计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (16): 148-155.DOI: 10.3778/j.issn.1002-8331.1906-0089

• 模式识别与人工智能 • 上一篇    下一篇

基于卷积自编码神经网络的心电信号降噪

陈健,刘明,熊鹏,孟宪辉,杨林   

  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
  • 出版日期:2020-08-15 发布日期:2020-08-11

ECG Signal Denoising Based on Convolutional Auto-encoder Neural Network

CHEN Jian, LIU Ming, XIONG Peng, MENG Xianhui, YANG Lin   

  1. Key Laboratory of Digital Medical Engineering of Hebei Province, College of Electronic Information Engineering, Hebei University, Baoding, Hebei 071002, China
  • Online:2020-08-15 Published:2020-08-11

摘要:

心电信号由于在采集过程中会受到外界环境的干扰导致其形态特征被严重淹没,从而对医生的诊断和远程智能分析造成干扰。基于此,提出了一种基于卷积自编码神经网络的心电信号降噪算法。该方法利用自编码器的编码、解码特性,通过卷积的方法构建深层神经网络来学习从含噪心电信号到干净心电信号的端对端映射。卷积层捕获心电信号的细节特征,同时消除噪声;解码部分能够对特征图进行上采样并恢复心电信号细节,从而得到干净的心电信号。实验中采用信噪比和均方根误差为指标,将该方法与小波阈值法、S变换法、BP神经网络法和指导滤波法进行比较。实验结果表明,该降噪方法整体降噪精度更优,同时信号的低频成分也得到了很好的保持。该方法可做到在消除心电信号中复杂噪声的同时完整保留心电信号的形态,为心血管疾病的智能诊断和心电图的特征检测奠定了基础。

关键词: 心电信号, 降噪, 自编码器, 卷积神经网络

Abstract:

The ECG signal is seriously submerged due to interference from the external environment during the acquisition process, thus causing interference to the doctor’s diagnosis and remote intelligent analysis. Based on this, this paper proposes an ECG signal denoising algorithm based on convolutional auto-encoder neural network. The method utilizes the encoding and decoding characteristics of the auto-encoder, and constructs a deep neural network by convolution to learn the end-to-end mapping from the noisy ECG signal to the clean ECG signal. The convolution layer captures the details of ECG signals content while eliminating noise; the decoding portion can upsample the feature map and restore the ECG signal details to obtain a clean ECG signal. In the experiment, the signal-to-noise ratio and root mean square error are used as indicators, and the method is compared with wavelet threshold method, S-transform method, BP neural network method and guided filtering method. The experimental results show that the overall noise reduction accuracy of the noise reduction method is better, and the low frequency components of the signal are also well preserved. The method can eliminate the complex noise in the ECG signal while completely retaining the shape of the ECG signal, which lays a foundation for the intelligent diagnosis of cardiovascular disease and the characteristic detection of ECG.

Key words: ECG signal, denoising, auto-encoder, convolutional neural network