Study on quantitative methods in near infrared spectroscopy analysis

This paper discussed three types of quantitative methods used to build NIR models: PLS (Partial Least Square), BP (Back Propagation Neural Network) and SVM (Support Vectors Machines). One hundred and thirty two unit wheat samples were used as experimental materials. The result indicates that SVM has good performance. There also discussed SVM parameters: the penalty parameter (c) and the kernel parameters (gamma) of RBF (Radial Base Function) kernel, which strongly affect prediction ability of the SVM model. The result indicates that c affects the self-prediction ability of SVM model and gamma affects the prediction ability of SVM model when the other parameters are fixed.