.png){kind=icon}
.png){kind=icon}
Return
A Gradient-Optimized Dual-Branch Network for EEG Emotion Recognition in Hearing and Hearing-Impaired Individuals
.png){kind=icon}
PRE
AI
W
N
J
Y
Z
Y
DOI：10.1109/jsen.2026.3708116.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Electroencephalogram (EEG) is widely employed in emotion recognition due to its direct reflection of neural activity. However, its nonstationary nature and noise interference pose challenges for feature extraction and model generalization. To address this, this article proposes a gradient-optimized dual-branch network (GDBN), comprising spatio-temporal local feature extraction (ST-LFE) and spatio-temporal global feature extraction (ST-GFE). ST-LFE employs multiscale convolutions to capture dynamic temporal sequences and local brain region features, making it suitable for modeling the relatively distinct emotion-related activation patterns observed in hearing individuals. ST-GFE employs global convolutions and an adaptive weighting mechanism to enhance key features, improving modeling capabilities for cross-region and irregular activations. To mitigate training instability caused by inconsistent gradient update directions between branches, a condition number-constrained gradient optimization strategy was carried out. This enhances the coherence of the gradient matrix, thereby improving the model’s convergence and generalization performance. The proposed model was evaluated on the SEED dataset for hearing individuals and the HIED dataset for hearing-impaired individuals. Experimental results demonstrate that the model achieves an emotion classification accuracy of 92.62% on the SEED dataset and a classification accuracy of 82.13% on the HIED dataset, fully reflecting its superior classification performance and strong robustness.
Keywords:
Dual-branch
electroencephalogram
emotion recognition
gradient optimization strategy
multiscale convolutions
Journal
IF:
4.5
Papers:
2.1W
Citations:
7.3W
Organization
Cited Papers