EEGGAN-Net: enhancing EEG signal classification through data augmentation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interface (BCI) technology offers significant potential for improving the quality of life for individuals with disabilities.
• Current limitations in electroencephalography (EEG) signal classification accuracy hinder the widespread adoption of BCI systems in real-world applications.

Purpose of the Study:

• To develop a novel EEG signal classification model, EEGGAN-Net, to overcome accuracy limitations in BCI applications.
• To enhance the classification efficacy of EEG signals by integrating advanced data augmentation and feature extraction techniques.

Main Methods:

• Implemented a novel EEG signal classification model, EEGGAN-Net, incorporating Conditional Generative Adversarial Network (CGAN) data augmentation.
• Utilized a cropped training strategy and a Squeeze-and-Excitation (SE) attention mechanism to improve feature assimilation.
• Evaluated the model's performance on the BCI Competition IV-2a and IV-2b datasets.

Main Results:

• EEGGAN-Net achieved a classification accuracy of 81.3% (kappa=0.751) on the BCI Competition IV-2a dataset.
• The model attained 90.3% classification accuracy (kappa=0.79) on the BCI Competition IV-2b dataset.
• Performance surpassed four other Convolutional Neural Network (CNN)-based decoding models.

Conclusions:

• The combination of data augmentation and attention mechanisms is crucial for extracting generalized features from EEG signals.
• EEGGAN-Net significantly enhances the overall proficiency of EEG signal classification for BCI applications.
• This approach holds promise for advancing the capabilities of assistive technologies for individuals with disabilities.