Prediction of fat in intact cereal food products using near-infrared reflectance spectroscopy

To evaluate the feasibility of an intact product approach to the near-infrared (NIR) determination of fat content, a rapid acquisition spectrometer, with an InGaAs diode-array detector and custom built sampling device, was used to obtain reflectance spectra (1100-1700 nm) of diverse cereal food products. Fat content reference data were obtained gravimetrically by extraction with petroleum ether (AOAC Method 945.16). Using spectral and reference data, partial least-squares regression analysis was applied to calculate a NIR model (n = 89) to predict fat in intact cereal products; the model was adequate for rapid screening of samples, predicting the test samples (n = 44) with root mean square error of prediction (RMSEP) of 11.8 (range 1.4-204.8) g kg(-1) and multiple coefficient of determination of 0.98. Repeated repacking and rescanning of the samples did not appreciably improve model performance. The model was expanded to include samples with a broad range of particle sizes and moisture contents without reduction in prediction accuracy for the untreated samples. The regression coefficients for the models calculated indicated that spectral features at 1165, 1215 and 1395 nm, associated with CH stretching in fats, were the most critical for model development. Published in 2005 for SCI by John Wiley & Sons, Ltd.