signal can help us achieve the early diagnosis and warning of many heart diseases. With the development of modern signal processing technology and biomedical engineering technology, the research on heart sound analyzing have a great development from traditional artificial auscultation for qualitative analysis, to quantitative analysis of characteristic traits. But, the PCG signal acquisition process unavoidable be disturbed by noise and interference from other organs sound(like lung sound), it will seriously affect the accuracy of subsequent analysis for heart sound characteristic. So how to achieve effective noise reduction and extract characteristic information accurately is important content in heart sound research area.Since the heart sound signal and the noise generally are nonlinear mixed, the traditional time-frequency domain analysis method could not effectively remove the noise. Empirical model decomposition(EMD) is a data-driven nonlinear decompose method, it is suitable for non-stationary signals analysis. It has high adaptability, because of its time-frequency resolution can change with the signals' characteristics. The EMD method has achieved good results in the separation of heart and lung sounds. However, due to the low SNR of heart sound and the overlapping between noise and signal, the intrinsic mode functions(IMF) obtained by EMD will be mixed with each other, which would seriously affect the separation effects. Accordingly, this paper introduces the single channel singular value decomposition(SSVD) method on the basis of EMD. It can further decomposition the IMFs into a number of feature components, and selecting the important feature components to reconstruct the heart sound signal. This method reduces the noise of heart sound by using the quasi-periodicity of heart sound in the time domain. Base on the above, we propose a complete framework named empirical model decomposition-cycle aligned singular value decomposition(EMD-CASVD) to reduce the noise and extract the signalSpecific studies include: First, we do research on physiological characteristics and time-frequency domain characteristics of heart sound and analysis the characteristics of noise between the heart sound acquisition process, and we propose an adaptive heart sound positioning and segmentation method to locate the main components of heart sound. Second, we study the EMD principle and its application on heart sound decomposition, and then use EMD method to decompose the heart sound signal into some IMFs, effectively filter out part of the noise by using the IMF's narrowband characteristics. Third, in order to solve the mode mixed problem of IMFs, we propose a special single singular value decomposition method which make the decomposition matrix constructed by aligned heart sound signal cycle, and use it to further separation the noise and signal components of each IMF. Fourth, we propose a method to filter the feature components by the comprehensive information on the energy and waveform shape of each feature component, and then use the extracted feature components to reconstruct the heart sound signal.We separately test the EMD-CASVD method and the traditional wavelet threshold(WT) de-noising method. Experimental simulation results show that EMD-CASVD method proposed in this paper has better de-noising performance, adaptability and noise robustness than WT.