EEG Classification Based On Time Sequences

Personal identification by EEG(Electroencephalogram)has a great potential in ascending security and accuracy of identification,due to merits that EEG calls for intravital detection and it is hard to be forged,sensitive to forcing,facile to update,etc.EEG identification used to require multi-channel signals collected by complicated devices and experimental setting,which put practical application into inconvenience.Therefore,this research concentrates on methods of utilizing single-channel EEG signal for EEG-based personal identification.EEG-based personal identification is a specific problem in EEG classification,so it shares multiple methods with other EEG classification problems.Previous research always classify EEG signals by three stages including preprocessing,feature extraction and classification.With special structure and powerful fitting capacity,neural networks can unify the three stages so that directly classify EEG signals according to their sequential information.This paper thus apply neural networks on EEG-based personal identification.As discussed above,results of this paper are showed as follows.1.EEG signals of 53 subjects are collected for diverse experiment objectives and research contents.Including signals from 40 persons in their rest status and 13 subjects in their different emotional status,the signals expand existing database and support this research.2.Explore how signal dimension influence identification accuracy.Accordingly based on one-dimension representation,design ResNet-EEG algorithm by apply ResNet(Residual Network)method on EEG personal identification.Experiment results show that designed ResNet-EEG surpass baseline on identification accuracy to a large extent.3.Merge LSTM(Long short-term memory)with different network structures including RCNN(Recurrent Convolutional Neural Network),ResNet-EEG,etc,so that to establish LSTM-based variable-length EEG personal identification algorithm.The designed algorithm makes EEG personal identification more practical by decreasing the necessary length of EEG signals for identification,and making real-time identify result available.4.On augmented database,quantitatively explore how multiple elements influence the accuracy of EEG personal identification.The elements cover physical status,imagining status,tired status,emotional status,etc.According to various experiment results,suggestions are given for actual application of EEG personal identification system.