Vision-Based Multi-Scale Structural Modeling of Water Loss during Morchella Drying

Drying-induced water loss in edible mushrooms is closely associated with structural reorganization, yet quantitative links between surface appearance and dehydration remain unclear. This study developed an interpretable computer-vision framework to predict water loss (WL) of Morchella during hot-air drying using multi-scale visual descriptors. A total of 420 valid RGB images were analyzed across seven drying stages. Morphological features, International Commission on Illumination Lab* (CIELAB) color descriptors, and pore-level structural parameters were extracted to characterize geometric shrinkage, chromatic migration, and cavity evolution. You Only Look Once version 8 (YOLOv8)- and Slicing Aided Hyper Inference (SAHI)-assisted segmentation achieved a mask precision of 0.789, mask recall of 0.798, and mask mAP50 of 0.852. RandomForest with within-sample baseline normalization (X−X₀) achieved the best WL prediction, with R² of 0.912, RMSE of 0.087, and MAE of 0.066. SHapley Additive exPlanations (SHAP) analysis showed that structural features contributed more strongly than color features. These results support vision-guided drying assessment and quality-oriented control in mushroom processing.