The increasing demand for high-quality grapes necessitates rapid and objective quality assessment methods to overcome the limitations of traditional subjective and destructive techniques. This study investigated the feasibility of using hyperspectral imaging combined with machine learning for non-destructive quality evaluation of fresh grapes. Hyperspectral data were acquired from four table grape varieties (“Rose”, “Yongyou”, “Xiahei”, and “Jumbo”), and their Soluble Solids Content (SSC) was measured, which varied significantly among varieties. We extracted texture features using the Gray-Level Co-occurrence Matrix (GLCM) from images at key wavelengths, which were a combination of those selected by the Successive Projections Algorithm (SPA) and sensitive wavelengths. Comparative models for variety classification (qualitative) and SSC prediction (quantitative) were built using Extreme Learning Machine (ELM), Convolutional Neural Network (CNN), and Partial Least Squares (PLS) with full-range spectra and texture features as inputs. The results showed that the ELM model using full-range spectra was superior for both tasks, achieving a classification accuracy of 97.56% and, for SSC prediction, an Rp2 of 0.75 and an RMSEP of 0.81. Notably, CNN models also showed considerable robustness. Our findings confirm that combining hyperspectral imaging with machine learning is a viable strategy for fresh grape quality assessment.
Chen et al. (Fri,) studied this question.