EFFECTS OF APODIZATION FUNCTION, ZERO FILLING, BACKGROUND SPECTRA, AND ABSORBANCE TRANSFORMATION ON MID-INFRARED CALIBRATIONS FOR FEED COMPOSITION

Abstract

ABSTRACT Research has demonstrated that diffuse reflectance mid-infrared spectroscopy can, like near-infrared diffuse reflectance, be used to quantitatively determine the composition of ground samples of forages and soils without the need for KBr dilution. While it has been demonstrated that the accuracy of calibrations developed using mid-infrared spectra can be equal to or better than that achieved using near-infrared spectra, the influence of factors such as apodization function has not been determined. Results based on the spectra of 173 treated forage samples obtained using a DigiLab FTS-60 spectrometer have demonstrated that many parameters associated with mid-infrared spectra have little or no effect on partial least squares calibrations. Additional zero filling of spectra had little effect other than to increase the derivative gaps found to produce optimal calibrations, but calibrations developed using Kubelka-Munk transformed data, as opposed to absorbance data, were not as accurate. Choice of apodization function also had little effect, although slightly better results were found using triangular or weak Norton-Beer. Likewise, the frequency of taking a background spectrum did not seem to have any great effect on calibrations, although results were slightly better with hourly or daily acquisitions as opposed to one for each sample as is done in the near-infrared.