MES-CTNet: A Novel Capsule Transformer Network Base on a Multi-Domain Feature Map for Electroencephalogram-Based...

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) based emotion recognition is crucial for human-computer interaction.
• Traditional methods struggle to effectively integrate multi-domain features from EEG signals.
• A need exists for advanced models that can capture complex spatio-temporal-frequency EEG patterns.

Purpose of the Study:

• To propose a novel deep learning architecture, MES-CTNet, for enhanced EEG-based emotion recognition.
• To leverage multi-domain features by combining spatial, frequency, and temporal characteristics.
• To improve the accuracy and robustness of emotion recognition from EEG data.

Main Methods:

• Developed a novel capsule Transformer network (MES-CTNet) incorporating Efficient Channel Attention (ECA) and Squeeze and Excitation (SE) blocks within a multichannel capsule neural network (CapsNet).
• Constructed a multi-domain feature map by integrating space-frequency-time characteristics of EEG signals.
• Employed a Transformer-based temporal coding layer for global perception of continuous emotion features.

Main Results:

• Achieved high accuracy on the DEAP dataset: 98.31% for valence and 98.28% for arousal.
• Demonstrated superior performance on the SEED dataset with 94.91% accuracy for cross-session tasks.
• Outperformed traditional EEG emotion recognition methods in experimental evaluations.

Conclusions:

• The proposed MES-CTNet effectively extracts and integrates multi-domain EEG features for superior emotion recognition.
• The novel architecture offers a broader observational perspective, significantly enhancing classification accuracy.
• MES-CTNet holds substantial theoretical and practical value for advancing EEG-based emotion recognition applications.