Quantization-aware nonlinear diffractive neural network for robust optical image reconstruction under fabrication constraints
Optics Express · 16.11.2026 · 34 (23) · 43663
DOI: 10.1364/oe.607289 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optics Express, online erschienen am 04.11.2026. Verfasst von Junyu Wang und 13 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Gefördert von: Ningbo Municipal Science and Technology Innovation Yong jiang 2035 Key R&D Plan, Natural Science Foundation of Ningbo Municipality, Natural Science Foundation of Zhejiang Province.
Abstract (Originalsprache)
Diffractive neural networks provide a compact platform for optical image reconstruction, but practical implementation is constrained by finite phase resolution, spatial phase complexity, and interlayer misalignment. A physics-constrained nonlinear diffractive neural network is developed by integrating quantization-aware training (QAT), circular total-variation (TV) regularization, and Kerr nonlinear phase modulation within an angular-spectrum propagation model. Under hard 8-level phase deployment, QAT provides the dominant improvement in reconstruction fidelity, while circular-TV regularization reduces phase spatial complexity with a limited fidelity penalty, and Kerr modulation provides a complementary nonlinear degree of freedom. Perturbation-aware training further improves tolerance to both transverse and axial interlayer misalignments. These results establish a simulation-based design framework for diffractive optical image reconstruction under finite phase resolution, phase-complexity, and assembly constraints.
Förderung
- Ningbo Municipal Science and Technology Innovation Yong jiang 2035 Key R&D Plan
- Natural Science Foundation of Ningbo Municipality
- Natural Science Foundation of Zhejiang Province
- Graduate Student Scientific Research and Innovation Project of Ningbo University
Zitieren
APA
Wang, J., Lu, J., Li, H., Wang, K., Zhou, J., Xiang, Z., Wen, Y., Cao, X., Xu, W., Yan, C., Wang, Z., Zhou, Y., Xu, K., & Jiang, R. (2026). Quantization-aware nonlinear diffractive neural network for robust optical image reconstruction under fabrication constraints. Optics Express, 34(23), 43663. https://doi.org/10.1364/oe.607289
Vancouver
Wang J, Lu J, Li H, Wang K, Zhou J, Xiang Z, et al. Quantization-aware nonlinear diffractive neural network for robust optical image reconstruction under fabrication constraints. Optics Express. 2026;34(23):43663. doi:10.1364/oe.607289
BibTeX
@article{Wang2026,
author = {Wang, Junyu and Lu, Jie and Li, Haoyu and Wang, Kexiang and Zhou, Junxiu and Xiang, Zeyang and Wen, Yun and Cao, Xinyu and Xu, Weitian and Yan, Chengjun and Wang, Zhihao and Zhou, Yuting and Xu, Ke and Jiang, Ran},
title = {Quantization-aware nonlinear diffractive neural network for robust optical image reconstruction under fabrication constraints},
journal = {Optics Express},
year = {2026},
volume = {34},
number = {23},
pages = {43663},
doi = {10.1364/oe.607289}
}
EUVIMED