Conditional probabilistic-based domain adaptation for cross-subject EEG-based emotion recognition.

Area of Science:

• Affective computing
• Neuroscience
• Machine learning

Background:

• Electroencephalogram (EEG)-based emotion recognition is crucial in affective computing.
• EEG signals are non-stationary and non-linear, leading to significant individual differences.
• Existing domain adaptation methods struggle with individual-dependent background signals, causing classification confusion.

Purpose of the Study:

• To propose a novel conditional probabilistic-based domain adversarial network (CPDAN) for cross-subject EEG-based emotion recognition.
• To address the limitations of existing methods in handling individual-dependent background signals.
• To improve the accuracy and robustness of EEG emotion recognition across different subjects.

Main Methods:

• CPDAN utilizes separate branch networks to distinguish background and task-specific emotional features from EEG signals.
• Employs domain-adversarial training to minimize global and local domain discrepancies.
• Reduces intra-class distance and enlarges inter-class distance for better feature representation.

Main Results:

• CPDAN framework demonstrates superior performance over comparison methods on SEED and SEED-IV datasets.
• Achieved a significant average accuracy improvement of 22% on the SEED-IV dataset compared to existing methods.
• Effectively mitigates the impact of individual differences in EEG signals for emotion recognition.

Conclusions:

• The proposed CPDAN framework offers an effective solution for cross-subject EEG-based emotion recognition.
• Separating background and task features is critical for improving recognition accuracy.
• CPDAN advances the field of affective computing by enhancing the reliability of EEG emotion recognition systems.