In-situ and non-destructive monitoring of chemical components in tobacco leaf during curing based on multimodal data and machine learning

Precise, in-situ, and non-destructive monitoring of chemical composition is critical for optimizing tobacco curing, yet remains challenging under complex and dynamic curing conditions. In this study, a multimodal online monitoring framework was developed by integrating visible-light imaging with temperature and relative humidity sensing to predict key chemical components during the curing process. Colorimetric features extracted from RGB, Lab, and YUV color models, together with environmental parameters, were used as inputs to machine learning models trained on datasets collected from two representative tobacco-growing regions, and external data was to test the models’ performance and robustness. The proposed framework achieved high predictive accuracy for carbohydrates and selected amino acids, with validation R 2 values exceeding 0.90 and satisfactory performance on external test data. SHapley Additive exPlanations (SHAP) analysis further revealed that relative humidity and V value (chrominance component in the YUV color model) were the dominant predictors for most components, while temperature played a key role in alanine variation. These results demonstrate the feasibility of using multimodal visual and environmental data for low-cost, in-situ estimation of chemical composition during tobacco curing, providing a practical tool for process monitoring and quality optimization. • In-situ multimodal acquisition system integrates image, temperature, and humidity monitoring during tobacco leaf curing. • Machine learning-based multimodal data analysis enables accurate prediction of chemical composition. • SHAP interpretation reveals the contribution of multimodal features to prediction models. • Visualization of compositional distribution captures dynamic changes throughout the curing process.