COMPARISON OF NEAR-INFRARED SPECTROSCOPY CALIBRATION METHODS FOR THE PREDICTION OF PROTEIN, OIL, AND STARCH IN MAIZE GRAIN

This study compared the accuracy of protein, oil, and starch content prediction in maize grain, using near-infrared calibrations developed from three types of spectral data: 1100-2500-nm ground grain diffuse reflectance, 1100-2500-nm whole grain diffuse reflectance, and 680-1235-nm whole grain diffuse transmission. The best constituent predictions were by equations developed from ground grain diffuse reflectance data. Standard errors of prediction were comparable to the standard errors of the reference methods, and r2 valves were all greater than 0.70. Equations developed from whole grain diffuse transmission and, to a greater extent, reflectance spectral data had higher SEP, especially for protein, and significantly lower r2 values for oil and starch prediction than corresponding ground grain equations. The performances of multiterm linear and partial least-squares regression calibration techniques were comparable for each set of spectral data.