Cross-Session Emotion Recognition by Joint Label-Common and Label-Specific EEG Features Exploration.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) is a robust, objective measure for emotion recognition due to its resistance to camouflage.
• EEG signals exhibit multi-rhythm and multi-channel properties, allowing for diverse feature extraction.
• Existing studies often overlook the critical task of distinguishing common versus specific features across emotional states.

Purpose of the Study:

• To propose a novel model for exploring both common and specific features in EEG signals for emotion recognition.
• To address the challenge of semi-supervised, cross-session EEG emotion recognition.
• To provide a quantitative method for identifying label-common and label-specific EEG features.

Main Methods:

• Developed a Joint label-Common and label-Specific Features Exploration (JCSFE) model.
• Employed the $\ell _{2,1}$-norm to extract label-common EEG features.
• Utilized the $\ell _{1}$-norm for extracting label-specific EEG features.
• Incorporated graph regularization to enforce local invariance in data.

Main Results:

• The JCSFE model achieved superior emotion recognition performance compared to state-of-the-art methods on SEED-IV and SEED-V datasets.
• Demonstrated the model's effectiveness in semi-supervised, cross-session EEG emotion recognition.
• Successfully provided a quantitative approach to differentiate common and specific EEG features.

Conclusions:

• The JCSFE model offers a significant advancement in EEG-based emotion recognition by dissecting feature types.
• This approach enhances understanding of neural correlates of emotions.
• The method is validated for its effectiveness and quantitative insights into EEG features.