The Optimal Local Model Selection for Robust and Fast Evaluation of Soluble Solid Content in Melon with Thick Peel and Large Size by Vis-NIR Spectroscopy

Abstract

Soluble solid content (SSC) is one of the most important factors determining the quality and price of fresh fruits. However, the accurate assessment of SSC in some kinds of fruits with thick peel and large size is relatively difficult because it is important to select a suitable measurement position on this type of fruit. In this study, ‘Hami’ melon was used as the object of study, the visible and near-infrared (Vis-NIR) spectroscopy with spectral range of 550–950 nm was acquired from three positions (calyx, equator, and stem) of each sample to observe the effect of measurement positions on SSC assessment of whole ‘Hami’ melon. Three local models (calyx-region, equator-region, and stem-region models) and one global model based on the partial least squares (PLS) were developed with different preprocessing methods. Comparing all the established models, the results showed that the equator-region model and global model had the similar predictive performance which was better than ones of calyx-region and stem-region models. For improving the performance of models, the equator-region model and global model were further optimized based on different variable selection algorithms, including competitive adaptive reweighted sampling (CARS), uninformative variable elimination (UVE), combination algorithm CARS-SPA (successive projections algorithm), and combination algorithm UVE-SPA, respectively. And also, the linear multispectral PLS models and the nonlinear multispectral least squares support vector machine (LSSVM) models were established and compared using those selected characteristic variables, respectively. The results indicated that the performance of equator-region multispectral models was slightly superior to those of global multispectral models, and the optimal equator-region multispectral models were UVE-SPA-PLS (RP = 0.9143 and RMSEP = 0.8359) and CARS-SPA-LSSVM (RP = 0.9134 and RMSEP = 0.8958). The overall results indicated that it was feasible to develop the models using only the equator position information for detecting the SSC of whole ‘Hami’ melon. This study can provide some valuable references for building a fast and robust multispectral prediction model for SSC assessment in some kinds of fruits with thick peel and large size such as watermelon.