PLS Modeling the Starch Contents of Corn Data Measured Through Different NIR Spectrometers

Abstract

A variety of filter wavelength region selection algorithm, including loading weight PLS (PLS-LW), regression coefficient PLS (PLS-RC), variable importance on PLS (PLS-VIP) and selectivity ratio PLS (PLS-SR) and significant multivariate correlation (PLS-SMC) are considered in modeling the starch contents of corn with corn spectral data. Corn samples were measured on three different NIR spectrometers known as M5, Mp5 and Mp6. Hence, the class of filter PLS methods were imposed on each data set obtained from different spectrometers. Filter PLS can select influential wavelength region of spectral data, through Leave-One-Out (LOO) cross validation procedure. The performance of each fitted PLS on each spectrometer data set was measured with root mean square error for prediction (RMSEP), which reveals the PLS-SR (pvalue=0.001) and Mp6 (p-value=0.073) select the wavelength region which best explains the variation in starch corn contents.