Design And Classification Of Respiratory Detection System For Child Pneumonia

With the rapid growth of the country’s development level and the continuous development of industrialization and urbanization,our country has made remarkable achievements.But the ecological environment has suffered,air pollution and urban air pollution has become more and more serious,the air quality index has been consistently high,which has caused a serious burden on the human respiratory system.Nowadays,more and more urban and rural residents die of respiratory diseases,among which infectious pneumonia is a common respiratory disease.Moreover,the prevalence of pneumonia in the elderly and children is much higher than that in adults.Therefore,it is extremely important clinical significance to study a fast,simple,non-invasive,and radiation-free method to assist doctors in diagnosing pediatric pneumonia.The details works of the paper are as follows:1.Establish a pediatric pneumonia detection system.The advantages and disadvantages of electrical sensors and optical sensors has been studied,and the transmission characteristics and application scenarios of various optical sensors were also compared.Then,a Mach-Zehnder(M-Z)interferometer with low cost,simple optical path and high sensitivity was decided to construct the sensing part of the detection system.In this part,the respiratory vibration signals(RSP)of pneumonia and non-pneumonia in children was collected,which were caused by respiratory vibration of human body.(RSP is caused by human breathing sound).2.Sample introduction and data classification.A total of 318 samples were collected in this experiment.The 318 samples were divided into pneumonia and nonpneumonia samples by comparing the clinical symptoms of the subjects and the diagnosis given by the doctors.Among them,the sample number of children with pneumonia was 139,and the other 179 were non-child pneumonia sample.3.Preprocessing of RSP.In this part,discrete cosine transform(DCT)was used to compress the original signal,and compare the reduction degree of signal reconstruction under the two methods.At the same time,wavelet decomposition and reconstruction algorithm(Mallat)was used to denoise the signal.Using the above two signal preprocessing methods,a relatively clean RSP was obtained by batch processing of 319 samples.4.Feature extraction of RSP.In this part,it discusses how to select an appropriate spectrum estimation model to calculate the RSP for pediatric pneumonia and nonpediatric pneumonia.Through the correspondence between the autocorrelation coefficient and partial autocorrelation coefficient of RSP signal,it was decided to use Burg algorithm in the autoregressive model(AR model)to do spectrum estimation analysis and feature extraction for RSP signals.In the meantime,the frequency spectrum parameters extracted by Hilbert transform are combined with the concept of order moment,which was one of the characteristics that distinguish the two types of RSP.In this experiment,a total of six characteristic parameters were extracted.Furthermore,in order to conveniently analyze RSP and reduce the pressure of batch processing samples,a set of simple based on numerical analysis software graphical user interface was designed,which was used for rapid visual analysis of signals.5.Construct a variety of pediatric pneumonia classification models based on machine learning,and perform model evaluation on them.According to the characteristics of feature data sets,the advantages and disadvantages of machine learning classification algorithms were compared.For the task of categorizing RSP for children with pneumonia and non-child pneumonia,we conducted experimental comparison and model evaluation on three models of Support Vector Machine(SVM),Decision Tree(DT)and ensemble learning.Based on multiple evaluation indicators of the two classifications,on classification accuracy,recall,precision,F1-score and AUC comparison,the classification effect of SVM on these two groups of people was better than DT and ensemble learning,and the final classification accuracy was 86.11%.6.Discuss the impact of penalty factors and kernel function parameters on SVM,and compare the optimization results of multiple optimization algorithms for these two parameters.The experiment proves that the setting of parameters will affect the classification accuracy of SVM.Therefore,this paper used three different optimization algorithms,artificial bee colony algorithm(ABC),genetic algorithm(GA),and differential evolution algorithm(DE),were used to further optimize and compare the two parameters of the SVM,taking the classification accuracy as the fitness function.The experiment showed that ABC had the best optimization effect on SVM,and the optimization result was 88.67%.The above experimental resulted show that children’s respiratory signals were collected by optical fiber vibration sensing,and then the classification algorithm was extracted and optimized according to the features,and the classification accuracy can reach up to 88%.This study also provided a new method for the rapid,efficient and non-invasive diagnosis of pneumonia in children.