Multi-view domain adaption based multi-scale convolutional conditional invertible discriminator for cross-subject...

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Cross-subject Electroencephalogram (EEG) emotion recognition aims to classify emotions across individuals using neural electrical patterns.
• Existing models face challenges in feature extraction, convergence, and negative transfer, limiting reliable cross-subject emotion detection.

Purpose of the Study:

• To develop a robust model for accurate cross-subject emotion recognition from EEG signals.
• To overcome limitations of existing methodologies in generalization and feature discrimination.

Main Methods:

• Utilized signal preprocessing and Short-Time Geodesic Flow Kernel Fourier Transform (STGFKFT) for feature extraction.
• Employed multi-view sheaf attention for enhanced feature discrimination and a Multi-Scale Convolutional Conditional Invertible Puma Discriminator Neural Network (MSCCIPDNN) for generalization.
• Incorporated efficient computational techniques and the puma optimization algorithm for robustness and convergence.

Main Results:

• The proposed framework achieved high accuracy rates of 99.5% on the SEED dataset, 99% on SEED-IV, and 99.50% on the DEAP dataset.
• Demonstrated superior performance in feature extraction, discrimination, and generalization compared to existing methods.

Conclusions:

• The developed framework shows extraordinary success in cross-subject EEG emotion recognition.
• The integration of advanced techniques effectively captures emotional features and overcomes limitations in current methodologies.