Classification of EEG Motor Imagery Tasks Using Convolution Neural Networks.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalograph (EEG) signals are complex and nonlinear, posing classification challenges.
• EEG is crucial for Brain-Computer Interfaces (BCI), enabling control for assistive devices like powered wheelchairs.
• Motor Imagery (MI) tasks are commonly used to generate EEG signals for BCI applications.

Purpose of the Study:

• To classify Electroencephalograph (EEG) signals from subjects performing four Motoric Imagery (MI) tasks.
• To evaluate the effectiveness of a Convolutional Neural Network (CNN) for EEG signal classification.
• To compare the performance of layered versus stacked input data configurations for EEG classification.

Main Methods:

• EEG datasets were acquired from subjects trained on four MI tasks.
• Signals were denoised using Bump Continuous Wavelet Transform (CWT) within the 8-32 Hz brainwave range.
• A CNN architecture, based on the Visual Geometry Group (VGG-16) network, was employed for classification, utilizing all 22 electrodes (10-20 system placement).
• EEG data was transformed into scaleograms using CWT.

Main Results:

• The study evaluated two input data configurations: layered and stacked.
• The layered image dataset configuration achieved a higher classification accuracy.
• An average accuracy of 68.33% was obtained for the two-class classification using the layered approach.

Conclusions:

• Layered image representation of EEG data, when processed by a VGG-16 based CNN, shows promise for classifying motor imagery tasks.
• This approach offers a potential improvement in EEG signal classification accuracy for BCI applications.
• Further research can explore optimizing CNN architectures and data preprocessing techniques for enhanced BCI performance.