Super-resolution reconstruction of structured illumination microscopy by untrained physics-informed neural networks with low-rank spectral constraints
Optics Continuum · 15.08.2026 · 5 (8) · 2724
DOI: 10.1364/optcon.606801 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optics Continuum, online erschienen am 06.08.2026. Verfasst von Jie Geng und 4 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shanghai, EURAMET - European Association of National Metrology Institutes.
Abstract (Originalsprache)
Structured illumination microscopy has been widely used in super-resolution imaging. Its resolution can be improved to 2× or even greater than that of traditional microscopy. However, existing reconstruction algorithms suffer from artifacts caused by model mismatch or poor model generalizability. In this paper, we introduce frequency-domain low-rank constraints to regularize the solution space, thereby suppressing noise and artifacts unrelated to the real features in the images. Meanwhile, to avoid improper setting of models, a parametric representation of the point spread function is adopted and integrated with the differentiable optical forward model. Ultimately, a physically driven optimization model is constructed without training datasets. Through dynamic joint optimization and a spectral separation mechanism, resolution enhancement that is consistent with theoretical expectations is realized. Experimental results demonstrate substantial enhancements in terms of structural fidelity and noise suppression compared to conventional structured illumination microscopy. This low-rank physics-informed neural network (LR-PINN) method provides a promising solution for image super-resolution reconstruction in complex scenarios.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Shanghai
- EURAMET - European Association of National Metrology Institutes
- State Key Laboratory of Crystal Materials
Zitieren
APA
Geng, J., Zheng, G., Hua, Z., Li, D., & Zhang, X. (2026). Super-resolution reconstruction of structured illumination microscopy by untrained physics-informed neural networks with low-rank spectral constraints. Optics Continuum, 5(8), 2724. https://doi.org/10.1364/optcon.606801
Vancouver
Geng J, Zheng G, Hua Z, Li D, Zhang X. Super-resolution reconstruction of structured illumination microscopy by untrained physics-informed neural networks with low-rank spectral constraints. Optics Continuum. 2026;5(8):2724. doi:10.1364/optcon.606801
BibTeX
@article{Geng2026,
author = {Geng, Jie and Zheng, Gang and Hua, Zhijie and Li, Di and Zhang, Xiangchao},
title = {Super-resolution reconstruction of structured illumination microscopy by untrained physics-informed neural networks with low-rank spectral constraints},
journal = {Optics Continuum},
year = {2026},
volume = {5},
number = {8},
pages = {2724},
doi = {10.1364/optcon.606801}
}
EUVIMED