Improvement of Soluble Solids Content Prediction in Navel Oranges by Vis/NIR Semi-Transmission Spectra and UVE-GA-LSSVM

The objective of this research is to improve soluble solids content (SSC) prediction in navel oranges by visible/near infrared (Vis/NIR) semi-transmission spectra and uninformative variable elimination-genetic algorithm-least squares support vector machine (UVE-GA-LSSVM). Spectra of navel oranges were acquired using a QualitySpec spectrometer in the wavelength range of 350 similar to 1,000 nm. After applying spectral pretreatment methods, UVE-GA was used to select variables, then LSSVM with three kernel functions (RBF kernel, linear kernel, polynomial kernel) was used to develop calibration models. The results indicate that Vis/NIR semi-transmission spectra combined with UVE-GA-LSSVM has good performance on assessing SSC of navel oranges, and SSC is improved. The R(2)s and RMSEPs of SSC for RBF kernel, linear kernel, and polynomial kernel in prediction set are 0.850, 0.848, 0.849 and 0.419, 0.421, 0.420 %, respectively.