Research On Detection Algorithm Of Heart Sound And Lung Sound

At present,the research on heart sound signal or lung sound signal is carried out separately for one signal.This single kind of research may ignore the pathological changes of heart and lung.Therefore,this thesis makes a comprehensive study on heart sound signal and lung sound signal.In this thesis,three kinds of heart sounds including normal heart sound,heart murmur,additional heart sound and three kinds of lung sounds including normal lung sound,wheezing sound and burst sound are collected,and these six kinds of heart and lung sound signals are detected and analyzed.The general detection steps are divided into three steps: preprocessing,feature extraction and classification and recognition.In this paper,each kind of cardiopulmonary sound is associated with the possible diseases.Through the identification of six kinds of cardiopulmonary sounds,the diagnosis of cardiovascular and respiratory diseases corresponding to each kind of cardiopulmonary sound is realized.In view of the problems of different duration and unstable signal in the collected cardiopulmonary audio segment,six kinds of cardiopulmonary audio signals are pre emphasized,windowed,framed and standardized.The sampling rate is 44100 Hz,and2000 frames are intercepted for subsequent processing.In order to classify and identify the six cardiorespiratory sounds,their corresponding characteristic parameters need to be extracted.In this paper,seven features of six kinds of cardiopulmonary sounds are extracted from time domain,frequency domain and cepstrum domain respectively,and combined into a 40 dimension original feature set.The time domain features include short-time energy,short-time average amplitude and zero crossing rate;Frequency domain features include short-time Fourier transform and linear prediction coefficient;The characteristics of cepstrum domain include 12 th order Mel cepstrum coefficient and 12 th order linear cepstrum coefficient.In order to improve the classification and recognition rate without loss of effective information,the relief algorithm is used to optimize the 40 dimension original feature set.The BP neural network classifier is used to classify and detect the original feature set and relief feature set of 500 cardiopulmonary sound test samples respectively.The average recognition rates are65.52% and 95.62%.The experimental data show that the recognition effect of relief feature set is better.In order to further improve the recognition rate,GA-BP neural network improved by genetic algorithm is used to detect and classify the original feature set and relief feature set of six kinds of cardiopulmonary sounds.The average recognition rates are 79.20% and97.33%.Compared with the average recognition rate when BP neural network is used as classifier,it can be seen that the average recognition rates of the two feature sets are increased by 13.68% and 1.71% respectively.From the experimental comparison results,the GA-BP classification effect of the relief feature set of cardiopulmonary sounds is the best.