Research On R Peak Location And Arrhythmia Recognition Method Of ECG Signal Based On Deep Learning

With the rapid development of deep learning theory and technology,the recognition of arrhythmia with deep neural network has become a research hotspot in the field of intelligent medical health.However,the arrhythmia recognition model based on deep learning lacks efficient R peak location method and spatio-temporal feature extraction method of ECG signal,which affects the accuracy of arrhythmia recognition and seriously restricts its wide application in computer-aided arrhythmia diagnosis.In view of the above problems,this thesis takes ECG signal as the research object.Based on deep learning methods and theories,it focuses on ECG signal R peak positioning method and arrhythmia recognition method.The main research contents are as follows:1)QRS complex detection and R peak location method based on improved U-net.This method extracts the spatial and temporal features of ECG signals by convolutional neural network and bidirectional long-term and short-term memory network respectively,and adds residual paths when connecting quickly,so as to solve the problem of feature difference between encoder and decoder in U-net network.Finally,the threshold filtering algorithm is used to locate the R peak.The experimental results of MIT-BIH database,CPSC2019 database and PTB-XL database show that the R peak positioning accuracy of this method is better than that of the existing similar methods.2)Arrhythmia recognition method based on multi-scale feature fusion neural network model.Based on the above R-peak positioning method,this method uses the current heart beat,the QRS complex of the current heart beat,and the ECG signals of three scales before and after the current heart beat to extract the spatio-temporal features through the automatic encoder.Then,each group of spatio-temporal features is weighted by the spatio-temporal attention module.Finally,the three groups of weighted features are weighted by the channel attention module,and the arrhythmia classification is performed by the Softmax function.The experimental results on MIT-BIH database show that the accuracy of arrhythmia classification reaches 99.52 %,which is better than the existing similar methods.In this thesis,the R peak positioning accuracy of ECG signal and the spatio-temporal feature extraction ability of arrhythmia detection model are studied,which effectively solves the problem of arrhythmia recognition accuracy.The proposed method will lay a theoretical and technical foundation for artificial intelligence fusion arrhythmia recognition,and provide new ideas for improving the efficiency of ECG signal analysis.