基于深度残差网络的单通道脑电信号身份识别
首发时间:2018-06-04
摘要:在多种生物特征识别模态中,用脑电波(Electroencephalogram,EEG)进行身份识别具有抗伪造性和抗胁迫性等诸多优势。单通道脑电信号采集设备以其简单易用性,提高了脑电信号应用于身份识别的实用性。然而,单通道脑电信号通道数少,仅能提供较少的信息。因此,针对单通道脑电信号身份识别任务,建立高准确率和高鲁棒性的模型是值得研究的。传统的脑电信号身份识别方法通常涵盖预处理、特征提取和用分类器分类等多个步骤,而利用卷积神经网络(Convolutional Neural Network,CNN)可简化相关过程。深度残差网络(Residual Network,ResNet)在卷积神经网络的基础上,加入了跨层连接以减小随层数加深而引起的性能退化。本文将深度残差网络的跨层连接思想应用于 BUPT-MCPRL 脑电信号数据集,并在 23 人数据集和 100 人数据集上都取得了高于传统方法的身份识别准确率。
For information in English, please click here
Single-channel EEG-based personal identification based on ResNet
Abstract:Electroencephalograph (EEG) provides resistance to forgery and coerciveness and can be employed for personal identification. Single-channel EEG device improves operability in real application but obtains less information in the collected signals. Therefore, effective and robust algorithm for single-channel EEG-based personal identification is of great significance. Previous methods usually consist of preprocessing, feature extraction and classification and these phases can be simplified by applying convolutional neural network (CNN). Deep residual nets (ResNets) use shortcuts to avoid degradation during accelerate layers of neural network. In this paper, the experiment results on the two datasets of BUPT-MCPRL verify the effectiveness of the proposed method.
Keywords: Signal and information processing EEG Personal identification ResNets
引用
No.****
同行评议
勘误表
基于深度残差网络的单通道脑电信号身份识别
评论
全部评论0/1000