EUVIMEDEUVIMED entdecken
Crossref

Enhancing Remote Heart Rate Monitoring With FedQ²: Focusing on Motion Interference in Federated Learning

Bowen Zhao, Chang-Hong Fu, Hong Hong, Long Shi

IEEE Sensors Journal · 15.10.2025 · 25 (20) · 39054-39063

DOI: 10.1109/jsen.2025.3603871 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 15.10.2025. Verfasst von Bowen Zhao und 3 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 38 Literaturangaben. Gefördert von: Jiangsu Science and Technology Program, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Remote photoplethysmography (rPPG) has emerged as a promising noncontact sensing technology for continuous heart rate monitoring in dynamic environments. While conventional rPPG algorithms are prone to motion artifacts in unconstrained scenarios, centralized deep learning methods face limitations from insufficient data diversity and privacy concerns. Federated learning provides a privacy-preserving approach for decentralized model training, offering potential solutions to these limitations. However, federated training suffers a significant performance drop when data from individual camera nodes are contaminated by noise, especially motion artifacts. This article proposes FedQ2, a federated learning framework that integrates signal quality-aware aggregation and adaptive data weighting to address motion interference in distributed rPPG systems. Specifically, FedQ2 introduces three innovations: 1) proposing a model-aware signal quality module for real-time evaluation of data quality; 2) a dual-weight aggregation mechanism that jointly optimizes model contributions from edge devices based on both data quantity and signal quality metrics; and 3) building a federated learning evaluation system for the first time on different public datasets to validate the effectiveness of the FedQ2. The experimental results demonstrate that while baseline federated averaging (FedAvg) achieves competitive performance in intra- and inter-dataset evaluations, its accuracy degrades significantly when motion-corrupted data from individual nodes participate in training. The proposed FedQ2 effectively mitigates these interference effects through its quality-driven aggregation strategy, ensuring robust performance across varying motion conditions.

Förderung

  • Jiangsu Science and Technology Program
  • National Natural Science Foundation of China

Zitieren

APA

Zhao, B., Fu, C., Hong, H., & Shi, L. (2025). Enhancing Remote Heart Rate Monitoring With FedQ²: Focusing on Motion Interference in Federated Learning. IEEE Sensors Journal, 25(20), 39054-39063. https://doi.org/10.1109/jsen.2025.3603871

Vancouver

Zhao B, Fu C, Hong H, Shi L. Enhancing Remote Heart Rate Monitoring With FedQ²: Focusing on Motion Interference in Federated Learning. IEEE Sensors Journal. 2025;25(20):39054-39063. doi:10.1109/jsen.2025.3603871

BibTeX
@article{Zhao2025,
  author = {Zhao, Bowen and Fu, Chang-Hong and Hong, Hong and Shi, Long},
  title = {Enhancing Remote Heart Rate Monitoring With FedQ²: Focusing on Motion Interference in Federated Learning},
  journal = {IEEE Sensors Journal},
  year = {2025},
  volume = {25},
  number = {20},
  pages = {39054--39063},
  doi = {10.1109/jsen.2025.3603871}
}
Kostenlos starten →