Predicting brix of intact pears by a portable NIR spectrometry with LS-SVM

Abstract

A portable near-infrared (NIR) spectrometry was developed to predict brix of intact pears nondestructively. The spectra of 190 pears collected in the wavelength range of 800–950nm, were dealt with the preprocessing methods of smooth, derivative and standard normal transformation (SNV). The calibration models for brix were developed by least squares support vector machines (LS-SVM), partial least squares (PLS) and multiple linear regression (MLR) with the calibration set of 135 pears. 45 remaining samples were used to evaluate the performance of them. Meanwhile, the capabilities of LS-SVM with different kernel function (RBF_kernel and lin_kernel) were comparatively analyzed. By contrast, the combination of LS-SVM with the RBF kernel, SNV preprocessing and PLS latent variables gave more excellent predictions of brix in the pears, with coefficients of correlation (r) and standard error of prediction (SEP) of (0.87, 0.48°Birx).The results showed that the portable NIR combination with LS-SVM was a feasible method to predict brix of intact pears nondestructively.