Cereal Chem 61:158 - 165.  |  VIEW ARTICLE
Optimization of Mathematical Treatments of Raw Near-Infrared Signal in the Measurement of Protein in Hard Red Spring Wheat. I. Influence of Particle Size.

K. H. Norris and P. C. Williams. Copyright 1984 by the American Association of Cereal Chemists, Inc. 

Diffuse reflectance does not vary linearly with absorber concentration. Therefore, if linear correlation techniques are used for near-infrared reflectance spectroscopy, mathematical treatments of the reflectance data are required. Data treatments found to give a linear correlation with protein content of wheat include log 1/R, dR/R, Delta (log 1/R), d2 (log 1/R), and (1-R)2/2R. These data treatments were evaluated for performance in predicting protein content for wheat samples ground with differe nt grinders to give a wide range of particle sizes. The spectral data were also normalized by dividing by the same data treatment at a refence wavelength. Particle size had a marked effect on the near-infrared spectra, and this effect carried through to produce large errors in protein prediction without normalization of spectral data. Normalization of data effectively removed the particle-size effect such that second derivative divided by second derivative at optimum wavelengths gave an average bias error of -0.02% with a standard deviation of bias of 0.12% protein for six sample lots with mean particle size varying from 161 to 327 microm. Normalized second- derivative treatment of data gave the best performance in predicting protein content of samples varying widely in particle size. Inclusion of particle-size variation in the calibration samples improved the performance of all the other data treatments, making several of them equal to the normalized second derivative.