Related Experiment Video
Updated: Jun 18, 2026
Applications of EEG Neuroimaging Data: Event-related Potentials, Spectral Power, and Multiscale Entropy
Published on: June 27, 2013
MS-STGAN: A dual-branch multi-scale spatio-temporal generative adversarial framework for incomplete EEG-based emotion
Cheng Cheng1, Jingjing Zhang1, Yikang Cheng2
1Institute of Psychological and Brain Sciences, Liaoning Normal University, Dalian, 116029, China; Key Laboratory of Brain and Cognitive Neuroscience, Liaoning Province, Dalian, 116029, China.
This study introduces a novel Multi-Scale Spatio-Temporal Generative Adversarial Network (MS-STGAN) for robust emotion recognition from incomplete electroencephalography (EEG) data. The method effectively reconstructs missing EEG signals, improving accuracy in real-world applications.
Area of Science:
- Neuroscience
- Machine Learning
- Signal Processing
Background:
- Electroencephalography (EEG) offers high-resolution emotion recognition.
- Real-world EEG data often suffers from incompleteness due to sensor failures or preprocessing errors.
- Existing methods struggle with reconstructing reliable EEG data.
Purpose of the Study:
- To propose a novel generative adversarial network for reconstructing complete EEG data from incomplete inputs.
- To enhance emotion recognition accuracy in the presence of missing EEG data.
- To validate the proposed method on benchmark datasets.
Main Methods:
- A Multi-Scale Spatio-Temporal Generative Adversarial Network (MS-STGAN) was developed.
- A spatio-temporal dual-branch generator utilizing Graph Convolutional Networks (GCNs) and BiMamba was designed.
- Multi-scale 2D convolution and deconvolution layers were incorporated for diverse feature extraction.
- A generative adversarial framework with a discriminator was employed to ensure data authenticity.
Main Results:
- The MS-STGAN effectively reconstructed complete EEG representations from masked data.
- The spatial branch captured inter-regional dependencies, while the temporal branch encoded emotional dynamics.
- Experiments on DEAP and SEED-IV datasets demonstrated superior performance and generalization ability.
- Each component of the MS-STGAN was validated for its effectiveness.
Conclusions:
- The proposed MS-STGAN significantly improves emotion recognition from incomplete EEG data.
- The network architecture effectively handles missing data by reconstructing spatio-temporal features.
- MS-STGAN shows strong potential for real-world emotion recognition applications.
