Motor Imagery-Based Brain-Computer Interface Using Fusion of Deep Convolutional Neural Network with Wavelet Scattering Network

Abstract

Brain-Computer Interface (BCI) is a system which is used to interact with the computer system and can be used to control different assistive devices by utilizing the brain signals such as Electroencephalography (EEG). Motor Imagery (MI) is considered one of the prominent fields of BCI systems which are based on EEG signals. These systems have the capability to restore motor ability in humans. Though a good deal of Machine Learning (ML) approaches were investigated in recent years, studies that explore BCI with Deep Learning methods or Wavelet Scattering Transforms have not been extensively used. Also, conventional classification methods show longer computational time and they are incapable of processing non-linear and non-stationary EEG signals. The proposed system aims to explore the area of a calibration-free or a subject-independent model by integrating Deep Convolutional Neural Network (CNN) with Wavelet Scattering Network by utilizing feature maps learned from both CNN and Scattering networks to tackle the non-linearity and non-stationarity of the EEG signals and thereby to improve the classification accuracy of the model to build a more robust and generalized MI-based BCI system. The proposed model outperforms the state-of-the-art techniques, achieving an accuracy of 87% and 93% on BCIC IV 2a and SMR-BCI datasets respectively.