深度残差网络在脉搏信号亚健康检测中的应用

Application of Deep Residual Network in Pulse Signal Sub-health Detection

传统的脉搏信号亚健康检测主要采取手工提取特征,这类方法容易受人为主观意志的影响,从而导致亚健康检测的识别率较低。针对这一问题,将深度残差网络方法应用于信号特征提取领域,提出一种适用于脉搏信号亚健康检测的深度残差网络模型。首先,针对实验中存在的脉搏信号样本数量不足的问题,在生成式对抗网络的基础上提出了一种脉搏信号的生成方法,对脉搏信号数据集进行扩增;然后针对脉搏信号的特点,改进深度残差网络,引入一维卷积,构建适用于脉搏信号亚健康的检测模型;最后,利用扩增之后的数据集训练该模型,对人体亚健康状态进行检测。实验结果表明,该方法能够有效地区分健康与亚健康状态,与现有的方法相比,可以取得更高的识别率。

The traditional pulse signal sub-health detection mainly adopts manual extraction of features,which is easily affected by the subjective will of human beings,resulting in the lower recognition rate of sub-health detection. Aiming at this problem,we apply the deep residual network method to the field of signal feature extraction,and propose a deep residual network model suitable for pulse signal sub-health detection. Firstly,aiming at the problem of insufficient samples of pulse signal in the experiment,a generation method of pulse signal is proposed based on the generative adversarial network,which can amplify the pulse signal data set. Then,according to the characteristics of pulse signal,the deep residual network is improved and one-dimensional convolution is introduced to construct a detection model suitable for the sub-health of pulse signal. Finally,the model is trained by using the data set after amplification to detect the sub-health state of the human body. The experiment shows that the proposed method can effectively distinguish between healthy and sub-health status,and achieve higher recognition rate than the existing methods.