Robust EEG-based Emotion Recognition Using an Inception and Two-sided Perturbation Model.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Automated emotion recognition using electroencephalogram (EEG) signals is crucial for brain-computer interfaces.
• Deep learning models for EEG-based emotion recognition are vulnerable to environmental noise and adversarial attacks.
• Enhancing model resilience against input perturbations is essential for reliable emotion recognition.

Purpose of the Study:

• To propose and validate the Inception feature generator and two-sided perturbation (INC-TSP) approach for robust EEG-based emotion recognition.
• To improve the elasticity of deep learning models against adversarial attacks and input uncertainties.
• To address the challenge of maintaining accurate emotion recognition in noisy or manipulated EEG data.

Main Methods:

• Developed an INC-TSP approach integrating an Inception module for EEG feature extraction.
• Implemented two-sided perturbation (TSP) to introduce worst-case perturbations to model weights and inputs.
• Validated the approach in a subject-independent, three-class emotion recognition task.

Main Results:

• The INC-TSP approach demonstrated robust performance in subject-independent emotion recognition.
• The proposed method enhances the model's elasticity against adversarial attacks and input perturbations.
• Maintained accurate emotion recognition despite input uncertainties.

Conclusions:

• The INC-TSP approach offers a promising solution for enhancing the robustness of EEG-based emotion recognition systems.
• This method effectively mitigates the impact of noise and adversarial attacks on deep learning models.
• The findings support the development of more reliable and secure brain-computer interfaces.