EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Physics-guided dataset homogeneity unlocks universal deep learning generalization in scattering media imaging

Xuyu Zhang, Haofan Huang, Dawei Zhang, Songlin Zhuang, Shensheng Han, Honglin Liu, Puxiang Lai

Advanced Photonics · 06.05.2026 · 8 (03) · 1-13

DOI: 10.1117/1.ap.8.3.036015 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Advanced Photonics, erschienen am 06.05.2026. Verfasst von Xuyu Zhang und 6 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, National Key R&D Program of China, Hong Kong Research Grant Council.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning (DL) has revolutionized imaging through scattering media, yet its widespread adoption is hindered by limited generalization, where models trained on specific datasets fail to perform reliably in unseen scenarios. Conventional wisdom attributes this limitation to feature-prior mismatches, but we identify a more root cause: a fundamental mismatch between the learned neural mapping and the system’s true physical inverse operator (T−1), driven principally by inhomogeneous spatial-intensity distributions in conventional training data. To overcome this, we introduce a physics-guided dataset homogeneity strategy. We demonstrate that enforcing spatial uniformity—ensuring all spatial modes in the region of interest are equally and sufficiently sampled—effectively aligns the learned weights with the physical transmission matrix. This approach ensures the network simulates the underlying physical laws rather than merely memorizing dataset-specific statistical biases. Specifically, by optimizing training datasets, we achieve unprecedented cross-dataset generalization: networks trained on simple digits successfully reconstruct complex face images. This physics-guided framework not only overcomes generalization barriers in scattering imaging but also establishes a universal principle for designing robust DL architectures. The conceptual repositioning of DL, from pure data-fitting to physics-simulating, is a big step forward for its reliable deployments in real-world imaging applications.

Förderung

  • National Natural Science Foundation of China
  • National Key R&D Program of China
  • Hong Kong Research Grant Council
  • Hong Kong Innovation and Technology Commission
  • Shenzhen Science and Technology Innovation Commission
  • Hong Kong Polytechnic University

Zitieren

APA

Zhang, X., Huang, H., Zhang, D., Zhuang, S., Han, S., Liu, H., & Lai, P. (2026). Physics-guided dataset homogeneity unlocks universal deep learning generalization in scattering media imaging. Advanced Photonics, 8(03), 1-13. https://doi.org/10.1117/1.ap.8.3.036015

Vancouver

Zhang X, Huang H, Zhang D, Zhuang S, Han S, Liu H, et al. Physics-guided dataset homogeneity unlocks universal deep learning generalization in scattering media imaging. Advanced Photonics. 2026;8(03):1-13. doi:10.1117/1.ap.8.3.036015

BibTeX
@article{Zhang2026,
  author = {Zhang, Xuyu and Huang, Haofan and Zhang, Dawei and Zhuang, Songlin and Han, Shensheng and Liu, Honglin and Lai, Puxiang},
  title = {Physics-guided dataset homogeneity unlocks universal deep learning generalization in scattering media imaging},
  journal = {Advanced Photonics},
  year = {2026},
  volume = {8},
  number = {03},
  pages = {1--13},
  doi = {10.1117/1.ap.8.3.036015}
}
Kostenlos starten →