Enhancing Cross-Dataset EEG Emotion Recognition: A Novel Approach with Emotional EEG Style Transfer Network

TL;DR

This work tackles cross-dataset EEG emotion recognition by addressing inter-dataset style shifts with the Emotional EEG Style Transfer Network (E$^2$STN). The framework combines a transfer module that fuses source-content with target-style, a transfer-evaluation module that enforces content-, style-, and identity-consistent stylization, and a discriminative prediction module that leverages a dynamic graph neural network for classification. Through multi-objective optimization with content, style, identity, and cross-entropy losses, E$^2$STN achieves state-of-the-art results on 3- and 4-category cross-dataset tasks across SEED, SEED-IV, and MPED datasets, with statistically significant improvements. Analyses show the transfer component substantially boosts performance and reveal frontal/temporal brain regions as key contributors, providing neurophysiological insights. The approach offers a practical pathway to more generalizable affective BCIs by mitigating dataset biases and enabling robust cross-dataset emotion decoding in EEG signals.

Abstract

Recognizing the pivotal role of EEG emotion recognition in the development of affective Brain-Computer Interfaces (aBCIs), considerable research efforts have been dedicated to this field. While prior methods have demonstrated success in intra-subject EEG emotion recognition, a critical challenge persists in addressing the style mismatch between EEG signals from the source domain (training data) and the target domain (test data). To tackle the significant inter-domain differences in cross-dataset EEG emotion recognition, this paper introduces an innovative solution known as the Emotional EEG Style Transfer Network (E$^2$STN). The primary objective of this network is to effectively capture content information from the source domain and the style characteristics from the target domain, enabling the reconstruction of stylized EEG emotion representations. These representations prove highly beneficial in enhancing cross-dataset discriminative prediction. Concretely, E$^2$STN consists of three key modules\textemdash transfer module, transfer evaluation module, and discriminative prediction module\textemdash which address the domain style transfer, transfer quality evaluation, and discriminative prediction, respectively. Extensive experiments demonstrate that E$^2$STN achieves state-of-the-art performance in cross-dataset EEG emotion recognition tasks.

Figures (5)

• Figure 1: Distribution of EEG data in different subjects and datasets.
• Figure 2: Framework of E$^2$STN. E$^2$STN consists of three modules to obtain stylized emotional EEG representations containing emotion content of the source domain and statistical characteristics of the target domain, and meanwhile, make discriminative predictions for cross-dataset emotional EEG samples.
• Figure 3: Architecture of the encoder and decoder layer in the transfer module. "Add&Norm" signifies the addition of the residual connection and subsequent layer normalization.
• Figure 4: Confusion matrices of E$^2$STN results on cross-dataset experiments.
• Figure 5: Visualization of the dynamic graph distribution in the discriminative prediction module.