Related Experiment Video
Updated: Jul 17, 2026
08:45
Mapping Cortical Dynamics Using Simultaneous MEG/EEG and Anatomically-constrained Minimum-norm Estimates: an Auditory Attention Example
Published on: October 24, 2012
MSBiLSTM-Attention: EEG Emotion Recognition Model Based on Spatiotemporal Feature Fusion
Yahong Ma1, Zhentao Huang1, Yuyao Yang1
1Xi'an Key Laboratory of High Precision Industrial Intelligent Vision Measurement Technology, School of Electronic Information, Xijing University, Xi'an 710123, China.
Biomimetics (Basel, Switzerland)
|March 26, 2025
Summary
This study introduces a novel deep learning model for accurate emotion recognition from electroencephalogram (EEG) signals. The MSBiLSTM-Attention model automates feature extraction and classification, achieving high accuracy in emotion analysis.
Area of Science:
- Artificial Intelligence
- Neuroscience
- Human-Computer Interaction
Background:
- Emotional states significantly influence decision-making and social interactions.
- Sentiment analysis is crucial for human-computer emotional engagement, driving AI research.
- EEG-based emotion analysis faces challenges in feature extraction and classifier design.
Purpose of the Study:
- To develop a novel deep learning technique for automatic EEG feature extraction and classification.
- To improve the accuracy and convenience of EEG-based emotion recognition.
- To address limitations of manual preprocessing in existing deep learning approaches.
Main Methods:
- A novel deep learning model integrating multi-scale convolution and bidirectional long short-term memory networks with an attention mechanism (MSBiLSTM-Attention).
- Automatic extraction and merging of spatiotemporal features from raw EEG data.
- Classification of emotional EEG signals using a fully connected layer after attention-based feature selection.
Main Results:
- The MSBiLSTM-Attention model achieved high classification accuracies on the SEED dataset.
- Single validation accuracies reached 99.44% for three-class and 99.85% for four-class emotion tasks.
- Average 10-fold cross-validation accuracies were 99.49% (three-class) and 99.70% (four-class).
Conclusions:
- The proposed MSBiLSTM-Attention model demonstrates effectiveness in EEG-based emotion recognition.
- This approach offers a powerful solution for automated emotion analysis from neural signals.
- The findings highlight the potential of deep learning with attention mechanisms for advancing affective computing.
