Related Experiment Video
Updated: Nov 1, 2025
08:31
Conscious and Non-conscious Representations of Emotional Faces in Asperger's Syndrome
Published on: July 31, 2016
13.9K
Cross-Subject EEG Emotion Recognition With Self-Organized Graph Neural Network
Jingcong Li1,2, Shuqi Li3, Jiahui Pan1,2
1School of Software, South China Normal University, Guangzhou, China.
Frontiers in Neuroscience
|June 28, 2021
Summary
This study introduces a novel Self-Organized Graph Neural Network (SOGNN) for improved cross-subject emotion recognition using electroencephalogram (EEG) brain signals. The SOGNN dynamically constructs graph structures, achieving state-of-the-art performance in recognizing emotions across individuals.
Area of Science:
- Neuroscience
- Cognitive Science
- Machine Learning
Background:
- Emotion recognition from brain signals is crucial but challenging due to individual differences and low signal-to-noise ratio in EEG.
- Conventional methods struggle with cross-subject emotion recognition accuracy.
Purpose of the Study:
- To propose a novel Self-Organized Graph Neural Network (SOGNN) for enhanced cross-subject EEG emotion recognition.
- To dynamically construct graph structures for improved EEG signal analysis.
Main Methods:
- Developed a Self-Organized Graph Neural Network (SOGNN) with a dynamic graph construction module.
- Conducted leave-one-subject-out experiments on SEED and SEED-IV datasets.
- Investigated performance variations across different graph construction techniques and frequency bands.
Main Results:
- The SOGNN achieved state-of-the-art performance in cross-subject EEG emotion recognition.
- Visualized graph structures revealed insights consistent with neuroscience research.
- Demonstrated the model's effectiveness and adaptability across datasets and features.
Conclusions:
- The proposed SOGNN model significantly improves cross-subject EEG emotion recognition.
- Dynamic graph construction is effective for handling individual differences in EEG signals.
- The SOGNN offers a promising approach for advanced brain-computer interfaces and affective computing.
