CS-Unet: spectral reconstruction via compressive sensing–deep learning synergy for robust nanophotonic spectral imaging
Chinese Optics Letters · 2026 · 24 (4) · 043001
DOI: 10.3788/col202624.043001 ↗Kurz gefasst
Fachartikel in der Zeitschrift Chinese Optics Letters, 2026. Verfasst von Lianwei Huang und 4 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben.
Abstract (Originalsprache)
Conventional deep-learning-based reconstruction algorithms for computational spectral imaging degrade severely under nanophotonic device fabrication errors. In this work, a compressive sensing – deep learning coupled algorithm (CS-Unet) that synergistically integrates the two-step iterative shrinkage/thresholding algorithm with a Unet architecture is proposed. Without fabrication errors, CS-Unet delivers reconstruction performance comparable to Unet. However, with fabrication errors, Unet exhibits severe declines in peak signal-to-noise ratio (PSNR), spectral angle mappers (SAMs), and structural similarity index measure (SSIM) metrics, while CS-Unet maintains robust performance with minor degradations. This integration of compressive sensing and deep learning ensures robust generalization to manufacturing variations, enabling practical nanophotonic computational imaging systems.
Zitieren
APA
Huang, L., Tao, C., Zhou, Z., Chen, D., & Zhu, H. (2026). CS-Unet: spectral reconstruction via compressive sensing–deep learning synergy for robust nanophotonic spectral imaging. Chinese Optics Letters, 24(4), 043001. https://doi.org/10.3788/col202624.043001
Vancouver
Huang L, Tao C, Zhou Z, Chen D, Zhu H. CS-Unet: spectral reconstruction via compressive sensing–deep learning synergy for robust nanophotonic spectral imaging. Chinese Optics Letters. 2026;24(4):043001. doi:10.3788/col202624.043001
BibTeX
@article{Huang2026,
author = {Huang, Lianwei and Tao, Chenning and Zhou, Zhiyong and Chen, Daru and Zhu, Huanzheng},
title = {CS-Unet: spectral reconstruction via compressive sensing–deep learning synergy for robust nanophotonic spectral imaging},
journal = {Chinese Optics Letters},
year = {2026},
volume = {24},
number = {4},
pages = {043001},
doi = {10.3788/col202624.043001}
}
EUVIMED