Linking Cucumber Surface Color to Internal Hydration Level Using Deep Learning for Freshness Classification

Postharvest dehydration is a major determinant of cucumber freshness and marketability, yet early reductions in internal water status are difficult to detect using conventional quality assessment methods. This study presents a non-destructive, physiology-informed deep learning approach that links cucumber surface color and texture patterns to internal hydration level for automated freshness classification. A time-resolved dataset comprising 4160 RGB images of cucumber fruits was paired with gravimetrically determined relative water content (RWC), used as an objective indicator of internal hydration status. Based on RWC, fruits were classified into four freshness categories: Very Fresh (≥98%), Moderately Fresh (95–98%), Low Freshness (90–95%), and Spoiled (<90%). A custom convolutional neural network (CNN) was trained using standardized RGB images and evaluated on an independent test set. The model achieved an overall classification accuracy of 91.35% and a Cohen’s Kappa coefficient of 0.875, indicating strong agreement between predicted and actual freshness classes. Classification performance was highest for the extreme freshness states, with F1-scores exceeding 0.94 for Very Fresh and Spoiled fruits, while intermediate classes showed greater overlap, reflecting the gradual nature of postharvest water loss. Model interpretability analyses revealed that the CNN consistently focused on physiologically meaningful surface color and texture features associated with dehydration. Overall, these findings highlight the potential of physiology-informed deep learning to advance non-destructive freshness assessment in cucumbers, offering a realistic pathway toward hydration-based sorting, improved shelf-life management, and intelligent quality monitoring in modern postharvest supply chains.