Skip to main content
ARS Home » Research » Publications at this Location » Publication #77824
Title: RAMAN SPECTROSCOPY FOR THE DETERMINATION OF TOTAL DIETARY FIBER IN CEREAL PRODUCTS

Author
item Archibald, Douglas
item Kays, Sandra
item Barton Ii, Franklin

Submitted to: Pittsburg Conference on Analytical Chemistry and Applied Spectroscopy
Publication Type: Abstract Only
Publication Acceptance Date: 12/6/1996
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Dietary fiber in food products is currently determined by gravimetry using a laborious series of enzymatic and chemical digestions and separations. We are taking steps to replace chemical determination of total dietary fiber (TDF) with spectroscopic methods which will be quicker and cleaner. toward this goal we analyzed a representative and diverse set of cereal products from such groups as snack foods, breakfast cereals, and flours. The set contained a wide variety of grains and grain components. Moreover, products with high fat (>10%) and/or high sugar (>20%) content were included. The TDF reference values ranged from less than 1% to greater than 50%. Specimens were ground and their spectra measured with an FT-Raman system. The spectra were preprocessed to normalize intensities and remove baseline offset. For some models, first and second derivatives were used to extract features from the spectra. Partial-least-squares (PLS) calibration was performed on a set of 63 samples and the number of factors determined by cross-validation. The RMSECV of a 6 factor model was 2.00% TDF with this set. An independent validation set of 63 samples yielded a RMSEP of 2.78% using this model. The model error was greater for samples with higher levels of TDF, for certain product classes, and for a couple of specimens which appear to be too different from the others in the set. No outliers were removed for the above analysis. The reconstructed TDF Raman spectral response will be presented for several types of calibration models. Outliers and errors for sample groups will be compared to NIR-reflectance PLS models which obtain similar overall performance with the same set of samples (12 factors, RMSECV =2.12%, RMSEP = 2.38%).