EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

PINSA: Physics-Informed Neural Spectral Imaging for Plastisphere Differentiation

Jingyan Chen, Yanmin Zhu, Yuxing Li, Yuqing Cao, Edmund Y. Lam

ACS Photonics · 05.08.2026 · 13 (15) · 4387-4397

DOI: 10.1021/acsphotonics.6c01063 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACS Photonics, online erschienen am 25.07.2026. Verfasst von Jingyan Chen und 4 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research Grants Council, University Grants Committee.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Plastisphere, a microbial community colonizing plastic debris, poses increasing risks to ecosystems and to animal and human health by enabling the long-range transport of invasive and pathogenic species in aquatic environments. Current analytical approaches for plastisphere analysis remain confined to laboratory settings and rely on bulky, time-consuming, and costly instruments. Here, we present a physics-informed neural spectral imaging system (PINSA) for portable, low-cost, plastisphere differentiation. The system uses a customized color-film spectral encoder costing approximately USD 0.1 to optically compress hyperspectral information into a small number of snapshot measurements and reconstructs 451-channel hyperspectral image stacks using a physics-informed neural network constrained by the optical forward model. For classification, we introduce a dual-branch feature-fusion strategy that jointly exploits the spectrally detailed yet reconstruction error-prone recovered hyperspectral data and the low-dimensional yet lossless raw compressed measurements. By fusing both representations, the two branches compensate for each other’s weaknesses. Multiple feature extractors are applied to both branches, and the resulting embeddings are fused before the classification. Experimental results demonstrate that the system accurately recovers 451-channel hyperspectral images and substantially improves the plastisphere classification performance, achieving 85.3% accuracy in differentiating five categories of plastisphere communities sampled from distinct aquatic environments. This work establishes a rapid and cost-effective full-field compressed spectral imaging pathway for plastisphere screening, with potential utility as a preliminary environmental monitoring tool to prioritize samples for downstream microbiological or public-health-related analyses.

Förderung

  • Research Grants Council, University Grants Committee

Zitieren

APA

Chen, J., Zhu, Y., Li, Y., Cao, Y., & Lam, E. Y. (2026). PINSA: Physics-Informed Neural Spectral Imaging for Plastisphere Differentiation. ACS Photonics, 13(15), 4387-4397. https://doi.org/10.1021/acsphotonics.6c01063

Vancouver

Chen J, Zhu Y, Li Y, Cao Y, Lam EY. PINSA: Physics-Informed Neural Spectral Imaging for Plastisphere Differentiation. ACS Photonics. 2026;13(15):4387-4397. doi:10.1021/acsphotonics.6c01063

BibTeX
@article{Chen2026,
  author = {Chen, Jingyan and Zhu, Yanmin and Li, Yuxing and Cao, Yuqing and Lam, Edmund Y.},
  title = {PINSA: Physics-Informed Neural Spectral Imaging for Plastisphere Differentiation},
  journal = {ACS Photonics},
  year = {2026},
  volume = {13},
  number = {15},
  pages = {4387--4397},
  doi = {10.1021/acsphotonics.6c01063}
}
Kostenlos starten →