Quantification of Mildew Damage in Soft Red Winter Wheat Based on Spectral Characteristics of Bulk Samples: A Comparison of Visible-Near-Infrared Imaging and Near-Infrared Spectroscopy

Abstract

Mildew growth on wheat kernels reduces grain quality due to gray discoloration of kernels that has negative impact on the resulting flour color. Higher levels of mildew damage equate to lower levels of quality. In this study, grain spectra in the 400–1,000 (visible-shortwave-near-infrared) and 1,000–2,500-nm (near-infrared) wavelength ranges were investigated for their ability to quantify mildew damage in Soft Red Winter (SRW) wheat grown in eastern Canada. For each wavelength range, partial least squares (PLS) regression models were developed for various commonly used data pre-treatments. Spectra in the 400–1,000-nm region with a mean-centering pre-treatment were optimal for quantification of mildew damage in SRW wheat. A PLS model using this approach predicted mildew damage in an independent test sample set accurately achieving a root-mean-squared-error of 0.69 and ratio-performance-deviation of 3.84 with the classification accuracy of 96 % for predictions within ±1 level against the trained inspectors’ visual assessment of mildew damage scored on a nine-level scale.