Spectra coupled with color features to determine sugar content of fragrant pears using LS-SVM

Visible and near-infrared (VIS-NIR) based technique coupled with color features for determination of sugar content (SC) of fragrant pears was studied. The VIS-NIR reflectance spectra of one hundred fragrant pears were measured by a portable spectrometer in a measurement range between 350 and 1000 nm. SC and color features (CIE L*, a*, b*) were analyzed by the reference methods. The results obtained by least squares support vector machines (LS-SVM) regression with the data fusion of 15 selected wavebands spectral data and color information were higher than that of just using the same 15 wavebands spectral information as input by stepwise multiple linear regression (SMLR) and LS-SVM regression methods, and the best result was obtained by coupled with all color values (L*, a* and b*), with a root mean square error of prediction (RMSEP)= 0.561 °Brix, then coupled with color values (a* and b*), and L* value in sequence. The results suggested that it was technically feasible to detect fruit SC and improve its prediction accuracy by data fusion of color and spectral information.