Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Prediction of Kernel Density of Corn Using Single-Kernel Near Infrared Spectroscopy

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  Applied Engineering in Agriculture. 28(4): 569-574. (doi: 10.13031/2013.42071) @2012
Authors:   P. R. Armstrong, J. G. Tallada
Keywords:   Corn hardness, Kernel density, NIRS, Partial least squares regression

Corn hardness is an important property for dry and wet-millers, food processors, and corn breeders developing hybrids for specific markets. Of the several methods used to measure hardness, kernel density measurements are one of the more repeatable methods to quantify hardness. Near infrared spectroscopy (NIRS) provides an attractive method to measure kernel density as it can also measure other compositional attributes. Some commercial instruments do measure density of bulk samples. Single-seed NIRS, however, may provide additional information and capabilities by measuring density of individual kernels. This has potential applications for breeders or quality control wishing to look at variance within a sample and for sorting. This study examined the accuracy of NIRS to predict density from single seeds of corn. Absorbance spectra (904 to 1685 nm) were collected on single seeds from 67 food hybrids and 40 commodity hybrids. Moisture adjusted density measurements, using 12-g samples, were made using a gas pycnometer and used as the reference method in the development of the prediction equation. The best prediction model developed from partial least squares regression between averaged spectra and density values had a standard error of cross (SECV) validation of 0.018, coefficient of determination (R2) of 0.79 and the ratio of the standard deviation to the standard error for the cross-validation model (RPD) of 2.1. Predictions for a validation set of 35 samples yielded a standard error of prediction (SEP) equal to 0.016, R2 = 0.76 and the ratio of the standard deviation to the standard error for the cross-validation model (RPD) = 1.9. Other models developed using different spectral pretreatments yielded very similar statistics. Ten samples were subsequently sorted into low, medium, and high density fractions based on spectroscopic predictions. Pycnometer measurements on the fractions verified they were correctly sorted by density and are correlated to starch content (r = 0.42) and oil content (r = -0.39).

(Download PDF)    (Export to EndNotes)