A Data Augmentation Method for Motor Imagery EEG Signals Based on DCGAN-GP Network.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) signals are crucial for brain-computer interface (BCI) research, particularly in motor rehabilitation and control.
• Limited labeled EEG data hinders the development of robust BCI classifiers.
• Existing data augmentation techniques may not fully capture the complexity of EEG signals.

Purpose of the Study:

• To propose and evaluate a novel data augmentation method for EEG signals using an improved Deep Convolutional Generative Adversarial Network with Gradient Penalty (DCGAN-GP).
• To address the challenge of limited labeled data in BCI research.
• To enhance the robustness and accuracy of BCI classifiers for motor imagery tasks.

Main Methods:

• Raw EEG signals were transformed into two-dimensional time-frequency maps.
• A DCGAN-GP network was employed to generate synthetic time-frequency representations of EEG data.
• Classifiers were trained using both augmented (synthetic) and unaugmented (real) data on the BCI IV 2b dataset.

Main Results:

• Classifiers trained with DCGAN-GP-generated synthetic EEG data demonstrated enhanced robustness across multiple subjects.
• The use of synthetic data led to higher classification accuracy in distinguishing motor imagery tasks.
• The proposed method effectively generated realistic time-frequency representations resembling real EEG data.

Conclusions:

• The DCGAN-GP-based data augmentation method is effective in overcoming EEG data scarcity for BCI applications.
• Synthetic EEG data generated by DCGAN-GP significantly improves classifier performance and robustness.
• This approach offers a promising solution for advancing BCI technology and its real-world applications.