Studies On Classification Of ECG Signals Based On Convolution Neural Network

With the development of the Internet,people’s living habits have also changed.The habits of sleeping late and lack of exercise are affecting people’s physical and mental health.Especially in the situation of an aging population,heart disease has become one of the major diseases that threaten human health.The electrocardiogram shows the invisible working state of the heart in the form of an image.Electrocardiogram is an important reference for cardiac health in clinic.Accurate reading of the ECG waveform is beneficial to the judgment of the working state of the heart and to the diagnosis and treatment of heart diseases.The ECG signal directly collected from human body is a very weak bioelectric signal,which contains much noise.How to remove noise from the signal,extract useful ECG signals,and provide reliable reference for medical workers has always been a hot spot for scientific researchers.In this paper,the following three aspects of ECG signal analysis are mainly studied:1.To explore suitable noise reduction methods.The ECG signal contains a variety of noises.Denoising the signal is beneficial to the extraction and mining of ECG signal features as well as the classification of ECG signals.Based on the characteristics of the ECG signal and the combination of various factors,the traditional wavelet decomposition and threshold processing method is adopted to preprocess the signal,which effectively reduces the interference signal in the ECG signal.With the interference component reduced,the original characteristics of the signal become more obvious,and it is more conducive to the subsequent feature extraction and classification.2.To find better classification method of ECG signals by comprehensive analysis.The classification model in this paper is a deep learning classification model based on convolutional neural network algorithm.This paper compares and analyzes the methods based on sparse expression classification(SRC)and convolutional neural networks.The results show that the accuracy of classification method of convolutional neural networks is higher.This study is based on the MIT-BIH arrhythmia database provided by the Massachusetts Institute of Technology.The classification method of ECG signals applied in this paper is composed of wavelet decomposition,soft and hard threshold combination filtering,original signal reconstruction,and finally using the convolutional neural network algorithm in deep learning based on the four types of disease data in the MIT-BIH arrhythmia database.The accuracy of classification is99.9%.The convolution neural network can autonomously extract and fit the characteristics of the data and finally complete the classification.Compared with traditional algorithm,it avoids the complexity and limitation of manually extracted features,and has better robustness for processing high-dimensional and large-scale data.The convolution neural network uses convolution checking data for local perception and weight sharing,which greatly reduces the operating parameters and calculations of the network.After horizontal comparison of the experimental process and the experimental results,it can be seen that the signal classification method based on convolution neural network is obviously superior to the traditional classification method in terms of classification workload and classification accuracy.3.Data selection and analysis.