Limitations of single kernel near-infrared hyperspectral imaging of soft wheat for milling quality

Soft wheat milling quality assessment typically begins in the later stages of wheat breeding programmes and continues after cultivar release in commercial milling and processing operations. Near-infrared (NIR) hyperspectral (HS) image analysis, a technique that is gaining interest in food and pharmaceutical inspection research, was explored as an alternative procedure for milling quality evaluation. Three quality properties were studied, flour yield (the weight fraction of flour to whole grain), softness equivalent (a gauge of how easily flour is released from the kernel during break), and sucrose solvent retention capacity (SRC, related to arabinoxylans, which influence water absorption during dough mixing) because of their high degree of heritability and influence on soft wheat quality. NIR HS reflectance images (1000–1700 nm) of non-touching kernels were collected on more than 120 pure cultivars or advanced lines of soft red and white wheat. Five morphological properties (area, elliptical eccentricity and major and minor axis lengths, and ellipsoidal volume) and three spectral properties (principal component scores 1–3) were exhaustively examined in multiple linear regression models for each quality property and in nonparametric classification into low, medium, and high groups using texture properties (contrast, correlation, energy, and homogeneity) calculated from grey-level co-occurrence matrices of principal component scores images. Results indicated that softness equivalent exhibited the highest correlation with HS properties, while sucrose SRC had the lowest correlation. The combination of morphological and spectral properties produced better models than either property group alone. However, because of the inherent chemical and physical complexities of wheat, HS imaging will not be sufficient so as to replace actual pilot milling procedures.