An Improved Convolutional Neural Network Framework For Efficient Motor Imegary Electroencepholography (EEG) Classification Using Common Spatial Pattern (CSP)

Abstract

This study develops an optimized Convolutional Neural Network (CNN) architecture based on the Common Spatial Pattern (CSP) method to improve the classification performance of motor-imagery-based brain-computer interface (BCI) systems. The use of brain activity is to communicate and control external devices by people with motor limitations has demonstrated encouraging outcomes for motor-imagery-based BCIs. However, the electroencephalography (EEG) signals can be difficult to extract pertinent information from, which limits the classification accuracy of these systems. The CSP algorithm, which successfully extracts discriminative features from multi-channel EEG signals, is combined in this study's unique approach with a CNN framework that makes better use of the learnt spatial filters. The preprocessing of the EEG data, use of the CSP algorithm to extract pertinent features, and training of the CNN with an optimized architecture make up the technique. Using publicly accessible benchmark datasets, the performance of the proposed strategy was assessed and compared to current state-of-the-art techniques (DNN). Improvements in classification accuracy (i.e., 98.80% was achieved against the 88.10%) of the existing approach (DNN-FBCSP), decreased computing complexity, and greater usability of motor-imagery-based BCIs are predicted consequences of this research, which was ultimately developed assistive technologies for people with motor disorders.