一种改进的PPG信号稀疏分解身份识别方法

An Improved Identification Method Based on Sparse Coding of PPG Signal

光电血容积脉搏波(photo plethysmo graph,PPG)信号作为人体的一种固有的生理信号,包含了人体大量的病理、生理信息,不同个体之间的PPG信号存在着很大的差异,具有很好的保密性和唯一性。文中基于个体间PPG信号的特异性,利用改进的匹配追踪(matching pursuit,MP)稀疏分解算法对不同个体的PPG信号进行分解表示,以此来提取个体PPG信号20个特征值,然后结合PPG信号的1个时域特征值,组成21个融合特征,最后利用决策树分类算法建立分类模型进行分类识别。该方法直接以PPG信号的单个周期波形作为分解对象,不需要对波形进行复杂的变换处理,降低了特征提取的运算复杂性;在过完备原子库上,通过提取的特征可以最大程度地还原PPG信号,提高了特征提取的准确性。最终实验证明,提出的基于PPG信号稀疏分解和机器学习的身份识别方法,识别率可以达到98.3%。

As the inherent physiological signal in human body,the photo plethysmo graph(PPG)contains countless pathological and biological information.There is big difference between PPG signals in different human bodies which makes it confidential and unique.Based on the specialty of individual PPG signals,we expound on the application of matching pursuit(MP)sparsely coding individual PPG signals which extracts 20 matrix eigenvalues of individual PPG signals.Then we explain how these matrix eigenvalues and one TD eigenvalues constitute 21 integrated characteristics.Lastly we illustrate how the decision tree method is employed to build a classification model for identification.This method directly uses the single periodic waveform of the PPG signal as the decomposition object,and does not require complex transformation processing on the waveform,which reduces the computational complexity of feature extraction.On the over-complete dictionary,the PPG can be restored to the greatest extent through the extracted features and the method improves the accuracy of feature extraction.Finally,the experiment has shown that the identification rate can reach 98.3%based on the identification technique through PPG sparse coding and machine learning proposed.