The objective of this study was to quantify the apparent characteristic values of tobacco leaves during the intensive curing process, to grasp the law of changes in water loss, and to achieve precise control of curing process parameters during the curing process.

Using the middle leaves of CB-1 as the material, the image and weight data of tobacco leaves during the curing process were collected, and 10 colour and texture features of tobacco leaves were extracted by image processing technology. The analysis took the optimised tobacco leaf characteristic value as the input variable, and used the grid support vector machine (GS-SVM), the genetic algorithm optimized BP neural network (GA-BP) and the extreme learning machine (ELM) algorithm to establish the tobacco leaf loss during the curing process. Water rate prediction was modelled, and evaluation and analysis of prediction accuracy and model verification and selection was carried out.

According to the feature variable clustering and correlation analysis, two colour features (a*/b*, R) and two texture features (Gradient entropy, gradient unevenness), the test set root mean square error (RMSE) of the established grid SVM and GA-BP neural network and ELM model are 0.0117, 0.0139, 0.0140 respectively, and the test set determination coefficient (R²) are 0.9973, 0.9962, 0.9961, respectively. The GS-SVM model had the highest accuracy.

The prediction model of the moisture loss rate of tobacco leaves during the intensive curing process established by machine learning meets the needs of real-time detection, and the GS-SVM model has the best prediction effect. The research results can provide a basis for the optimisation of intensive curing technology, and lay a theoretical foundation for the development of an intelligent control system for tobacco curing.