EUVIMEDEUVIMED entdecken
Crossref

Using generative adversarial networks to correct for shell interference on Vis/NIR spectral acquisition

Kaihao Ji, Xuesong Jiang, Qing Chen, Linyun Xu, Zuping Yang, Hongping Zhou

Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy · 2025-11 · 341 · 126409

DOI: 10.1016/j.saa.2025.126409 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2025-11. Verfasst von Kaihao Ji und 5 weiteren Autor:innen. Beteiligte Einrichtung: College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjing 210037, China. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 33 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Thick skins and shells significantly hinder light penetration, resulting in spectra with low signal-to-noise ratios that impair the predictive performance of calibration models. To address this challenge, this study proposes DehullingNet, a generative adversarial network-based model designed to mitigate shell-induced interference on visible/near-infrared (Vis/NIR) spectral acquisition. DehullingNet learns a mapping between real shelled spectra and real dehulled spectra, enabling spectrum-to-spectrum translation. Compared with traditional calibration transfer methods such as procrustes analysis and piecewise direct standardization, DehullingNet offers superior capability in modeling nonlinear spectral relationships. This advantage is attributed to its carefully designed learning strategy and network architecture. Key modules, including a fully connected layer, Wasserstein distance, and spectral normalization, are incorporated to effectively prevent mode collapse. To validate the method's effectiveness, pecan moisture content detection was used as a case study. Models were constructed using generated dehulled spectra, real shelled spectra, and real dehulled spectra. The optimal model based on generated dehulled spectra achieved accuracies of 84.64% and 84.17% on the training and test sets, respectively. These represent improvements of 8.57% and 15% over the optimal model based on real shelled spectra. Overall, DehullingNet effectively mitigates shell interference in spectral modeling and provides reliable data support for non-destructive fruit quality assessment using Vis/NIR spectroscopy.

Schlagwörter

Stichwörter der Autor:innen

Generative adversarial networksNon-destructive detectionShellSpectral correctionVis/NIR spectroscopy

Autor:innen und Institutionen

  • College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjing 210037, China.
  • Co-Innovation Center of Efficient Processing and Utilization of Forest Resources, Nanjing Forestry University, Nanjing 210037, China; College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjin…
  • College of Marxism, Nanjing Forestry University, Nanjing 210037, China.

Zitieren

APA

Ji, K., Jiang, X., Chen, Q., Xu, L., Yang, Z., & Zhou, H. (2025). Using generative adversarial networks to correct for shell interference on Vis/NIR spectral acquisition. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 341, 126409. https://doi.org/10.1016/j.saa.2025.126409

Vancouver

Ji K, Jiang X, Chen Q, Xu L, Yang Z, Zhou H. Using generative adversarial networks to correct for shell interference on Vis/NIR spectral acquisition. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy. 2025;341:126409. doi:10.1016/j.saa.2025.126409

BibTeX
@article{Ji2025,
  author = {Ji, Kaihao and Jiang, Xuesong and Chen, Qing and Xu, Linyun and Yang, Zuping and Zhou, Hongping},
  title = {Using generative adversarial networks to correct for shell interference on Vis/NIR spectral acquisition},
  journal = {Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy},
  year = {2025},
  volume = {341},
  pages = {126409},
  doi = {10.1016/j.saa.2025.126409}
}
Kostenlos starten →