Exploring EEG based Authentication for Imaginary and Non-imaginary tasks using Power Spectral Density Method

1st International Conference on Mechanical Electronic and Biosystem Engineering

Zhi Chin, Tze^1 ; Saidatul, A.^1 ; Ibrahim, Z.^2

Biosignal Processing Research Group (BioSIM), School of Mechatronic Engineering, Universiti Malaysia Perlis, Perlis, Arau, Malaysia^1

Faculty of Technology, University of Sunderland, St Peter's Campus, SR6 0DD, Sunderland, United Kingdom^2

关键词: Biometric technology;    Classification accuracy;    Human authentication;    K-nearest neighbours;    Linear discriminant analysis;    Power spectral density method;    Statistical features;    Undergraduate students;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/557/1/012031/pdf
DOI  :  10.1088/1757-899X/557/1/012031

来源: IOP

【 摘 要 】

Biometric technology has swiftly emerged as a go-to solution for improving cyber security especially in financial fraud and security threats. EEG based-authentication is best of security in cyber security application as it is unique and cannot be replicated. The aim of this study is to investigate the possibility of adopting imaginary or non-imaginary task for human authentication. In this study, twenty subjects were recruited from undergraduate students with age ranging from 19 to 30 years old. The subject must be healthy and right-handed. The subjects were required to perform non-imaginary task (left hand or right hand movement) and imaginary task (just need to imagine the movement of left hand or right hand). Duration for each task is 1 minute and take rest for 1 minute between the tasks. Truscan EEG device (Deymed Diagnostic, Alien Technic, Czech Republic) with 19 channels were used to collect EEG data with 1024 Hz frequency sampling and the impedance is kept below 5 kOhm. Bandpass filter was used in pre-processing to extract alpha (8-13Hz) and beta (14-30Hz) waves. The signal was segmented and the power spectral density were calculated by Welch's method and Burg's method. The statistical features (mean, median, mode, variance, standard deviation, minimum and maximum) were obtained from PSD were used as input of classifier. K-nearest neighbour classifier (KNN) and Linear Discriminant Analysis (LDA) were applied for classification. In conclusion, Welch method gives the highest classification accuracy which is 98% for beta waves from channel C4 with the K-nearest neighbour classifier. Imaginary task shows the higher classification accuracy which is 98.03% instead of non-imaginary task which is 94.95%. Thus, imaginary task is more suitable for authentication.