Enhanced EEG Emotion Recognition Using MIMO-Based Denoising and Band-Wise Attention Graph Neural Network.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalogram (EEG) signals are crucial for brain-computer interface (BCI) systems, particularly for emotion recognition.
• Noise in EEG signals significantly degrades the performance of emotion recognition systems.
• Existing methods like Band Feature Extraction Neural Network (BFE-Net) have limitations in band-wise feature extraction and aggregation.

Purpose of the Study:

• To propose a noise-robust band-attention BFE-Net framework to enhance EEG-based emotion recognition.
• To improve upon the conventional BFE-Net by addressing noise contamination and feature extraction limitations.

Main Methods:

• Implemented multiple-input, multiple-output (MIMO)-based preprocessing using multichannel minima-controlled recursive averaging and generalized eigenvalue decomposition for noise reduction.
• Introduced an attention-based band aggregation mechanism with a band-wise self-attention model for sophisticated feature fusion.

Main Results:

• The proposed model demonstrated improved performance on the SEED and SEED-IV datasets under a subject-independent protocol.
• Outperformed the state-of-the-art BFE-Net by 3.27% and 3.34% on the respective datasets.
• Validated the effectiveness of MIMO noise reduction and frequency-centric attention in enhancing BCI system reliability and generalization.

Conclusions:

• Rigorous MIMO noise reduction techniques significantly improve EEG signal quality for BCI applications.
• Frequency-centric attention mechanisms enhance feature fusion, leading to more accurate emotion recognition.
• The proposed noise-robust band-attention BFE-Net framework offers a more reliable and generalizable solution for BCI systems.