EUVIMEDEUVIMED entdecken
Crossref

DSGTANet: a semi-supervised dual-stream graph-temporal adaptation network for cross-subject continuous EEG emotion recognition

Chen Lu iD, Li-Lin Jie iD, Ying Xiong, Jia Yuan, Wang-Yang Xu iD

Journal of Neural Engineering · 01.12.2026 · 23 (6) · 066017

DOI: 10.1088/1741-2552/aeba18 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Neural Engineering, online erschienen am 13.11.2026. Verfasst von Chen Lu und 4 weiteren Autor:innen. Beteiligte Einrichtung: School of Instrument Science and Optoelectronic Engineering, Nanchang Hangkong University, Nanchang 330063, People's Republic of China. Hauptthemen laut MEDLINE: Electroencephalography, Emotions, Brain-Computer Interfaces. Die Arbeit nennt 78 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Jiangxi Province.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Objective. Continuous emotion recognition using electroencephalogram (EEG) can be formulated as valence and arousal regression in a two-dimensional affective space rather than discrete emotion classification. It has applications in brain–computer interfaces (BCIs), including mental-disorder regulation and human–computer interaction. In practice, however, its application is limited by scarce emotion labels, inter-subject distribution shifts, and heteroscedastic regression noise with sample-dependent variance. To address these issues, this study aims to develop a semi-supervised framework for accurate cross-subject continuous EEG emotion recognition. Approach. We propose a novel Dual-Stream Graph-Temporal Adaptation Network (DSGTANet). First, we collect an EEG dataset from 40 participants covering six directional dynamic emotional-transition tasks between calm and happy, sad, or nervous states, with continuous valence and arousal labels at 5s granularity. Then, DSGTANet is designed to consist of a spatial structural stream, a temporal dynamic stream, and a reliability-driven fusion and alignment module. Specifically, the spatial structural stream learns robust EEG channel-topology features through a learnable graph structure, structural augmentation, and a self-supervised consistency constraint. Meanwhile, the temporal dynamic stream employs a Transformer to capture long-range temporal dependencies in emotion evolution and introduces label-conditioned domain adversarial learning with an exponential moving average (EMA) teacher to enhance pseudo-labeling for unlabeled samples. Furthermore, the reliability-driven module explicitly estimates dual-stream predictive uncertainty via heteroscedastic regression and performs precision-weighted product-of-experts fusion along with reliability-weighted conditional distribution alignment. Main results . DSGTANet yielded the lowest valence and arousal regression errors among the compared methods on both the self-collected dataset and DEAP. On the self-collected dynamic-transition dataset, the mean MSE and MAE across the six transition tasks and two affective dimensions were 0.246 and 0.300, respectively. Across the two datasets, the average relative reductions in MSE and MAE were 4.78% and 4.35%, respectively, compared with the strongest baseline. Significance. These findings highlight the considerable potential of the proposed DSGTANet for real-world EEG emotion decoding under conditions of limited labeled data and substantial subject variability.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

ElectroencephalographyEmotionsBrain-Computer InterfacesHumansGraph Neural NetworksMaleFemaleAdultYoung AdultReproducibility of Results

Stichwörter der Autor:innen

EEGcontinuous emotion recognitioncross-subject adaptationgraph-temporal modelingsemi-supervised learning

Autor:innen und Institutionen

  • School of Instrument Science and Optoelectronic Engineering, Nanchang Hangkong University, Nanchang 330063, People's Republic of China.
  • School of Civil Aviation, Nanchang Hangkong University, Nanchang 330063, People's Republic of China.
  • School of Economics and Management, Nanchang Hangkong University, Nanchang 330063, People's Republic of China.

ORCID-Profile: Chen Lu, Li-Lin Jie, Wang-Yang Xu

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Jiangxi Province

Zitieren

APA

Lu, C., Jie, L. L., Xiong, Y., Yuan, J., & Xu, W. Y. (2026). DSGTANet: a semi-supervised dual-stream graph-temporal adaptation network for cross-subject continuous EEG emotion recognition. Journal of Neural Engineering, 23(6), 066017. https://doi.org/10.1088/1741-2552/aeba18

Vancouver

Lu C, Jie LL, Xiong Y, Yuan J, Xu WY. DSGTANet: a semi-supervised dual-stream graph-temporal adaptation network for cross-subject continuous EEG emotion recognition. Journal of Neural Engineering. 2026;23(6):066017. doi:10.1088/1741-2552/aeba18

BibTeX
@article{Lu2026,
  author = {Lu, Chen and Jie, Li-Lin and Xiong, Ying and Yuan, Jia and Xu, Wang-Yang},
  title = {DSGTANet: a semi-supervised dual-stream graph-temporal adaptation network for cross-subject continuous EEG emotion recognition},
  journal = {Journal of Neural Engineering},
  year = {2026},
  volume = {23},
  number = {6},
  pages = {066017},
  doi = {10.1088/1741-2552/aeba18}
}
Kostenlos starten →