Work

Measurement configuration as a primary determinant of spectral quality and analytical reliability in miniaturized NIR spectroscopy

Year: 2026
Type: article
Abstract: Near-infrared spectroscopy (NIRS) is widely recognised for enabling rapid and non-invasive analysis, facilitating its integration into non-specialised environments for routine quality monitoring. However, the development of NIR methods often prioritises chemometric modelling over the evaluation of sampling and measurement configurations, limiting the understanding of how instrumental versatility affects analytical performance. This study assesses three sample-presentation strategies implemented using a low-cost miniaturized NIR device (DLP NIRscan Nano, Texas Instruments) and designed to emulate realistic measurement scenarios: static transmittance in a quartz cuvette, dynamic transmittance using a flow-through cell to simulate on-line analysis, and transflectance measurements using a fiber-optic probe. A dataset comprising three pure commercial olive oils and 112 samples adulterated with controlled proportions of vegetable oils was used as a chemically relevant matrix to determine whether the three configurations preserved the spectral information required for qualitative assessment, supervised classification, and quantitative prediction. Principal component analysis revealed configuration dependent spectral structures associated with the separation of pure and adulterated oils, indicating that each sampling interface emphasized different spectral features. For developing models, the 112 adulterated samples were divided into a calibration set ( n = 88) and an independent external validation set ( n = 24). PLS-DA showed that static transmittance over 930–1500 nm provided the strongest classification of low, medium, and high adulteration levels, achieving an external accuracy of 87.50% and a balanced accuracy of 85.71%. For quantitative PLS regression, static transmittance over 930–1300 nm produced the best external prediction performance (R 2 pred = 0.859, RMSEP = 3.742%, and RPD = 2.73). Dynamic transmittance retained moderate screening ability under continuous flow conditions, whereas the fiber-optic probe preserved some quantitative information but did not support reliable classification. Overall, the results demonstrate that the sampling interface is a key analytical design variable and that the optimal spectral range depends on the analytical objective. The study provides a practical framework for adapting a single miniaturized NIR platform to controlled, continuous-flow, and remote food quality screening scenarios.
Language: English
Field: Chemistry
Open Access status: hybrid
Estimated APC paid: $4,000
OpenAlex ID: W7208819858
FWCI:
Cites: 33
Cited by:
Related to: 10