基于深度全卷积提升网络的心电信号降噪
ECG SIGNAL DENOISING BASED ON DEEP FULLY CONVOLUTIONAL BOOSTING NETWORK
杨畅 1刘慧妍 1刘明1
作者信息
- 1. 河北大学电子信息工程学院河北省数字医疗工程重点实验室 河北保定 071002
- 折叠
摘要
针对传统的降噪方法难以在不丢失心电信号下准确去除复杂噪声问题,提出一种基于深度全卷积提升网络(FCBN)的心电信号降噪方法.该方法利用全卷积网络的局部连接的特性来保留心电信号波形细节信息,通过提升(Boosting)算法堆叠多个FCN网络形成深度神经网络,多级输入原始信号,保留心电信号的深层信息特征,提高整体网络的降噪性能.实验结果表明,该方法与小波阈值法、S变换法、BP神经网络法和卷积自动编码器比较,信噪比提高明显且均方根误差较小,同时可保留更多心电信号的波形形态信息.
Abstract
The traditional method of denoise is difficult to accurately remove the complex noise without losing the ECG signal,an ECG signal denoising method is proposed based on the deep full convolutional Boosting network(FCBN).This method used the characteristics of the local connection of the full convolutional network to retain the detailed information of the ECG signal waveform,and stacked multiple FCN networks through the Boosting algorithm to form a deep neural network.The original signal in multiple stages were inputted,and it retained the deep information of the ECG signal features to improve the denoising performance of the overall network.Experimental results show that compared with wavelet threshold method,S transform method,BPNN method and convolutional autoencoder,this method has obvious improvement in SNR and smaller RMSE,while retaining more ECG signal waveforms morphological information.
关键词
心电信号/降噪/全卷积网络/提升算法Key words
ECG signal/Denoising/Fully convolutional network/Boosting algorithm引用本文复制引用
基金项目
国家自然科学基金(61703133)
国家自然科学基金(61673158)
河北省自然科学基金(F2018201070)
河北省青年拔尖项目(BJ2019044)
出版年
2024
国家科技期刊平台
NSTL
万方数据