SECT: A Method of Shifted EEG Channel Transformer for Emotion Recognition.

Area of Science:

• Human-Computer Interaction (HCI)
• Neuroscience
• Signal Processing

Background:

• Electroencephalographic (EEG) emotion recognition is crucial for HCI.
• Existing EEG emotion datasets lack diversity, primarily using data from unimpaired subjects.
• There is a need for diverse datasets, including those from hearing-impaired individuals.

Purpose of the Study:

• To collect and analyze EEG signals from 30 hearing-impaired subjects experiencing six distinct emotions.
• To develop and evaluate a novel deep learning model, the Shifted EEG Channel Transformer (SECT), for enhanced EEG emotion recognition.
• To improve the diversity and applicability of EEG-based emotion recognition systems.

Main Methods:

• Collected EEG data from 30 hearing-impaired subjects watching emotion-inducing video clips (happiness, inspiration, neutral, anger, fear, sadness).
• Utilized frequency domain features: Power Spectral Density (PSD) and Differential Entropy (DE), up-sampled via cubic spline interpolation.
• Proposed the Shifted EEG Channel Transformer (SECT) model with two layers: Channel Transformer (CT) for global brain regions and Shifted Channel Transformer (S-CT) for localized regions.

Main Results:

• Subject-dependent experiments achieved accuracies of 82.51% (PSD) and 84.76% (DE) for six-emotion classification.
• Subject-independent experiments on public datasets yielded 85.43% (SEED, 3-class), 66.83% (DEAP, Valence 2-class), and 65.31% (DEAP, Arousal 2-class).

Conclusions:

• The SECT model demonstrates significant potential for accurate EEG emotion recognition, particularly in diverse populations like the hearing-impaired.
• The proposed method effectively captures both global and localized brain region information for improved emotion classification.
• This research contributes to more inclusive and robust HCI applications through enhanced EEG-based emotion recognition.