Motor Imagery EEG Signal Classification Using Deep Learning Approach

Abstract

Motor Imagery electroencephalography (MI EEG) data are employed in brain computer interface (BCI) systems to identify the intention of participants. Several factors, such as poor signal to noise ratios and a scarcity of high-quality samples, complicate the classification of MI EEG signals. For BCI systems to operate well, it is necessary to analyse MI-EEG signals. Recent successful applications of deep learning methods have been observed in pattern recognition and other domains. Conversely, there have been few successful implementations of deep learning algorithms in BCI systems, particularly those based on machine intelligence. Brain-computer interfaces (BCI) can be crucial in facilitating communication with the external environment for those with movement impairments. Deep learning has achieved remarkable success across the many domains. Nevertheless, deep learning has achieved only limited progress in the analysis of Electroencephalogram (EEG) information. The present study suggests a novel approach to address the problem by integrating the Continuous Wavelet Transform (CWT) with deep learning-based transfer learning approach. Continuous Wavelet Transform (CWT) converts one-dimensional EEG signals into a two-dimensional representation of time, frequency, and amplitude images. This allows us to explore existing deep networks via transfer learning. The present work assesses the efficacy of the suggested methodology by utilising a publicly accessible dataset from the BCI competition VI-2b. Our study attained a promising validation accuracy of 81.72% by comparing the findings of the approach with previous efforts on the same dataset. A comparative analysis of the proposed algorithm with existing algorithms demonstrates its superior performance in classification tasks. The approach can enhance the classification accuracy of motor imagery (MI)-based braincomputer interfaces (BCIs) and BCI systems designed for individuals with impairment.