Non-destructive prediction of soluble solids content of pear based on fruit surface feature classification and multivariate regression analysis

Improving the prediction accuracy of SSC is the unremitting pursuit in the field of nondestructive optics. The Vis/NIR transmission spectra of 'Korla' pear were collected with a portable spectrometer instrument developed by ourselves. In order to study the effects of fruit surface feature such as peel color on determination of SSC and classification of surface locations, three types of SSC prediction models (separate location model, global locations model and average spectra model) and four types of surface location classification models (spectra regions: 550-950, 550-780, 780-950 and 550-700 nm) were built based on the full wavelengths (FWs) and effective wavelengths (EWs), respectively. Results showed that the prediction model of EWs-separate location achieved a best result, the correlation coefficient of prediction set (Rpre) were 0.9408 and 0.9463 for sunlit and shaded side samples, respectively, meanwhile, the classification model based on the 68 EWs selected from 550 to 950 nm achieved an optimal result with the correct classification rate of 97.78% and 96.67% in calibration and prediction sets, respectively. Overall mentioned results above illustrated the fruit surface feature was sensitive to the models of SSC prediction and fruit location classification. Therefore, a compensation model of SSC prediction that is robust and accurate, as well as insensitive to fruit surface feature was built by combining the EWs-classification model with EWs-separate location prediction model, the Rpre and root mean square error of prediction were 0.9368 and 0.5256 °Brix, respectively. In addition, the EWs-global locations model exhibited a negligible effect on the surface feature, although its prediction accuracy had a little inferior to the optimal compensation model. Hence, a complex compensation model with higher prediction accuracy of SSC showed a considerable potential for portable spectrometer instrument.