Cascaded wavelength selection method for near-infrared spectroscopy based on simulated annealing bee colony algorithm

Abstract

This article combined the artificial bee colony algorithm with the simulated annealing algorithm to construct the simulated annealed bee colony algorithm (SABC), and combined it with the interval partial least squares (iPLS) to construct a simulated annealing bee colony interval partial least squares algorithm (SABC-iPLS), which was carried out intelligent selection of near-infrared spectral (NIRS) feature intervals. The modeling performance of SABC-iPLS was compared with that of synergy iPLS (SiPLS) and backward iPLS (BiPLS). And then, the SABC was adopted to conduct the secondary optimization of the selected feature intervals of SiPLS, BiPLS, and SABC-iPLS, and three cascaded wavelength selection methods of SiPLS-SABC, BiPLS-SABC, and DSABC-PLS were constructed to eliminate redundant variables in feature intervals. The best modeling performance of DSABC-PLS was further verified by applying corn stover and soil datasets. For validation set of DSABC-PLS models, the determination coefficients of the cellulose in corn stover and organic matter in soil were 0.9437 and 0.9875 with the relative root mean square errors of 1.4616% and 0.7664%, respectively, which could satisfy the actual detection needs. The DSABC-PLS cascaded wavelength selection effectively reduced the variable dimension and model complexity, the prediction ability of the regression model was elevated effectively.