On-line NIR predicting soluble solids content of intact pears combination with wavelet transform and support vector regression

The objective of this paper was to predicting soluble solids content of intact pears using on-line near-infrared spectroscopy (NIRS) combination with wavelet transform (WT) and least squares-support vector machine (LS-SVM). Spectra of 200 pears were collected in the wavelength range of 840~950 nm at the speed of 5 pears per second. All samples were divided into two sets: calibration set (n=150) and validation set (n = 50). The spectra were pretreated with the preprocessing method of multiplicative scatter correction (MSC), first derivative (1st D), second derivative (2nd D) and wavelet transform (WT). Partial least squares (PLS) and LS-SVM models were developed with the treated spectra. By comparison the LS-SVM model was super to PLS ones with r of 0.87 and RMSEP of 0. 43oBrix using WT treated spectra. The results indicated that LS-SVM combined with WT could be utilized as a precision method in predicting SSC of intact pears.