ECG Signal Classification Based On Incorporation Of CNN&SVR By D-S Evidence Theory

As a vital killer of human health,cardiovascular disease has shown an increasing trend in recent years.Cardiovascular patients usually show symptoms of arrhythmia in the beginning,so early diagnoisi of cardiac arrhythmias in patients is essential.Traditional clinical diagnosis of arrhythmias is to analyzes the pathological information in ECG by doctors.Automatic analysis technology uses computers to analyze the ECG,which can free up the doctors’ energy and time,and monitoring the ECG in real time efficiently.Therefore,constructing an automatic classification system of arrhythmia plays a positive role in the prevention and diagnosis of cardiovascular diseases.This paper uses convolutional neural networks and support vector machines to learn heartbeat features,and incorporate the outputs by Evidence Theory.The specific research contents are as follows:(1)Since the ECG signal has the characteristics such as low-amplitude and low-frequency,it’s susceptible to noise interference.In this paper,according to the different frequencies of noises,various kinds of filters are designed for ECG signal preprocessing.Filter the low-frequency niose by using a median filter to improve the baseline drift in the signal.Removing the high-frequency noise by using wavelet transform method.The characteristics of the ECG signal can be well matained after denoising.(2)For feature extraction part,P wave,QRS complex and T wave need to be completely intercepted in order to extract the pathological information of ECG signal.This article intends to use the differential threshold method for QRS complex detection and segmentation of heartbeat samples.Time-domain feature and Frequency-domain feature are extracted to represent the characteristics of heartbeat signals.(3)Classification of ECG signals.The convolutional neural networks and support vector regression are used to classify the ECG signals.CNN model is used for time-domain feature learning.CNN features self-learning and self-adaptive capabilities.It has advantage in high-level features mining due to its complex network structure.This paper building 1-D CNN based on the traditional CNN model to classify and recognize the ECG signals.Multiple support vector regression machines are used for frequency domain feature learning and classification of ECG signals.The outputs of the two models are incorporated by D-S Evidence Theory to obtain the final classified results.Experimental results show that,compared with single classifier,it has higher accuracy.