Prediction of the level of astringency in persimmon using visible and near-infrared spectroscopy

Early control of fruit quality requires reliable and rapid determination techniques. Therefore, the food industry has a growing interest in non-destructive methods such as spectroscopy. The aim of this study was to evaluate the feasibility of visible and near-infrared (NIR) spectroscopy, in combination with multivariate analysis techniques, to predict the level and changes of astringency in intact and in the flesh of half cut persimmon fruits. The fruits were harvested and exposed to different treatments with 95% CO2 at 20 °C for 0, 6, 12, 18 and 24 h to obtain samples with different levels of astringency. A set of 98 fruits was used to develop the predictive models based on their spectral data and another external set of 42 fruit samples was used to validate the models. The models were created using the partial least squares regression (PLSR), support vector machine (SVM) and least squares support vector machine (LS-SVM). In general, the models with the best performance were those which included standard normal variate (SNV) in the pre-processing. The best model was the PLSR developed with SNV along with the first derivative (1-Der) pre-processing, created using the data obtained at six measurement points of the intact fruits and all wavelengths (R2 = 0.904 and RPD = 3.26). Later, a successive projection algorithm (SPA) was applied to select the most effective wavelengths (EWs). Using the six points of measurement of the intact fruit and SNV together with the direct orthogonal signal correction (DOSC) pre-processing in the NIR spectra, 41 EWs were selected, achieving an R2 of 0.915 and an RPD of 3.46 for the PLSR model. These results suggest that this technology has potential for use as a feasible and cost-effective method for the non-destructive determination of astringency in persimmon fruits.