Exploration And Research Of EEG-Based Identification Using Deep Learning

With the development of the Internet finance and other industries,personal biometrics have been widely used to identify the individual identity of organisms.At present,features of the face and fingerprint are included in the most frequently used biometric features.Electroencephalogram(EEG),also as a unique identity feature,has been paid increasing attention by researchers because of its advantage of not being imitated and reproduced.However,there are still a great number of key problems to be solved in individual identity recognition based on EEG.For example,the acquisition status of EEG that can be used for identity recognition is unclear.Besides,the traditional data features of EEG are too single as they are the frequency domain features of single band EEG,which are simply classified by machine learning.In terms of the above problems,there is a further exploration and improvement in this research.EEG-based identification using deep learning in this research refers to finding out the differences between subjects from EEG image data for deep learning modeling,so as to identify the subjects.The main tasks and findings of this research are as follows:(1)This research focuses on the wave band characteristics of EEG and the methods of signal analysis;The implementation principle as well as the advantages and disadvantages of different algorithms in machine learning and deep learning are compared and analyzed;And the basic structure and functions of convolutional neural networks in deep learning are summarized;The Alex Net framework of the classical neural network in deep learning is also studied.(2)In this research,the multi-band data of EEG are converted into multi-spectral image data,and the data features of identification of EEG are expanded in a multi-dimensional way,which solves the problem of single data features of traditional identification of EEG.Considering the data set form of the image,EEG-based identification using deep learning adopts VGG-16 neural network-a kind of deep learning framework implemented in Python to automatically extract features of image data of the EEG and realizes the classification and recognition in the end.The highest rate of accuracy of EEG-based identification shown in the research reaches 86.79%,which is 10% higher than that of traditional EEG-based identification.(3)It is found that the effect of identification is not only related to the relevant algorithms and models,but also in connection with the degree of brain load through the research.In order to study the specific relationship between brain load and recognition effect,this research also compares the effect of identification under different brain loads through experiment and data analysis,the results of which shows that the lower the brain load in the data sample is,the higher the accuracy of identification of EEG will be.To some extent,this result provides a reliable basis for the optimal collection of EEG-based identification data in the future.