基于卷积神经网络的动态脉搏语义分割与心率估计方法

Dynamic pulse semantic segmentation and heart rate estimation based on convolution neural network

通过监测脉搏信号进行心率监测的腕表设备,存在易受噪声干扰、心率估计结果不够准确的问题。为探究能适应多种动态噪声场景、鲁棒性强的心率估计方法,本研究提出了一种基于卷积神经网络的动态脉搏语义分割与心率估计算法。将卷积神经网络与自注意力机制相结合,搭建深度学习模型进行脉搏信号起始点的分割,通过数据后处理抑制异常分割点的影响,并经过计算脉搏信号间期获得心率估计结果。实验结果具有较强的相关性和一致性,平均误差为1.29 BPM,明显优于其他模型。本研究可为基于脉搏信号的心率估计提供一种新的解决思路,具有重要研究意义。

The wrist wearable devices monitor the heart rate by monitoring the pulse signal,but they are vulnerable to noise interference,resulting in inaccurate heart rate estimation.To explore a robust heart rate estimation method that can adapt to a variety of dynamic noise scenarios,we proposed a dynamic pulse semantic segmentation and heart rate estimation algorithm based on convolution neural network.This deep learning model combined the convolution neural network with the self-attention mechanism,could segment the starting points of the pulse signal.The influence of abnormal segmentation points was suppressed by data post-processing,and the heart rate estimation result was obtained by calculating the pulse signal interval.The experimental results had strong correlation and consistency.The average error of heart rate estimation was 1.29 BPM,which was superior to the other models.This study can provide a new solution for heart rate estimation based on pulse signal,and has important research significance.