Related Experiment Video
Updated: Aug 13, 2025
Mapping Cortical Dynamics Using Simultaneous MEG/EEG and Anatomically-constrained Minimum-norm Estimates: an Auditory Attention Example
Published on: October 24, 2012
Temporal relative transformer encoding cooperating with channel attention for EEG emotion analysis
Guoqin Peng1, Kunyuan Zhao2, Hao Zhang2
1Yunnan University, Kunming, 650500, China; Department of Psychology and Behavioral Sciences, Zhejiang University, Hangzhou, 310013, China; Department of Psychiatry of Sir Run Run Shaw Hospital, Zhejiang University School of Medicine, Hangzhou, 310013, China.
This study introduces novel temporal relative (TR) and channel-attention (CA) mechanisms for electroencephalogram (EEG) emotion recognition. The proposed model effectively utilizes both temporal and spatial EEG data, outperforming existing methods.
Area of Science:
- Neuroscience
- Computer Science
- Artificial Intelligence
Background:
- Electroencephalogram (EEG)-based emotion computing is a growing field within brain-computer interfaces.
- Existing methods often fail to fully leverage the inherent temporal and spatial characteristics of EEG signals.
- Standard transformer position encoding struggles with continuous temporal data like emotions.
Purpose of the Study:
- To propose a novel approach for EEG emotion recognition that simultaneously utilizes temporal and spatial information.
- To introduce a temporal relative (TR) encoding mechanism for better temporal feature extraction in transformers.
- To present a channel-attention (CA) mechanism to weigh the importance of different EEG channels.
Main Methods:
- A novel temporal relative (TR) encoding mechanism was developed to capture the continuous temporal nature of EEG signals.
- A channel-attention (CA) mechanism was introduced to analyze the contribution of individual EEG channels.
- The TR and CA mechanisms were integrated into a transformer architecture for simultaneous temporal and spatial feature utilization.
- Experiments were conducted on the DEAP dataset for binary and 5-class discrete emotion classification.
Main Results:
- The proposed model demonstrated superior performance in binary classification tasks for valence, arousal, dominance, and liking.
- The model also achieved state-of-the-art results in the 5-class discrete emotion category classification task.
- Experimental results confirmed the effectiveness of the combined temporal and channel attention mechanisms.
Conclusions:
- The developed model effectively integrates temporal and spatial information from EEG signals for enhanced emotion recognition.
- The novel TR and CA mechanisms significantly improve the performance of EEG-based emotion computing.
- This approach offers a promising advancement in the field of brain-computer interfaces and affective computing.
