Research On Denoising And Feature Extraction Of Lung Sound Signal Based On EMD

With the rapid development of modern medical technologies,lung sound signal analysis,one of the most important pulmonary disease diagnosis methods,has gradually become the focus of studies.lung sound signal that contains abundant information of physiology,pathology and organ function is not only a sound signal but also a typical non-linear and non-stationary multi-component signal,of which,the research on acquisition,denoising,classification and recognition are of great clinical significance assisting physicians to achieve efficient diagnosis and treatment.EMD(Empirical Mode Decomposition)algorithm,the main research contents of this thesis,which is widely used in the time-frequency analysis field is an excellent adaptive data-driven method for analyzing multi-component non-linear and non-stationary signals.Therefore,this thesis mainly studies the EMD algorithm and the processing and analysis of lung sound signal.The specifics are as follows:(1)An improved RPSEMD algorithm based on correlation coefficient is proposed,which is combined with wavelet soft threshold to study the denoising of lung sound signal.Aiming at the mode mixing problem in EMD algorithm,the causes of mode mixing problem and two ideas for solving mode mixing problem are analyzed firstly.The typical algorithms of these two ideas include EEMD and RPSEMD.Summarizing the characteristics of two typical algorithms and embedding the cyclic de-correlation operation in the RPSEMD process to better solve the mode mixing problem and reduce the loss of detailed information.Finally,based on the above research,aiming at the denoising problem in lung sound signals processing,the improved RPSEMD algorithm combined with the soft threshold of wavelet is used to denoise the lung sound signals.The effectiveness of this method is verified by the experiments on the collected lung sound signals.(2)Research on Feature Extraction of Lung Sound Signal Based on Hilbert-Huang Transform.Aiming at the shortcomings of the existing signal analysis methods in time-frequency analysis,the algorithm idea of Hilbert-Huang transform is analyzed.Based on the EMD algorithm,the Hilbert transform is performed on the intrinsic mode function,and the complete 3D time spectrum of time,frequency and amplitude can be obtained by Hilbert-Huang transform.By performing a Hilbert-Huang transform on a respiratory cycle data of each type of lung sound signal,the Hilbert boundary spectrum is obtained,and it can be found that the Hilbert boundary spectrum of the three types of lung sound signals has obvious differences.Analyzing and extracting the effective feature vectors from them,and constructing classification models based on artificial neural network algorithm,support vector machine and K nearest neighbor algorithm to identify three kinds of lung sounds.Experiments show that the method can effectively identify normal sound,crackle and wheeze.