EEG-based human recognition using Hjorth-parameters and LSTM technique

Document Type : Original Research Articles.

Authors

Department of Information System, Faculty of Computers and Information, Mansoura University, Egypt

Abstract

 There is increasing interest in assessing the feasibility of using Electroencephalography (EEG) signals in biometric purposes. Using deep learning techniques has achieved great performance in classification-based systems in general. However using them in EEG–based human recognition systems still limited, this was the main motivation which encouraged the authors to investigate using of these techniques in EEG-based human recognition system. In this paper, the authors suggested a framework that uses the three Hjorth parameters to enhance the Long-Short Term Memory (LSTM) performance for Electroencephalography (EEG)-based human recognition systems. The proposed framework also investigates the ability to optimize two critical factors of EEG-based biometrics, which are the number of channels and the time needed for acquiring data. The proposed approach has been tested on a public data set, which is the public Texas data repository to verify the improvement of recognition and its reliability through the data recording duration of the eight minutes. The study evaluates two optimizers, namely, Stochastic Gradient Descent optimizer and Conjugate Gradient Descent. The results show a significant improvement in the LSTM performance using the proposed framework, by applying the fusion of features with the Hjorth parameters and using Conjugate Gradient Descent optimizer (CGD).

Keywords

Files

Share

How to cite

Statistics