Digitalization of wheat mold odor based on controlled volatile release: A large-scale study.

This study presents a large-scale framework for the digitalization of wheat mold odor based on controlled volatile release and standardized gas acquisition. Seven wheat-varieties harvested from 2022 to 2025 in three provinces of China were cultivated into four spoilage levels (normal, mild, moderate, and severe mildew), and a graphene-based sensor array was developed for multidimensional odor detection. A dual pre-treatment strategy integrating temperature-regulated volatilization and cooling-assisted dehumidification was implemented to ensure stable and comparable signal acquisition. Under optimized conditions, 1491 odor response curves from the first three batches were used for machine learning modeling, while an independent fourth batch (n = 503) was used for external validation. For binary classification (normal vs. moldy wheat), the optimized Light Gradient Boosting Machine achieved 95.8% accuracy, 95.5% sensitivity, 96.8% specificity, and an AUC of 0.984. This proposed approach enables rapid, non-destructive mold assessment and supports standardized grain quality monitoring.