From physicochemical to sensory quality: Non-destructive profiling and grading of kiwifruit using a self-designed handheld Vis/NIR spectrometer combined with chemometrics.

Abstract

Kiwifruit is a nutrient-rich fruit with high levels of vitamin C and bioactive compounds, contributing to its nutritional value and consumer acceptance. However, as a climacteric fruit, its physicochemical properties and sensory attributes change rapidly during ripening, posing challenges for postharvest quality evaluation. Therefore, rapid and non-destructive assessment methods are essential. In this study, a self-developed handheld visible/near-infrared (Vis/NIR) spectrometer was employed to characterize multiple dimensions of kiwifruit quality. Partial least squares (PLS) models were developed to predict key physicochemical parameters, including flesh firmness (FF), soluble solids content (SSC), and pH, achieving good predictive performance (R2ₚ = 0.915-0.975). Sensory attributes, such as fruity aroma, sweetness, and freshness, were further predicted using deep learning models. The differences in sensory prediction performance were associated with their relationships with physicochemical properties, with more strongly correlated attributes showing better spectroscopic predictability. In addition, a classification strategy based on SSC, FF, and pH was developed for kiwifruit quality grading, where the PCA-RF model achieved the best performance among all evaluated models, with a test set classification accuracy of 97.2%. Overall, the proposed method shows potential for applications in postharvest quality control and consumer-oriented grading.