Online Sequential EEG Emotion Recognition with Prototypical Alignment Based Transfer Model.

Area of Science:

• Neuroscience
• Computer Science
• Human-Computer Interaction

Background:

• Deep learning for emotion recognition requires large datasets and struggles with new data.
• Subject-specific training data can inflate model accuracy, lacking generalizability.
• Online learning models need efficient adaptation to new subjects and data.

Purpose of the Study:

• To develop an innovative online sequential electroencephalogram (EEG) emotion recognition method.
• To enhance subject independence and rapid adaptation in emotion recognition models.
• To improve accuracy and simplify network architecture using transfer learning and adversarial networks.

Main Methods:

• Utilized a cross-subject transfer learning model in an online learning environment.
• Implemented selective parameter pruning and reinitialization for rapid subject adaptation.
• Employed an enhanced Domain Adversarial Neural Network (DANN) to align features across emotional categories.

Main Results:

• Achieved average accuracies of 80.04% on the SEED dataset and 62.78% on the SEED-IV dataset.
• Demonstrated superior performance compared to existing mainstream online learning methods.
• Showcased rapid adaptation to new subjects under limited sample conditions with high accuracy.

Conclusions:

• The proposed online sequential EEG emotion recognition method offers practical application potential.
• The approach effectively addresses limitations of traditional deep learning models in emotion recognition.
• This method provides a robust solution for real-time, subject-independent emotion detection.