HYBRID VARIABLE SELECTION IN VISIBLE AND NEAR-INFRARED SPECTROSCOPY FOR NON-INVASIVE CLASSIFICATION OF SPIRULINA POWDERS

摘要

Spirulina are free-floating filamentous cyanobacteria which are cylindrical, multicellular trichomes in an open left-hand helix. Spirulina is favored as a human dietary supplement as well as a whole food. As it contains an unusually high amount of protein, between 55% and 77% by dry weight, composed with all essential amino acids, and easy to be digested, Spirulina powder is welcome by consumers. As the specific protein content and amino acids proportions depend upon the source, it is important to classify Spirulina powder from different species and manufacturers. This paper evaluated the feasibility of using visible and near infrared (Vis-NIR) spectroscopy for the non-invasive discrimination of Spirulina platensis powder. Full-spectrum obtained a 93.33% correct answer rate (CAR) for the discrimination of three varieties. Several spectral wavelength variable selection algorithms were operated. A hybrid variable selection algorithm of interval partial least squares (iPLS) and successive projections algorithm (SPA) obtained the optimal model with 96.67% of CAR for the prediction set, in which iPLS obtained 135 variables for SPA calculation. Compared to the SPA calculation on the full-spectrum, visible spectra or NIR spectra, iPLS can better reduce the calculation time of SPA. These results show the possibility for the non-invasive discrimination of Spirulina platensis powder using Vis-NIR spectroscopy, and iPLS-SPA is a good spectral variable selection algorithm.