State recognition and temperature rise time prediction of tobacco curing using multi-sensor data-fusion method based on feature impact factor

Abstract

The quality and economic value of cigarettes depend heavily on the quality of tobacco curing, which is greatly influenced by the temperature rise time during the process. Accurately setting the temperature rise time is crucial in creating the right curing environment to achieve high-quality tobacco leaf curing. However, accurate recognition of the different stages of tobacco curing is a complex process that involves multiple sources of feature data for comprehensive analysis. The impact factor of each feature changes with the curing conditions, making it difficult for existing models to account for errors and adjust to changes in data-fusion schemes. To address these challenges, this study proposes a dynamic data-fusion scheme based on the voting method strategy using the extreme gradient enhancement algorithm and data-fusion method. Additionally, a comprehensive baking prediction model (CBPM) is introduced, which integrates the state recognition fusion model and multiple temperature rise time prediction models to accurately recognize curing states and set the temperature rise time accordingly. CBPM is designed to incorporate curing expertise and can automatically adjust the data-fusion strategy with the change of impact factors of multi-source data to meet the demand for comprehensive data analysis in prediction. Through experimental comparison, it is concluded that CBPM outperforms existing models, with a recognition macro F1-Score of the curing state at 98.4% and a mean absolute error of prediction of the temperature rise time of 0.541. This data-fusion and model integration method can provide effective technical support for other curing production fields.