Non-invasive Classification of Laver Using Visible and Near-infrared Spectroscopy

Abstract

Visible and near infrared (NIR) spectroscopy was utilized to classify the verities of laver. As there are almost six hundreds of NMR variables which would cause poor classification and long calculation time, uninformative variables should be eliminated. Successive projections algorithm (SPA) was applied to select the effective variables from the full-spectrum (FS). Finally 13 variables were selected, and were inputted into least-square support vector machine (LS-SVM) to do the classification. A better result of 96.55% correct answer rate of SPA-LS-SVM model was obtained, compared to that of the principal component analysis (PCA)-LS-SVM model. It was proved that SPA is effective algorithm for spectra variable selection. As a conclusion, Vis-NIR spectroscopy is a feasible way to distinguish laver varieties fast and accurately.