Multi-angle polarization spectral reconstruction from sparse measurements based on non-negative matrix factorization
Applied Optics · 20.11.2026 · 65 (33) · 11689
DOI: 10.1364/ao.613718 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Optics, online erschienen am 10.11.2026. Verfasst von Lilong Tan und 3 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben.
Abstract (Originalsprache)
Multi-angle polarization spectral data are valuable for target recognition and environmental monitoring, but complete measurements are time-consuming. In this paper, a polarization spectral reconstruction method based on non-negative matrix factorization (NMF) is proposed. Angle-independent spectral bases are first extracted from the training data measured at an incident angle of 40°, separating the angle and spectral characteristics. A genetic algorithm is then used to select six characteristic wavelengths from 601 candidate wavelengths to reduce the number of measurement channels. Based on the reconstruction residuals of each angle under the extracted spectral bases, a priority ranking is performed, and the training angles are compressed from 78 to 25–41 with a training-set accuracy loss of less than 2%. For an unseen observation angle of a sample for which the offline calibration has been established, the full spectrum from 400 to 1000 nm can be reconstructed using measurements at only the six selected wavelengths. For each sample and test set, R 2 and RMSE are first calculated for each observation angle over the 601 wavelength channels and then averaged over all angles in that test set. Using all 78 training angles, the minimum mean R 2 and maximum mean RMSE across the four samples are 0.9520 and 0.0030, respectively, for the same-incidence held-out-angle test; across the two cross-incidence tests at 30° and 50°, the corresponding worst-case values are 0.8619 and 0.0044. After training-angle reduction, the minimum mean R 2 values across the four samples are 0.9687, 0.8583, and 0.9374 for the same-incidence held-out-angle test, the 30° incidence test, and the 50° incidence test, respectively, with maximum mean RMSE values of 0.0127, 0.0097, and 0.0102. The proposed method provides an effective solution for the sparse acquisition of multi-angle polarization spectral data.
Zitieren
APA
Tan, L., Meng, Z., Yang, Z., & Zhang, Z. (2026). Multi-angle polarization spectral reconstruction from sparse measurements based on non-negative matrix factorization. Applied Optics, 65(33), 11689. https://doi.org/10.1364/ao.613718
Vancouver
Tan L, Meng Z, Yang Z, Zhang Z. Multi-angle polarization spectral reconstruction from sparse measurements based on non-negative matrix factorization. Applied Optics. 2026;65(33):11689. doi:10.1364/ao.613718
BibTeX
@article{Tan2026,
author = {Tan, Lilong and Meng, Ziyang and Yang, Zhiyong and Zhang, Zhiwei},
title = {Multi-angle polarization spectral reconstruction from sparse measurements based on non-negative matrix factorization},
journal = {Applied Optics},
year = {2026},
volume = {65},
number = {33},
pages = {11689},
doi = {10.1364/ao.613718}
}
EUVIMED