EUVIMEDEUVIMED entdecken
Crossref

Dynamic and Decoupled Spatial-Temporal Graph Neural Networks for Human Activity Recognition From Inertial Signals

Yang Chen, Hongkun Zhao, Siyuan Liu, Kang Li

IEEE Signal Processing Letters · 2026 · 33 · 3212-3216

DOI: 10.1109/lsp.2026.3718841 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Yang Chen und 3 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Gefördert von: National Key Research and Development Program of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Spatial-temporal graph neural networks have emerged as a prominent paradigm for human activity recognition from inertial signals. Despite significant progress, accurately modeling long-range dependencies and fine-grained action evolution remains challenging. Many existing frameworks predominantly use static decay strategies based on physical-distance priors, which may limit their flexibility in capturing non-linear global dynamics. Additionally, repeated spatial aggregation may homogenize node representations, while symmetric temporal aggregation is insensitive to sequence directionality and may discard temporal-order information. To address these issues, we propose a Spatial-Temporal Graph Neural Network architecture. First, an Adaptive Dynamic Decay Network learns edge-specific scaling coefficients from semantic interactions among axis-time nodes, enabling flexible refinement of cross-axis and cross-time dependencies. Second, a Decoupled and Recurrent Spatial-Temporal Aggregation Module separates ego-evolution from neighbor aggregation and employs recurrent temporal aggregation to preserve axis-specific information and the ordered evolution of inertial signals. Experiments on three public datasets demonstrate that the proposed architecture learns discriminative spatial-temporal representations and achieves competitive performance.

Förderung

  • National Key Research and Development Program of China

Zitieren

APA

Chen, Y., Zhao, H., Liu, S., & Li, K. (2026). Dynamic and Decoupled Spatial-Temporal Graph Neural Networks for Human Activity Recognition From Inertial Signals. IEEE Signal Processing Letters, 33, 3212-3216. https://doi.org/10.1109/lsp.2026.3718841

Vancouver

Chen Y, Zhao H, Liu S, Li K. Dynamic and Decoupled Spatial-Temporal Graph Neural Networks for Human Activity Recognition From Inertial Signals. IEEE Signal Processing Letters. 2026;33:3212-3216. doi:10.1109/lsp.2026.3718841

BibTeX
@article{Chen2026,
  author = {Chen, Yang and Zhao, Hongkun and Liu, Siyuan and Li, Kang},
  title = {Dynamic and Decoupled Spatial-Temporal Graph Neural Networks for Human Activity Recognition From Inertial Signals},
  journal = {IEEE Signal Processing Letters},
  year = {2026},
  volume = {33},
  pages = {3212--3216},
  doi = {10.1109/lsp.2026.3718841}
}
Kostenlos starten →