Classification of Raw EEG Motor Imagery Signals with Improved Convolutional Neural Networks for Use in Brain-Computer Interfaces

Abstract

The EEG motor imagery-based brain-computer interface (BCI) has drawn an interest from neuro-engineering researchers and is currently being utilized in various rehabilitative contexts. The performance of BCI systems may be negatively impacted by EEG motor imagery with a very low signal to noise ratio. This paper presents a novel technique for classifying EEG motor imagery, which is based on a continuous neural network and an improved convolutional neural network. The utilization of a continuous wavelet transforms incorporating three distinct mother wavelets enables the capture of a comprehensive EEG image through the amalgamation of time-frequency and electrode location. The purpose of this work is to reduce the computational complexity of a convolutional neural network while increasing its recognition of motor imagery tasks. The BCI competition IV dataset 2b was used to verify the suggested approach. The findings prove that the suggested methods outperform the current ones in terms of classification performance, making a brain-computer interface based on motor imagery a real possibility.