Comparison of Different Variable Selection Methods on Potato Dry Matter Detection by Hyperspectral Imaging Technology

Abstract:

In order to improve precision determination of dry matter content in potatoes by hyperspectral image technology, several variable selection methods such as PCA, siPLS, GA-PLS, UVE and competitive adaptive reweighed sampling (CARS) were compared. A combinatorial method named CARS-SPA (successive projections algorithm) was proposed to select variables from 678 wavelength variables. The number of wavelength variables was reduced to 27. A multivariate linear regression model (MLR) based on these 27 wavelength variables was developed to predict DM content with Rp of 0.86, and RMSEP of 1.06%. It was concluded that hyperspectral imaging technology could be used to detect potato dry matter concentration and CARS-SPA was a feasible and efficient algorithm for the spectral variable selection. 