基于注意力机制的毫米波雷达ECG信号反演算法研究

Study on ECG Signal Reconstruction Based on Millimeter Wave Radar with Attention Mechanism

关键词:  毫米波雷达,生命体征信号,ECG,卷积神经网络(CNN),注意力机制(Attention),小波变换

Key words:  Millimeter  Wave Radar,ECG,Convolutional  Neural Network (CNN),Attention,Wavelet  Transform

摘要:

针对毫米波雷达反演高精度的人体ECG心电信号,本文提出一种融合小波变换和注意力机制的AM-CNN-GRU神经网络的ECG信号重建方法。首先对雷达信号进行预处理,从中提取与生理特征相关的相位形变信号,并使用滑动窗口去除直流干扰；然后利用小波变换对形变信号进行分解,提取出与ECG信号显著相关的小波基分量,并在CNN-GRU网络模型的基础上,对小波分解的信号引入通道注意力机制,输入到网络中训练得到ECG信号。通过实验数据验证表明,本文提出模型相比传统的CNN-BiLSTM模型,在模型参数更轻量化的同时,取得了更高精度的ECG反演结果。

Abstract:

Aiming at the high-precision ECG signal inversion by Millimeter Wave Radar, this paper proposes an ECG signal reconstruction method based on AM-CNN-GRU neural network with the combination of wavelet transform and attention mechanism. Firstly, the radar signal is preprocessed to extract the phase deformation signal related to physiological characteristics, and the sliding window is used to remove the DC interference. Then, the deformation signal is decomposed by wavelet transform to extract the wavelet base component which have exhibited significantly relationship with the ECG signal. While the CNN-GRU network model is built as the main architecture, wavelet decomposed signal are fed into as input signal after using the channel attention mechanism. The measurement ECG signal and the simultaneous radar signal are trained into the network. The experimental results show that the proposed model is lighter in model parameters and achieves higher precision inversion results, compared with the traditional CNN-BiLSTM model, making it more suitable for the ECG reconstruction by Millimeter Wave Radar.