Spectral library searching: Mid-infrared versus near-infrared spectra for classification of powdered food ingredient

The objective of this study was to determine the potential of using mid- or near-infrared diffuse reflectance spectra to construct food ingredient spe ctral libraries for product identification and checking. Samples (106) cons isting of buttermilk, dehydrated onion, cheese and milk-egg powders, wheat hours, and two powdered seasonings were scanned "as is" (not diluted with K Br using diffuse reflectance) at 4 and 16 cm(-1) resolution in the mid-infr ared on a Digilab FTS-60 and a Perkin-Elmer Model 2000 and in the near-infr ared on an FTS-60 (4 and 16 cm(-1) resolution) and NIRSystems Model 6500 sc anning monochromator (10 nm bandwidth). A custom-made sample transport devi ce was used on the FTS-60, a rotating sample cup on the NIRSystems 6500, an d a stationary cell for the Perkin-Elmer 2000. Every third sample of each g roup was used as a test sample and searched against a library containing th e remaining samples. Results showed that only full spectrum based searches with the use of Euclidian distance or correlation (with or without a first derivative) were useful. All unknowns were correctly classified by using ne ar-infrared spectra generated on either the scanning monochromator or the F TS-60 (4 cm(-1) resolution) or by using any of the mid-infrared spectra. Re sults demonstrated that near- or mid-infrared spectral libraries of powdere d food ingredients can be used for product identification and checking.