Directional Spatial and Spectral Attention Network (DSSA Net) for EEG-based emotion recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion recognition from Electroencephalography (EEG) signals has advanced, but modeling complex spatial, spectral, and temporal features remains challenging.
• Accurate interpretation of multi-channel brain signals is crucial for understanding emotional states.

Purpose of the Study:

• To propose a novel framework, the Directional Spatial and Spectral Attention Network (DSSA Net), for enhanced emotion recognition from EEG signals.
• To effectively capture and model critical spatial-spectral-temporal features inherent in EEG data.

Main Methods:

• The proposed DSSA Net framework integrates three attention modules: Positional Attention (PA), Spectral Attention (SA), and Temporal Attention (TA).
• The PA module utilizes Vertical Attention (VA) and Horizontal Attention (HA) branches to identify active brain regions from various orientations.
• The framework processes multi-channel EEG signals to extract comprehensive feature representations.

Main Results:

• DSSA Net demonstrated superior performance compared to existing methods on three benchmark EEG datasets (SEED, SEED-IV, DEAP).
• Subject-dependent emotion recognition accuracies reached 96.61% (SEED) and 85.07% (SEED-IV).
• Subject-independent recognition achieved 87.03% (SEED) and 75.86% (SEED-IV), with valence and arousal accuracies of 94.97% and 94.73% on DEAP.

Conclusions:

• The DSSA Net framework effectively leverages spatial and spectral differences across brain regions and hemispheres.
• The proposed method significantly enhances classification accuracy for EEG-based emotion recognition.
• DSSA Net offers a promising approach for advanced brain-computer interfaces and affective computing.