HBUED: An EEG dataset for emotion recognition.
Journal:
Journal of affective disorders
Published Date:
Sep 15, 2025
Abstract
Emotion recognition via electroencephalogram (EEG) data is crucial for improving human-computer interaction. In practice, researchers require a substantial quantity of EEG samples to train and validate models. However, existing EEG datasets typically have a limited number of subjects. To address this issue, we present a large-scale EEG dataset, the Hebei University Emotional EEG Dataset (HBUED), specifically designed for research on human emotion recognition. Furthermore, this research presents a deep learning methodology aimed at improving emotion recognition performance by efficiently handling complicated samples in EEG-based emotion recognition. This method first constructs a dual-input network architecture to extract discriminative features of EEG signals from two perspectives for classification. Furthermore, this paper uses a parallel feature extraction module for EEG signals, which increases the number of neurons per layer by expanding the network width, thereby extracting more comprehensive feature information while avoiding overfitting caused by excessive network depth. In addition, a topological feature extraction module has been created to better capture the topological characteristics of EEG signals. Lastly, the proposed method is validated on both the self-constructed HBUED and the public DEAP datasets, with experimental results demonstrating its effectiveness. The HBUED datasets and the source code of the proposed method are publicly available at: https://tensorground.github.io/HBUED.github.io/.
