Nondestructive Analysis of Internal Quality in Pears with Self-made Near Infrared Spectrum Detector Combined with Multivariate Data Processing

Taste index (TI) is the ratio of soluble solid content (SSC) to titratable acidity (TA). This index can be used to determine the taste and ripening stage of pears and characterize their flavor. The objective of this paper was to use a portable near-infrared (NIR) spectrometer and multivariate data analysis algorithms for the nondestructive detection of the TI of pears. For calculation convenience, a global numerical transformation of the TI value was formulated. The optimal partial least squares regression (PLSR) model was based on the new value of TI (nVTI) at a TA percentage (PTA) of 90% and a TI percentage (PTI) of 10%, and the model exhibited good performance. The PLSR model based on full variables without pretreatment had the best results. However, the full-spectrum PLSR model contained 397 variables, and the large number of variables led to overfitting of calibration models; therefore, it was necessary to select effective wavelengths from full wavelengths. As for the TI property measurement, a genetic algorithm (GA) was used to select the characteristic wavelengths for measuring TI based on the NIR spectrum obtained by an NIR spectrometer. The number of useful variables selected by the GA was 81. The numbers of characteristic variables selected by the GA-SPA, GA-MCARS, and GA-BOSS were 19, 16, and 22, respectively. The PLSR models based on the variables selected by the GA-BOSS had the smallest number of partial least squares components, demonstrating the good performance of the model. The Rc, RMSECV, Rp, RMSEP, and RPD of the PLSR model based on the useful variables selected by the GA-BOSS method were 0.971, 0.074%, 0.983, 0.066%, and 4.91, respectively. The results show that the portable NIR spectrometer with multivariate data processing is a good tool for the rapid and nondestructive analysis of TI in pears.