Spatial-temporal network for fine-grained-level emotion EEG recognition.

Area of Science:

• Affective computing
• Brain-computer interfaces
• Neuroscience

Background:

• Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) are advancing machine understanding of human intentions.
• Current applications often overlook the nuances of short-term emotional states, termed fine-grained emotions.

Purpose of the Study:

• To develop a novel dataset for fine-grained emotion recognition.
• To propose an advanced neural network architecture for accurate emotion classification from EEG signals.

Main Methods:

• Construction of a unique EEG dataset encompassing two coarse-grained and four fine-grained emotions.
• Development of FG-emotionNet, a deep learning model for spatial-temporal feature extraction from raw EEG signals.
• Integration of feature fusion techniques and Long Short-Term Memory (LSTM) for enhanced temporal analysis and classification.

Main Results:

• The proposed FG-emotionNet demonstrated superior performance compared to existing methods in emotion recognition tasks.
• Subject-dependent and cross-session experiments validated the effectiveness of the developed model.
• The network effectively extracts spatial-temporal features, mitigating overfitting and improving classification accuracy.

Conclusions:

• The developed fine-grained emotion EEG dataset and FG-emotionNet provide a significant advancement in EEG-based emotion recognition.
• The proposed methodology offers a robust approach for understanding and classifying nuanced emotional states.
• This research paves the way for more sophisticated affective computing applications.