Dual-Encoder VAE-GAN With Spatiotemporal Features for Emotional EEG Data Augmentation

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Data scarcity is a major challenge in developing accurate electroencephalogram (EEG)-based emotion recognition models.
• Existing deep learning methods struggle with limited datasets, impacting model performance.

Purpose of the Study:

• To address data scarcity in EEG emotion recognition by generating high-quality artificial samples.
• To propose a novel DEVAE-GAN model that incorporates spatiotemporal features for enhanced sample generation.

Main Methods:

• EEG data preprocessed into differential entropy features across five frequency bands.
• Spatiotemporal features extracted using temporal and spatial morphology data.
• A dual encoder architecture trained to generate artificial EEG samples via latent variables.

Main Results:

• The proposed DEVAE-GAN model augmented the SEED dataset with artificial samples.
• Deep neural networks trained on augmented data achieved an average accuracy of 97.21%, a 5% improvement over the original dataset.
• Demonstrated similarity between generated and original data distributions, validating the model's effectiveness.

Conclusions:

• The DEVAE-GAN model effectively generates high-quality artificial EEG samples.
• Generated samples significantly improve the accuracy of emotion recognition models.
• This approach offers a viable solution for data scarcity in affective computing.