A rapid and nondestructive quantitative detection method for total and organic selenium in selenium-enriched kefir grain based on hyperspectral imaging

Abstract

Rapid monitoring of total and organic selenium content of Kefir grain is important for the development of selenium-enriched foods. First, a portable hyperspectral system was used to obtain spectra of selenium-enriched Kefir grain. Second, the characteristic variables were screened based on competitive adaptive reweighted sampling (CARS), bootstrapping soft shrinkage, genetic algorithm, and particle swarm optimization. Then, the quantitative prediction performance of extremely randomized trees, partial least squares (PLS), and least squares support vector machine models were compared. For the prediction of total selenium content, CARS-PLS performed the best, with a prediction set correlation coefficient and relative prediction deviation (RPD) value of 0.97 and 3.88, respectively. For organic selenium, the CARS-PLS predicted set correlation coefficients and RPD values were 0.94 and 2.97, respectively. CARS-PLS realized the rapid prediction of total and organic selenium content within 1 min. This provides new ideas for screening selenium-enriched microorganisms and rapid detection of selenium content.