基于心电特征改良和可变输入卷积神经网络的生物识别

Biometric recognition based on improved ECG feature with variable input convolutional neural network

在生物识别领域,基于心电的识别方法具有难复制、活体性、易采集的优势。本文提出了一种基于心电图特征改良的卷积神经网络,该网络能接受可变长度的心电信号,在基于心拍的生物识别中有效地解决了心拍长度不一致的问题,并对比了目前主流的解决方法。另外,本研究团队制作了USST-ID数据库,该数据库包含117个志愿者静息和运动后的心电记录。最后,在ECG-ID数据库和USST-ID数据库上验证了该方法的优势,准确率分别达到了98.75%和90.47%,均高于目前主流方法,尤其弥补了目前在同时含有静息和运动心电信号上进行生物识别的空白。

In the field of biometric identification,ECG based identification method has the advantages of being difficult to copy,in vivo and easy to collect.This paper proposes an improved convolution neural network based on ECG characteristics.The network can accept ECG signals of variable length and effectively solve the problem of inconsistent heart beat length in biometric recognition based on heart beat,and compare the current mainstream solutions.In addition,the research creates the USST-ID database,which contains resting and exercise ECG records of 117 volunteers.Finally,the advantages of this method are verified in ECG-ID and USST-ID databases,the accuracy rate reaches 98.75%and 90.47%respectively,both higher than the current mainstream solutions.Especially,the fruits make up the current gap in biometrics involving both resting and exercise ECG signals.