Study On ECG Signal Analysis And Arrhythmia Classification

Nowadays,the average life expectancy of people is gradually increasing with the development of technology and medicine.However,the more and more people are suffering from cardiovascular disease with the increase of their age,which is becoming a more serious problem in people’s normal life.Electrocardiogram(ECG)is an important indicator for diagnosing people’s heart health.In order to prevent or reduce the occurrence of heart disease,the effective detection and identification of ECG is more and more important.The use of computers to automatically diagnose ECG signals can effectively reduce the workload of doctors.To solve this problem,this article has conducted in-depth research on the analysis of ECG signals and arrhythmia classification.The main research contents include:(1)In ECG signal feature extraction aspect,this paper proposes a better adaptive feature extraction method-Ensemble Local Mean Decomposition(ELMD)method.This method first adds different Gaussian white noise to the ECG signal,and then performs local mean decomposition(LMD)to obtain several Product Function(PF)components,and obtains the mean value of the PF component after multiple decompositions.The process of adding noise multiple times and component averaging can overcome the model confusion problem in the basic LMD method.The first four superior PF components are selected to calculate the sample entropy,average power,singular value and standard deviation.The obtained eigenvector matrix can effectively represent the arrhythmia information in the ECG signal.(2)For the classification of arrhythmia ECG signals after feature extraction,this paper chooses to establish a classification model using support vector machines(SVM).In addition,the genetic algorithm is used to optimize the two parameters of the support vector machine: the penalty factor C and the kernel function parameters ?.The feature vector matrix decomposed by the ELMD method is sent to SVM for classification.From the classification results of the MIT-BIH arrhythmia database,the overall classification accuracy of the ELMD method has reached 99.61%,which is not only higher than 98.28% accuracy rate in the LMD method but also higher than the general method.This proves the effectiveness of the ELMD method.(3)Deep learning methods can automatically extract signal features and classify them.Therefore,this paper proposes an arrhythmia classification method combining Gram’s Angle Field(GAF)method and improved convolutional neural network.First,the GAF method is used to convert one-dimensional ECG signals into two-dimensional images.This method does not lose signal information in the process of converting ECG data,while retaining the time dependence of the original signals.Based on the idea of transfer learning,a transfer convolution neural network(TCNN)was designed to automatically classify ECG images.The first 13 layers of the network structure are migrated from the classic VGG16 convolutional neural network structure,and then a convolution layer,a fully connected layer,and a softmax layer are added as the entire network structure.After the imbalanced arrhythmia data was processed through data balance,the TCNN network was used to classify the normal ECG signals and the four types of arrhythmia ECG signals,and finally reached a total accuracy of 99.8%,which is higher than the general deep learning classification method.This proved the effectiveness of this proposed method.