EUVIMEDEUVIMED entdecken
Crossref

P 2 DNet: A Physics-Constrained Dual-Branch Network for Underwater Polarimetric Image Restoration

Xiangyue Zhang, Wen Sun, Yubo Zheng, Junlin Li, Xin Qu, Hongli Xu

IEEE Sensors Journal · 01.04.2026 · 26 (7) · 10813-10825

DOI: 10.1109/jsen.2026.3667137 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.04.2026. Verfasst von Xiangyue Zhang und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 51 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Fundamental Research Funds for the Central Universities.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Underwater optical imaging serves as a key sensing modality for marine exploration and environmental monitoring, providing critical visual and radiometric information about underwater environments. However, scattering by suspended particles and absorption by water severely degrade image contrast and visibility, thereby limiting the performance of optical sensing in complex underwater environments. Existing image restoration methods mainly rely on empirical color or spatial priors, and often lack physical interpretability and robustness under varying illumination. In contrast, underwater polarimetric imaging can effectively suppress scattered light and provide additional cues for accurate scene reconstruction. Building on this principle, this article presents a physics-constrained, polarimetric dual-branch network for underwater optical image restoration. The proposed framework jointly estimates the background light intensity and the degree of linear polarization (DOLP) of backscattered light, which serve as learnable physical priors for analytical restoration based on the canonical underwater imaging model. A brightness adaptation module is further introduced to compensate for spatially varying illumination, improving the reliability of physical-parameter estimation. In addition, we establish a large-scale polarization dataset, NEU-PURGB, to facilitate supervised learning and improve generalization across diverse real-world environments. On the NEU-PURGB validation set, the proposed P2DNet achieves 26.05-dB PSNR, 0.8973 SSIM, and 0.0866 LPIPS, outperforming existing methods and confirming its effectiveness in high-quality underwater image restoration. Experimental results demonstrate that the proposed framework achieves state-of-the-art performance with strong physical interpretability, effectively enhancing optical sensing capabilities in complex underwater environments.

Förderung

  • National Natural Science Foundation of China
  • Fundamental Research Funds for the Central Universities

Zitieren

APA

Zhang, X., Sun, W., Zheng, Y., Li, J., Qu, X., & Xu, H. (2026). P 2 DNet: A Physics-Constrained Dual-Branch Network for Underwater Polarimetric Image Restoration. IEEE Sensors Journal, 26(7), 10813-10825. https://doi.org/10.1109/jsen.2026.3667137

Vancouver

Zhang X, Sun W, Zheng Y, Li J, Qu X, Xu H. P 2 DNet: A Physics-Constrained Dual-Branch Network for Underwater Polarimetric Image Restoration. IEEE Sensors Journal. 2026;26(7):10813-10825. doi:10.1109/jsen.2026.3667137

BibTeX
@article{Zhang2026,
  author = {Zhang, Xiangyue and Sun, Wen and Zheng, Yubo and Li, Junlin and Qu, Xin and Xu, Hongli},
  title = {P 2 DNet: A Physics-Constrained Dual-Branch Network for Underwater Polarimetric Image Restoration},
  journal = {IEEE Sensors Journal},
  year = {2026},
  volume = {26},
  number = {7},
  pages = {10813--10825},
  doi = {10.1109/jsen.2026.3667137}
}
Kostenlos starten →