Improved prediction of wheat quality and functionality using near-infrared spectroscopy and novel approaches involving flour fractionation and data fusion

Abstract

The accurate and rapid determination of wheat quality is of great importance for the wheat supply chain. Near-infrared (NIR) spectroscopy has become an established method for this purpose. So far, however, predictions for most wheat quality characteristics are not accurate enough to replace reference measurements, with the exception of protein content. This study investigates the potential to improve the prediction of 41 wheat quality parameters (protein- and starch-related parameters, solvent retention capacity, farinograph, extensograph, alveograph) based on a flour fractionation approach (sieve fractionation, dough preparation, gluten washing) and data fusion using the established techniques of NIR spectroscopy and chemometrics. Results show that preprocessing of flour significantly altered the composition of the samples, which reflected in spectral differences of their NIR spectra. This also led to a change in the prediction accuracy for many wheat quality parameters. Compared to the prediction using flour spectra, flour fractionation with or without data fusion improved the RMSECV between 5.6 and 28.6% for 35 out of the 41 quality parameters tested, leading to R2CV between 0.80 and 0.96 for many of them. Gluten, dough, and the 50–75 µm and the 75–100 µm fractions were particularly important for the improved predictions. The best predictions were often based on data fusion of spectra from different sample types, demonstrating the importance of using complementary information from different data sources to improve predictions. The results underline the potential of this novel approach to be established in the industry as an extension of conventional NIR spectroscopy to improve wheat quality prediction.