Spatiotemporal graph prototype learning with adversarial training for UWB open-set gesture recognition
Measurement Science and Technology · 17.07.2026 · 37 (28) · 286102
DOI: 10.1088/1361-6501/ae7fab ↗Kurz gefasst
Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 10.07.2026. Verfasst von Fubiao Yang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 57 Literaturangaben.
Abstract (Originalsprache)
Open-set gesture recognition using ultra-wideband (UWB) radar remains challenging because practical sensing systems may encounter unseen gesture inputs and subject-related variations that are absent from the training data. To address these issues, this paper proposes a spatiotemporal graph prototype learning (STGPL) framework for UWB radar open-set gesture recognition. STGPL integrates hybrid spatiotemporal feature extraction, graph prototype learning, and adversarial feature disentanglement into a unified framework. Specifically, a hybrid dual-stream feature extractor combines atrous spatial pyramid pooling and a transformer encoder to capture multiscale range-time patterns and long-range temporal dependencies from radar range-time maps. A graph prototype learning module then calibrates sample features according to their relationships with known-class prototypes, which helps enhance known-class compactness and improve the separability between known and unknown samples. In addition, an adversarial feature disentanglement module based on a gradient reversal layer is introduced to reduce subject-related information in the learned representations, thereby improving recognition stability under subject-independent conditions. Experiments on the UWB-Gestures dataset demonstrate that STGPL achieves strong closed-set recognition and open-set recognition (OSR) performance. In the standard OSR evaluation, STGPL obtains average area under the receiver operating characteristic curve and open set classification rate values of 0.9291 and 0.9201 across five gesture class splits, outperforming representative OSR baselines. Further openness analysis, cross-subject evaluation, ablation studies, and feature visualizations confirm the effectiveness of the proposed modules for robust UWB radar gesture recognition in open-set environments.
Zitieren
APA
Yang, F., Yin, Y., Sun, S., Wang, Y., & Zhang, Y. (2026). Spatiotemporal graph prototype learning with adversarial training for UWB open-set gesture recognition. Measurement Science and Technology, 37(28), 286102. https://doi.org/10.1088/1361-6501/ae7fab
Vancouver
Yang F, Yin Y, Sun S, Wang Y, Zhang Y. Spatiotemporal graph prototype learning with adversarial training for UWB open-set gesture recognition. Measurement Science and Technology. 2026;37(28):286102. doi:10.1088/1361-6501/ae7fab
BibTeX
@article{Yang2026,
author = {Yang, Fubiao and Yin, Yanhua and Sun, Sibo and Wang, Yuanyuan and Zhang, Yi},
title = {Spatiotemporal graph prototype learning with adversarial training for UWB open-set gesture recognition},
journal = {Measurement Science and Technology},
year = {2026},
volume = {37},
number = {28},
pages = {286102},
doi = {10.1088/1361-6501/ae7fab}
}
EUVIMED