HS-CNN: a CNN with hybrid convolution scale for EEG motor imagery classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) motor imagery classification is crucial for assistive technologies and neurorehabilitation.
• Convolutional Neural Networks (CNNs) show promise but are limited by single convolution scales and data scarcity.
• Subject-specific optimal convolution scales and limited training data hinder current CNN-based EEG classification accuracy.

Purpose of the Study:

• To develop an improved CNN architecture for EEG motor imagery classification.
• To address limitations of single-scale CNNs and data augmentation challenges.
• To enhance classification accuracy in EEG-based applications.

Main Methods:

• Proposed a hybrid-scale CNN architecture incorporating multiple convolution scales.
• Implemented a data augmentation technique to address limited training data.
• Evaluated the method on two standard EEG motor imagery datasets.

Main Results:

• Achieved an average classification accuracy of 91.57% on one dataset.
• Obtained an average classification accuracy of 87.6% on a second dataset.
• Outperformed several existing state-of-the-art EEG motor imagery classification methods.

Conclusions:

• The hybrid-scale CNN with data augmentation effectively overcomes limitations of existing methods.
• The proposed approach significantly improves EEG motor imagery classification accuracy.
• This advancement holds potential for more robust brain-computer interfaces and neurorehabilitation tools.