Tobacco Curing State Recognition Based on a Multi-Sensor Feature Fusion Approach

The accurate identification of tobacco curing status is a complex process involving comprehensive analysis of multiple types of feature data sources. The influence of each type of feature on the state recognition model varies with the change in the curing process, and the existing models cannot reasonably utilize these feature data, which makes it difficult to improve the model recognition accuracy. To address these challenges, this paper proposes a dynamic data fusion scheme based on clustering methods and uses the LightGBM algorithm and data fusion method to build a state recognition fusion model that integrates six state recognition sub-models, which can accurately identify the curing status. SRFM aims to integrate curing expertise and automatically adjust the data fusion strategy according to the changes in multi-source data to meet the needs of comprehensive data analysis in recognition. Through experimental comparison, SRFM outperforms the existing models, with an accuracy of 99.8% in identifying the curing status and a macro F1-Score of 99.9%.