Research On Denoising Algorithm Of ECG Signal Based On Fully Convolutional Boosting Network

According to the "China Cardiovascular Health and Disease Report 2019",there are 330 million people diagnosed with cardiovascular diseases in China.The prevalence of cardiovascular diseases will continue to rise as the increasing proportion of the aging population.As a mature and noninvasive technology,electrocardiography can assess the physiological state of the heart and make early diagnosis of cardiovascular diseases.However,the collected ECG signal cannot be directly adopted as the basis for disease diagnosis since the collection of Holter can be interfered by patient’s muscle movement,and patch noise,etc.Therefore,denoising the collected ECG signal without losing the useful waveform information has become a hot issue for researchers.Therefore,this paper proposes an ECG signal denoising algorithm based on the fully convolutional boosting network to improve the denoising effect and accurately obtain the waveform characteristics of the ECG signal.The main contents of the paper are as follows:(1)Aiming at the problem that the traditional denoising method loses part of the ECG signal waveform information during denoising,an ECG signal denoising algorithm based on a deep fully convolutional boosting network is proposed.The overall network is based on the SOS boosting algorithm and integrates three fully convolutional neural networks based on autoencoders to extract the deep features of the ECG signal.The signal is enhanced by superimposing the original input signal and the output signal of the previous fully convolutional autoencoding network,and the enhanced signal is adopted as the input signal of the next fully convolutional autoencoding network.In this way,the whole network can reduce the noise and retain the waveform information of ECG signal as much as possible.The experimental results show that the output signal-to-noise ratio of the algorithm for filtering baseline wander,electrode motion and muscle artifacts is 18.858 dB,14.883 dB and 17.308 dB respectively,and the reconstructed waveform is highly coincides with the clean ECG signal waveform.(2)Aiming at the complicated problem of the ECG signal noise of the actual clinical center and further improving the noise reduction performance,an ECG signal denoising algorithm combining wavelet transform and fully convolutional boosting network is proposed.The SOS boosting algorithm is adopted to integrate the wavelet threshold denoising algorithm and the fully convolutional autoencoder network that two different denoising algorithms are integrated to improve the denoise effect of the whole network.At the same time,there are a variety of mixed noise interference due to the complex and changeable noise of the actual clinical application center electrical signal.In the experiment,different types of noise are mixed to verify the filtering effect of the denoising algorithm on complex noise.Experimental verification shows that the output signal-to-noise ratio of the algorithm to filter out mixed noise of electrode motion,baseline wander and muscle artifacts is 2.34 dB which is higher than that of the denoising algorithm on deep fully convolutional boosting network.The algorithm in this paper not only improves the denoising effect of complex noise,but also reduces the network model parameters and improves the generalization ability of the whole network.