Comparison and Optimization of Models for Determination of Sugar Content in Pear by Portable Vis-NIR Spectroscopy Coupled with Wavelength Selection Algorithm

Abstract

The portable device could help to obtain a complete follow-up of fruit quality in orchards and during post-harvest. Thus, it is an important step to develop portable and non-destructive technology for current and future research in fruit. In this study, the ability of portable visible-near infrared (Vis-NIR) spectroscopy to non-invasively determine sugar content in pear was studied. Partial least square regression (PLSR) was applied to establish calibration models based on the spectral signatures of three regions (550–1050, 650–950, 750–1050 nm) and four types of data sets (Set-I, Set-II, Set-III, and Set-IV), respectively, and the performance of models was compared to determine the optimal spectral calibration strategy. The spectral region of 650–950 nm was proved to be much better compared with other two spectral regions. Competitive adaptive reweighted sampling (CARS) algorithm was used to reduce redundancy and collinearity of the original spectral data based on the optimal spectral region for selecting the most important wavelengths. The CARS-PLSR was identified as the most effective method to calibrate the prediction models for sugar content determination, resulting in good coefficient of determination for prediction ( $$ {R}_P^2 $$ ) of 0.85–0.92 and root mean square error of prediction (RMSEP) of 0.27–0.20 for four types of data sets, respectively. The overall results show that the portable Vis-NIR spectroscopy is a promising tool for the non-destructive on-site evaluation of sugar content in pear, as well as affording the additional advantage of low cost.