Use of near-infrared spectroscopy for qualitative and quantitative analyses of grains and cereal products

The purpose of the first part of the study was to develop a simplified near-infrared reflectance (NIR) spectroscopy method for detecting insect larvae in individual wheat kernels. Discriminant analysis, based on Mahalanobis distances calculated from log 1/R data at only four discrete wavelengths, yielded better results for classification of sound and insect infested wheat kernels than principal component analysis (PCA) using the spectral region from 1100 to 1900 nm. This simplified technique was then used to detect 3- and 4-week-old larvae of granary and maize weevils in wheat kernels. A model developed from a calibration set containing sound kernels and kernels infested with 3- week-old larvae was applied to a validation set containing sound kernels, sound air-dried kernels, kernels containing 3-week-old larvae of granary and maize weevils, kernels containing 4-week-old larvae of granary and maize weevils, and infested air-dried kernels containing dead larvae of both species. Correct classification rates of 92, 98, 77, 73, 95, 98, 96, and 94%, respectively, were achieved. Additionally, 99% of sound kernels from ten different wheat varieties were correctly classified into their respective classes. First and second derivative spectral treatments did not improve classification results for 3-week-old infested kernels. NIR spectroscopy was also used to predict the degree of cook in products produced by HTST extrusion of corn meal. Corn meal was cooked with a Wenger TX-57 twin screw extruder using screw speeds ranging from 250 to 350 rpm, and moisture contents ranging from 13-20%, providing a wide range of pressures and shear conditions in the extruder barrel. Extruded samples were analyzed using reference methods that measure different aspects of cooking, including water absorption index (WAI), water solubility index (WSI), viscosity profile as measured with a Rapid Viscoanalyzer (RVA), hardness and fracturability as measured by Texture Profile Analysis. Calibrations for each parameter were developed using multiple linear regression (MLR) and partial least squares (PLS) regression. Correlations with r-value\u3e0.95 were achieved between the NIR and laboratory values. Relative predictive determinant (RPD) values ranged from 5.3 to 6.3 for the various parameters (except for hardness, and trough viscosity) indicating that the NIR measurements should be useful in quality control applications