Useful Tips for Constructing a near Infrared-Based Quality Sorting System for Single Brown-Rice Kernels

Abstract

Information required to develop a high-speed single kernel sorting machine for brown rice is presented. The effects of wavelength region, sample illumination geometry, kernel position and calibration method were examined. To obtain a high-accuracy system, the long wavelength region from 1100 nm to 1800 nm in transmittance mode should be used. An appropriate geometry for spectral acquisition and an alerting system for improper kernel position using principal component analysis are recommended. The choice of calibration methods, partial least squares or multiple linear regression, with similar accuracy is provided. Using the system proposed, sufficiently accurate calibration equations with standard errors of prediction of 0.24% w/w dry basis for moisture and 0.40% w/w dry basis for protein could be obtained.