Scanner‐Independent Calibrations of Handheld Near‐Infrared Spectra for Robust Quality Assessment of Vanilla Pods
Journal of Food Quality · 2026-01 · 2026 (1)
DOI: 10.1155/jfq/5564134 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Food Quality, online erschienen am 20.05.2026. Verfasst von Marina De Gea Neves und 8 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Austrian Science Fund.
Abstract (Originalsprache)
Vanillin is a high‐value flavor compound derived from Vanilla planifolia , whose grading in commercial supply chains remains challenging due to compositional heterogeneity of the pods and variability among handheld near‐infrared (NIR) scanners. In this study, scanner‐independent calibration models were developed for the rapid prediction of vanillin content and dry matter in vanilla pods using diffuse‐reflection handheld NIR spectroscopy and partial least squares (PLS) regression. Thirty‐nine batches from a Ugandan vanilla farm, spanning typical commercial grades and a broad range of vanillin (≤ 0.1–38.2 mg/g) and dry matter (71.3%–95.5% (w/w)), were measured at multiple positions on three pods per batch. More than 2000 raw spectra were acquired, including additional measurements from eight auxiliary scanners of the same model to emulate cross‐device deployment. Reference values for vanillin and dry matter were obtained by high‐performance liquid chromatography (HPLC) and oven‐drying, respectively, using composite extracts of three pods per batch to reflect real‐world grading practice. Spectral preprocessing included extended multiplicative scatter correction (EMSC) and truncation of water‐dominated regions for the vanillin model and standard normal variate (SNV), followed by first‐derivative Savitzky–Golay smoothing for the dry matter model. The PLS vanillin model yielded RMSE values (RMSEC/RMSECV/RMSEP) of 2.963/6.044/2.309 mg/g and corresponding R 2 values (Cal/CV/Pred) of 0.921/0.699/0.979, with an average absolute prediction error of 1.98 mg/g for five external test batches. The dry matter model achieved RMSEC/RMSECV/RMSEP values of 1.862/2.379/1.657% (w/w) and R 2 (Cal/CV/Pred) values of 0.906/0.846/0.932, with an average absolute prediction error of 1.42% (w/w). X‐loading analysis confirmed that both models rely on chemically meaningful NIR bands, dominated by water overtones and combination bands for dry matter and aromatic/CH‐related regions for vanillin. The proposed calibration framework is robust to pod‐to‐pod variability and scanner hardware fluctuations, enabling decentralized field deployment of a shared master calibration for on‐site quality assessment of vanilla pods.
Förderung
- Austrian Science Fund
Zitieren
APA
Neves, M. D. G., Pfeifer, F., Siesler, H. W., Bec, K. B., Grabska, J., Rutzinger, P., Huck, C. W., Wise, B. M., & Palacios, M. A. (2026). Scanner‐Independent Calibrations of Handheld Near‐Infrared Spectra for Robust Quality Assessment of Vanilla Pods. Journal of Food Quality, 2026(1). https://doi.org/10.1155/jfq/5564134
Vancouver
Neves MDG, Pfeifer F, Siesler HW, Bec KB, Grabska J, Rutzinger P, et al. Scanner‐Independent Calibrations of Handheld Near‐Infrared Spectra for Robust Quality Assessment of Vanilla Pods. Journal of Food Quality. 2026;2026(1). doi:10.1155/jfq/5564134
BibTeX
@article{Neves2026,
author = {Neves, Marina De Gea and Pfeifer, Frank and Siesler, Heinz W. and Bec, Krzysztof B. and Grabska, Justyna and Rutzinger, Peter and Huck, Christian W. and Wise, Barry M. and Palacios, Manuel A.},
title = {Scanner‐Independent Calibrations of Handheld Near‐Infrared Spectra for Robust Quality Assessment of Vanilla Pods},
journal = {Journal of Food Quality},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/jfq/5564134}
}
EUVIMED