Abstract

The objective of this research was to determine if near infrared spectroscopy could be used to discriminate among various food ingredients, and the influence of the type of instrumentation used on the discrimination. One hundred and six samples consisting of 15 milk-egg, 24 buttermilk, 14 cheese and 12 dehydrated onion powders, 20 wheat flours, 10 cheese and 11 ranch seasonings were scanned from 1100 to 2498 nm using diffuse reflectance on a NIRSystems model 6500 scanning spectrophotometer equipped with a rotating sample cup and a Digilab FTS-60 Fourier transform spectrometer (4 and 16 cm⁻¹ resolution) equipped with a custom-made sample transport device. All samples were in a powdered state and were used as is. Discriminant analysis was performed using “Mahalanobis Distance by Principal Component Analysis with Residuals” using the “Averaged Predicted Distance” indicator to determine the number of factors to use for each grouping, obtained from a one-out cross validation analysis. Discrimination models were developed using either all available samples, or 2/3 of the samples in each group with the other third used as a validation set. Results using all available samples demonstrated that calibrations capable of discriminating among the various groups of food ingredients could be developed (100% correct classification) using either spectrometer. Initial results using an independent test set were, however, less satisfactory with many samples often misclassified. While derivatisation did not appear to be of much benefit when all samples were used, it was found to be beneficial in reducing the misclassification rate when samples were divided into calibration and test sets. Also for the Fourier transform spectrometer, results were considerably better at 4 cm⁻¹ resolution than at 16 cm⁻¹ when samples were divided into calibration and validation sets. It was also determined that increasing the distance criteria for group inclusion from 3 to 10 or more resulted in greatly increased discrimination robustness with no increased risk of misclassification. In conclusion, results demonstrated that near infrared spectra of food ingredients obtained from either a scanning spectrophotometer or a Fourier transform spectrometer can be used for product discrimination.

© 1997 NIR Publications

Rapid detection of talc content in flour based on near-infrared spectroscopy combined with feature wavelength selection

Changhao Bao, Changhao Zeng, Jinming Liu, and Dongjie Zhang
Appl. Opt. 61(19) 5790-5798 (2022)

Optical diffuse reflectance accessory for measurements of skin tissue by near-infrared spectroscopy

R. Marbach and H. M. Heise
Appl. Opt. 34(4) 610-621 (1995)

Optical system for tablet variety discrimination using visible/near-infrared spectroscopy

Yongni Shao, Yong He, and Xingyue Hu
Appl. Opt. 46(34) 8379-8384 (2007)

Specialized source-detector separations in near-infrared reflectance spectroscopy platform enable effective separation of diffusion and absorption for glucose sensing

Jin Liu, Tongshuai Han, Jingying Jiang, and Kexin Xu
Biomed. Opt. Express 10(9) 4839-4858 (2019)

Dysplasia and tumor discrimination in brain tissues by combined fluorescence, Raman, and diffuse reflectance spectroscopies

Enrico Baria, Flavio Giordano, Renzo Guerrini, Chiara Caporalini, Anna Maria Buccoliero, Riccardo Cicchi, and Francesco Saverio Pavone
Biomed. Opt. Express 14(3) 1256-1275 (2023)