Classification Of Multi-beat ECG Signal Based On Improved SENet-LSTM

Electrocardiograph(ECG)plays an important role in the clinic as the main index to judge the state of the heart.However,the abnormality of the heart is usually a flash.If it is not captured immediately,it is difficult to be found in the hospital examination afterwards.Therefore,the portable ECG monitoring system came into being.In this system,the data to be processed is the ECG signal containing multiple beats,and the study of a single beat can not meet the actual needs.Based on this,this paper studies the classification of ECG signals with multiple beats.There are generally two kinds of ECG classification models in the existing literature:one is based on the traditional machine learning method.Although it has strong specificity,the feature extraction process is cumbersome and complex and the generalization ability of the model is poor;The other is to use long short term memory(LSTM)and convolutional neural networks(CNN)to automatically extract the characteristics of ECG signals and realize classification.Although the accuracy is high,they are all for ECG signals of single beat.This paper integrates the data in three public databases,intercepts multi beat ECG signals,and inputs them to CNN and cnn-lstm models built for single beat classification models in the existing literature after wavelet transform denoising.It is proved theoretically that for the initial values given in this paper,the variance of each output layer is equal.In the empirical study,it is found that the CNN and cnn-lstm models are effective in the classification of multi beat ECG signals,with the accuracy of 92.91% and 93.53%respectively,and the F1 score of 0.5618 and 0.5929 respectively.On this basis,in order to improve the classification effect,a classification model combining the improved squeeze and excitation network(SENet)and LSTM is proposed.The improved part is described in detail and the error analysis is given in theory.Input the non denoised multi beat data into the model proposed in this paper.In the empirical study,it is found that the accuracy rate of the model is 95.13%,and the F1 score is 0.6452,which is improved compared with CNN and CNN-LSTM.