Related Experiment Video
Updated: May 21, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
2.5K
EEG-based emotion recognition with autoencoder feature fusion and MSC-TimesNet model
Jibin Yin1, Zhijian Qiao1, Luyao Han2
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, China.
Computer Methods in Biomechanics and Biomedical Engineering
|March 17, 2025
Summary
This study introduces an advanced deep learning method for emotion recognition using electroencephalography (EEG) signals. The approach significantly improves classification accuracy by fusing features and employing a novel multi-scale convolutional neural network.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Signal Processing
Background:
- Electroencephalography (EEG) signals are valuable for emotion recognition due to their real-time nature and artifact resilience.
- Current EEG-based emotion recognition methods struggle with high-dimensional data and temporal dynamics, limiting performance.
Purpose of the Study:
- To develop a novel deep learning framework for enhanced EEG-based emotion recognition.
- To address limitations in feature integration and temporal pattern analysis in existing methods.
Main Methods:
- EEG signals were segmented and processed to extract Power Spectral Density (PSD) and Differential Entropy (DE) features.
- An autoencoder was utilized for feature fusion, enhancing representation before input into the MSC-TimesNet model.
- The proposed MSC-TimesNet model, incorporating multi-scale convolutional kernels, efficiently processed fused features to capture inter- and intra-period information.
Main Results:
- The proposed AEF-DL method achieved high classification accuracies on DEAP and Dreamer datasets.
- Dependent subject experiments yielded accuracies of 98.97% and 95.71%.
- Independent subject experiments demonstrated accuracies of 97.23% and 92.95%, surpassing existing approaches.
Conclusions:
- The AEF-DL method, integrating autoencoder fusion features and MSC-TimesNet, shows superior performance in EEG-based emotion recognition.
- The approach effectively handles high-dimensional features and complex temporal patterns, indicating broad applicability.
