Accuracy and biases in predicting the chemical and physical traits of many types of cheeses using different visible and near-infrared spectroscopic techniques and spectrum intervals

摘要

Near-infrared spectroscopy (NIRS) has been widelyused to determine various composition traits of manydairy products in the industry. In the last few years,near-infrared (NIR) instruments have become more andmore accessible, and now, portable devices can be easilyused in the field, allowing the direct measurementof important quality traits. However, the comparisonof the predictive performances of different NIR instrumentsis not simple, and the literature is lacking. Theseinstruments may use different wavelength intervals andcalibration procedures, making it difficult to establishwhether differences are due to the spectral interval, thechemometric approach, or the instrument’s technology.Hence, the aims of this study were (1) to evaluate theprediction accuracy of chemical contents (5 traits), pH,texture (2 traits), and color (5 traits) of 37 categoriesof cheese; (2) to compare 3 instruments [2 benchtop,working in reflectance (R) and transmittance (T) mode(NIRS-R and NIRS-T, respectively) and 1 portabledevice (VisNIRS-R)], using their entire spectral ranges(1100–2498, 850–1048, and 350–1830 nm, respectively,for NIRS-R, NIRS-T and VisNIRS-R); (3) to examinedifferent wavelength intervals of the spectrum within instrument,comparing also the common intervals amongthe 3 instruments; and (4) to determine the presenceof bias in predicted traits for specific cheese categories.A Bayesian approach was used to develop 8 calibrationmodels for each of 13 traits. This study confirmed thatNIR spectroscopy can be used to predict the chemicalcomposition of a large number of different cheeses,whereas pH and texture traits were poorly predicted.Color showed variable predictability, according to thetrait considered, the instrument used, and, within instrument,according to the wavelength intervals. Thepredictive performance of the VisNIRS-R portable devicewas generally better than the 2 laboratory NIRSinstruments, whether with the entire spectrum or selectedintervals. The VisNIRS-R was found suitable foranalyzing chemical composition in real time, withoutthe need for sample uptake and processing. Our resultsalso indicated that instrument technology is much moreimportant than the NIR spectral range for accurateprediction equations, but the visible range is usefulwhen predicting color traits, other than lightness. Specificallyfor certain categories (i.e., caprine, moldy, andfresh cheeses), dedicated calibrations seem to be neededto obtain unbiased and more accurate results.