Design And Implementation Of Respiratory Sound Classification System

The respiratory system is an important system for maintaining the normal functioning of organisms and metabolism of the organism.The occurrence of any lung disease is bound to cause abnormal respiratory system.As the mortality rate of respiratory diseases in China is increasing,all social circles has gradually begun to pay attention to the signal of respiratory sounds.The study of breath sound signals can not only effectively prevent and treat respiratory diseases,but also promote the development of clinical medicine and medical signal processing technologies and the development of related medical instruments.At present,classification and recognition of breath sound signals has become a concern at home and abroad hot spot.The purpose of this project is to design a respiratory sound classification and identification system to effectively classify common respiratory sounds.The main research content is to use the Cool Edit Pro software to perform preliminary processing on breath sound signals used in clinical medical standard listening training,and to cut each type of breath sound into several sample data as experimental data.After extracting the effective features of the experimental data,select the Support vector machine(SVM)and BP neural network were used as feature recognition models of breath sounds.Recognition experiments run on the MATLAB software platform.First,the sample data were pre-processed and endpoint detection to reduce the processing time of invalid data.Due to the differences in the energy levels of different types of breath sounds at different time periods,the two extracted short-time time domain parameters,namely short-term energy and short-time zero-crossing rate characteristic parameters,are used as feature parameters.And refer to the value of Mel cepstrum coefficient(12 dimensional MFCC)commonly used in speech recognition,Because of the non-stationary and dynamic characteristics of respiratory sounds,MFCC is optimized and its dynamic MFCC parameters(12 dimensional first order differential(35)MFCC)are extracted.The short-time time domain parameter method is combined with the respiratory sound characteristic parameters extracted bythe static and dynamic parameter method of MFCC as the combination characteristic parameters.The experimental results show that the extracted feature parameters satisfy the requirements of difference,unity and relevance the needs of training recognition,and the recognition effect of combined characteristic parameters is obviously better than that of two types of characteristic parameters uniquely identifies.Because of the many types of respiratory sounds,one kind of recognition model can not identify each class of respiratory sound signals accurately.Therefore,two types of recognition models are introduced in this paper.The normalized characteristic parameters as sample data,they were input into the BP neural network and SVM for training and recognition.The experiment shows that the BP network has a great difference in the recognition of different types of respiratory sounds(The coarse rales,sonorous wheezes,rales crepitus,medium rales recognition rate of 98%,but the recognition rate of bronchial breathing with dampness rales is less than 70%).The recognition rate of SVM for each type of respiratory sounds was stable,and the overall recognition rate can reach 97.857%.A stable and accurate identification sample library can be built.Based on the above theoretical methods and experimental research,the graphical design interface tool of MATLAB programming software is used to design and implement a breath sound classification and identification system.The recognition system can achieve the following functions:calling a preliminary processed breath sound sample,extracting feature parameters of the sample data to build a sample database.Input the feature parameters of the constructed sample library into the recognition model for training,and then extract the characteristic parameter input identification model of the breath sound to be identified for rapid classification and identification.The recognition results are displayed on the output side for reference by the testee.