Abstract

In order to quickly and accurately determine the laver’s harvest time, we adopt combination of modified uninformative variable elimination, successive projection algorithm and visible-near infrared spectroscopy (Vis-NIR) technology to achieve this goal; as mass of spectral data with noise cannot build a stable and efficient recognition model, the effective wavelength should be extracted from the whole spectra. Modified uninformative variable elimination (Muve) algorithm was used to eliminate uninformative variable and noise, successive projection algorithm (SPA) was used to eliminate relevant redundant information, and the remaining variables of 19 were obtained. Finally, the remaining 19 variables were used to establish recognition model using partial least squares vector machine (LS-SVM), and satisfactory prediction rate of 96.67% was obtained. Meanwhile, compared to other traditional variable selection algorithms, such as genetic algorithm (GA) and simulated annealing (SA) algorithm, the proposed algorithms have more advantages.

Key words: Laver, visible-near infrared spectroscopy (Vis-NIR), uninformative variable elimination (UVE), successive projection algorithm (SPA).