Intelligent Classification Model Design for Tobacco Roasting State Images Based on Deep Learning

In this study, a deep learning model TobALGF based on adaptive fusion of local-global features is proposed to address the key challenges in tobacco roasting state recognition, particularly the complex scenarios encountered in actual production environments. The model realises the dynamic fusion of local texture features and global semantic information through a dual-path network structure combined with LGFA and MSGAF modules, effectively handling the variations in leaf appearance and stacking patterns. Through extensive experimentation and optimization, the results show that on the dataset containing six roasting stages, the model recognition accuracy reaches 90.67%, and the precision and recall rates are 90.74% and 90.66%, respectively. These results significantly outperform existing methods, demonstrating the model's robust performance in real-world conditions and providing reliable technical support for the intelligent monitoring of the tobacco roasting process.