| An optical radiation measurement system was used to measure reflectance spectra of single wheat kernels from 400 to 2000 nm. Six classes of wheat were used for this study. The overall objectives of this study were to develop a precise and nondestructive method to identify genetically red and genetically white wheats using visible and near-infrared reflectance spectroscopy. Partial least squares (PLS), multiterm linear regression (MLR), principal component regression (PCR), and artificial neural networks (ANN) were used to develop models for identifying single wheat kernel color class.;In the L*a*b* color space, the white wheat (zero red genes) had larger L and b values and smaller a values than red wheat. The a value of red wheats increased as the number of red genes increased. This indicates the redness of wheat kernels increased as the number of red genes increased. There was a linear relationship between the degree of red pigmentation and the number of red genes resulting in a ;For PLS models, the highest classification accuracy was 98.8% obtained using 500-1700 nm wavelength region. For MLR models, the highest classification accuracy of 98.1% was obtained in the wavelength region covered by both the visible and the near-infrared regions. For PCR models, the highest classification accuracy was 98.3% in the wavelength region of 500-1700 nm. For ANN models, the highest classification accuracy was 98.8% with Log (1/R). Among the four classification methods, PLS and ANN models yielded the highest classification accuracy, and these two methods are recommended for single wheat kernel color classification.;The amount of radiation reflected by wheat kernels increased as kernel size increased. The predicted values of red and white kernels increased as kernel size increased. The kernel color varies with kernel orientation. The dorsal side had more effect on the wheat color classification than the crease side and the side view. |
