Research On EEG Identity Recognition Technology Based On Deep Learning

With the advent of the digital age,identity recognition is very important in people’s work and life.Compared with traditional identification technology,identification based on biometrics has been widely used because of its high security,convenience and no risk of forgetting;As a new type of biological characteristics,EEG has the advantages of high concealment and difficult to forge.More and more researchers begin to pay attention to this field.This paper mainly studies the tasks of personal identification based on EEG using Physionet dataset,including EEG data preprocessing,EEG data augmentation and identification based on convolutional neural network.The main work of this paper is as follows:(1)A data preprocessing scheme for resting EEG identification is designed.In order to filter the noise except EEG information,FIR filter is used to filter the resting state part of the dataset based on EEG component analysis;In order to obtain the optimal segmentation method,five segmentation methods are designed for comparison;In order to enhance the portability of the acquisition device,channel optimization is carried out,and 16 channels are selected from 64 channels to compare the recognition accuracy.(2)A data augmentation algorithm based on sliding window is proposed.In order to solve the problem of insufficient training samples for EEG identification,the idea of sliding window is used to double the resting state training samples of the dataset.The data augmentation algorithm can improve the accuracy of the classifier by about 3%.(3)An EEG identification algorithm based on convolution neural network is proposed.In this paper,a shallow convolutional neural structure is designed,which achieves 96.13% recognition accuracy on the dataset after data augmentation;This paper also designs a spatial conversion method combining ICA algorithm and neural network back propagation algorithm,which converts the collected EEG signals to the original EEG signals in the brain source space.The experimental results show that the recognition accuracy of 99.32% is achieved after the optimization of spatial conversion,which is better than the highest level of literature.