Multi-Channel Convolutional Neural Networks Architecture Feeding for Effective EEG Mental Tasks Classification

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Classifying mental tasks using electroencephalography (EEG) signals is challenging due to noise and spatial instability.
• Current state-of-the-art methods struggle with the inherent complexities of EEG data.

Purpose of the Study:

• To develop a robust multi-channel convolutional neural network (CNN) architecture for improved mental task classification.
• To address the limitations of existing methods in handling noisy and spatially unstable EEG signals.

Main Methods:

• A multi-channel CNN with adaptively optimized parameters was designed.
• The architecture utilizes a frequency-domain feeding scheme, analyzing EEG signal frequency sub-bands.
• Feature mapping is supported by two convolutional layers and a fully connected layer.

Main Results:

• The proposed method achieved nearly 70% classification accuracy on Dataset V of the BCI Competition III.
• It demonstrated superior performance compared to alternative methods, with a 1.2% improvement in accuracy.
• The solution exhibited high generalization capability (∼5%) and outperformed established networks like AlexNet and VGG-16.

Conclusions:

• The developed frequency-domain multi-channel CNN offers a significant advancement in EEG-based mental task classification.
• The approach provides higher accuracy and better generalization than current state-of-the-art solutions.
• Adaptive parameter optimization tailored to signal characteristics is key to the method's success.