A soft-sensing method for corn composition content using NIRS and LS-SVR

Abstract

A soft-sensing method for oil, protein and starch content in the corn is developed using near-infrared reflectance spectroscopy(NIRS) and least square support vector regression(LS-SVR) techniques, and the feasibility of using different NIR spectrometers for analysis is also examined. Firstly, 90 corn samples are scanned using NIR spectrometers. Then, the original NIRS are processed with multiplicative scatter correction(MSC), Savitzky-Golay second derivative analysis and principal component analysis(PCA). Finally, the soft-sensing model for corn composition content is built using LS-SVR algorithm. The research results show that correlation coefficient (Rc) of NIRS calibrated and actual oil, protein and starch content measured by chemical method are 0.947, 0.969 and 0.948 respectively. It is proved that soft-sensing method has strong robustness for agricultural products.