Lung Sound Signal Reconstruction And Respiratory Cycle Segmentation Combined With EMD And Wavelet Transform

Lung sound signals contain a large amount of physiological information that can reflect abnormalities in the human body.Lung sound auscultation is helpful in the diagnosis of lung sound signals.Traditional auscultation has the characteristics of simplicity and non-invasive.However,in the process of diagnosing lung diseases,the traditional auscultation method takes too long,which is also a significant defect.In particular,during the 2019 COVID-19 pneumonia outbreak,the virus was highly infectious and pathogenic,and the diagnosis results were not timely,which led to the epidemic becoming more serious.Therefore,computerization and automation of lung sound signals is a development trend in the field of lung sound signals.In the process of automatic diagnosis of lung sound signals,segmentation of lung sound signals based on the respiratory cycle is the most basic and one of the most important steps.The accuracy of respiratory cycle segmentation determines the late feature selection and final classification results.In this paper,empirical mode decomposition(EMD)and wavelet transform(WT)are used to study lung sound signal reconstruction and respiratory cycle segmentation.The main contents are as follows:Firstly,it systematically introduces the relevant knowledge of lung sound and the research status of lung sound signal reconstruction at home and abroad.At the same time,it analyzes the current common reconstruction algorithms and points out the existing advantages and problems.Based on clinicians’ labeling of lung sound signals,this article organizes and forms a lung sound dataset.In order to further compensate for the singularity of the dataset,some lung sound datasets are also supplemented from the public data R.A.L.E.Secondly,in the case of excessive background noise in the collected lung sound data,four methods are considered,namely,digital filter filtering,adaptive filter filtering,Wiener filtering,and spectral subtraction filtering.Through experimental demonstration,adaptive filter filtering is finally selected as the filtering method in this work.Subsequently,an improved lung sound reconstruction algorithm was proposed to segment the respiratory cycle,and the method was improved and optimized.By comparing with wavelet reconstruction algorithm and EMD decomposition reconstruction algorithm.Using short-term energy analysis of lung sound signals,the respiratory cycle of lung sound signals is divided to verify the effectiveness of the improved lung sound reconstruction algorithm.This article utilizes an improved lung sound reconstruction algorithm to segment the respiratory cycle of different types of lung sound data in the dataset.In the self built database,the improved lung sound reconstruction algorithm improves the accuracy of respiratory cycle segmentation for normal lung sound signals by 11.3% and 2.13%,respectively,compared to the wavelet reconstruction algorithm and EMD reconstruction algorithm.The improved lung sound reconstruction algorithm improves the accuracy of respiratory cycle segmentation for wet rales in abnormal lung sound signals by 6.81% and 0.94%,respectively,compared to the wavelet reconstruction algorithm and EMD reconstruction algorithm.The improved lung sound reconstruction algorithm improves the accuracy of respiratory cycle segmentation for wheezing sounds in abnormal lung sound signals by 5.08% and 0.19%,respectively,compared to wavelet reconstruction algorithm and EMD reconstruction algorithm.The improved lung sound reconstruction algorithm improves the accuracy of respiratory cycle segmentation for phlegm sounds in abnormal lung sound signals by 1.67% and 1.88%,respectively,compared to the wavelet reconstruction algorithm and EMD reconstruction algorithm.In addition,the improved lung sound reconstruction algorithm also improves the accuracy of respiratory cycle segmentation on public datasets compared to wavelet reconstruction and EMD reconstruction.