Milling quality prediction from single maize (Zea mays L.) kernels using the MicroNIR spectrophotometer

Milling quality of maize is a critical property that defines its suitability for milling and end-product quality. Plant breeding programmes have the opportunity to improve this characteristic and the evaluation of the success of breeding efforts is best performed on a single-kernel basis. X-ray micro-computed tomography (µCT) volume and density measurements, i.e. the vitreous-to-floury endosperm ratio (V:F), kernel density (KD) as well as the percentage vitreous endosperm (%VE) as obtained using near infrared (NIR) hyperspectral imaging (HSI) are three descriptors of milling quality. Milling quality as defined by percentage chop (%chop) correlated with these descriptors (r = -0.87, -0.61 and -0.72, respectively for V:F, KD and %VE). These quantification methods are, however, not ideal for fast screening as the availability of instruments are limited and experienced technicians are needed, along with tedious data analysis. The MicroNIR is a portable spectrophotometer designed for hand-held applications and suitable for single kernel analysis. The aim of this study was to investigate the use the MicroNIR to predicting maize milling quality from single kernels. NIR spectra of 297 individual maize kernels were acquired germ-up as well as germ-down, using a MicroNIR 1700. The maize kernels were placed, one at a time, in a hollowed-out Teflon (PFTE) disk was completely covered with the MicroNIR’s connectable collar during scanning. In order to keep the optimal focal distance of 3 mm constant, in spite of maize kernels differing in thickness, a range of hollowed-out disks were developed, varying in depth from 5 mm to 10 mm. MicroNIR software was used to perform the spectroscopic measurements and the data was saved in Microsoft Excel format for analysis in Unscrambler v.10.3 software. PCA and PLS regression modelling, using mean-centered and SNV pre-treated spectra (n = 594) were performed using spectra from 1000 to 1680 nm. PCA scores plot shows clear clusters of good and poor milling samples. Prediction statistics for %chop, V:F, KD and %VE were: R2V = 0.51, 0.41, 0.6 and 0.51; RMSEP = 3.56%, 0.85, 0.03 g.cm-3 and 8.18%; and RPD = 1.44, 1.35, 1.67 and 1.44, respectively. The kernel density models indicated application possibilities on a single kernel basis, as predictions suitable for screening were achieved. Although accuracy needs to be improved with further work, the MicroNIR offers potential to replace the current hugely expensive equipment (X-ray CT scanner, NIR-HSI system) required to determine milling quality from single maize kernels.