Wheat grain colour analysis by digital image processing II. Wheat class discrimination

Abstract

The extent to which wheat grain colour, objectively measured by video colorimetry, can be used to distinguish kernel type according to wheat class and variety was investigated. Discriminant analyses were performed based on mean red (R), green (G) and blue (B) pixel reflectance (tristimulus) features obtained by colour digital image analysis. Colour data collected from individual kernels of 10 cultivars representing six Canadian wheat classes were used to develop discriminant models. Pairwise discrimination between selected varieties representing the different wheat classes was achieved with considerable success. Over all pairwise trials, 88 % correct varietal classification was achieved on average. In pairwise trials between certain red-grained varieties (cvs), e.g. Neepawa, Norstar or Glenlea and amber durum cv. Wakooma or white spring wheat cv. Owens, correct classification exceeded 96%. A more demanding discrimination problem of correctly classifying grain of single wheat varieties according to official grade class was also posed. Correct classification scores for individual varieties varied from 34 to 90%. Average correct classification scores for the Canada Western Soft White Spring, Amber Durum and Red Spring classes of wheat were 76, 76 and 62 %, respectively. Relatively lower scores of 56 and 34 % were achieved for the Canada Western Hard Red Winter and Canada Prairie Spring wheat classes. The average correct classification for hard red spring type kernels was approximately 90 %.