摘要
针对当前动态心率测量方法中存在心率监测准确度不高的缺点,提出使用深度学习算法提取光电容积脉搏波(photoplethysmograph,PPG)中的心率值。方法采集了15名身体健康的受试者不同运动速度下的PPG信号,并通过有抗干扰能力的心电(electrocardiogram,ECG)设备同步采集他们的ECG信号,将具有较强干扰的PPG信号作为堆栈自编码(stacked autoencoder,SAE)网络的输入信号,并将ECG信号作为网络标签,然后使用深度学习算法对自编码网络进行训练,以将有较强干扰PPG信号拟合为具有准确心率特征的类正弦波信号,从而实现对运动状态下干扰严重的PPG信号进行心率的提取。将SAE网络输出信号与对应ECG信号进行比较,结果显示,运动心率测量的平均误差为1.1658 bpm,表明深度学习算法对于心率测量的有效性,也为运动心率信号测量提供了一种新的途径。
The main disadvantage of currentmethods ofdynamic heart rate measurement is the low accuracy. In order to improve the problem, deep learning algorithm was introduced to extract the photoplethysmograph（PPG） of heart rate value. In this paper, the pulse signals of 15 healthy subjects participated in the experiment was acquired under the different veloeityas the input of stacked auto-encoders network （SAE）. At the same time, electrocardiograph （ECG） signal as the label of that network was gathered by a standard ECG collector whichhas high anti-interference. Combining with the deep learning algorithm, SAE was trained, in which the pulse signal with strong interference was fitted to thesignalof sine-like wave with the characteristic of accurate heart rate, in order to realize the extraction of heart rate under the condition of serious disturbance under sports conditions. The experimental results show that compared with the output signal of SAE, the proposed method obtains smaller error value of the heart rate （ 1. 165 8 bpm）, which showsthe effectiveness of heart rate measurementusing deep learning algorithm, and provides a new way for the sportiveheart ratemeasurement.
出处
《电子测量与仪器学报》
CSCD
北大核心
2017年第12期1912-1917,共6页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(81271659
61773408)资助项目
关键词
运动心率测量
深度学习
脉搏
堆栈自编码网络
sportiveheart ratemeasurement
deep learning
pulse
stack auto-encoders network
