Abstract: To promote the industrial classification of raw tobacco and to realize automation production and intelligent sorting of tobacco leaves, a binary-classification sorting approach for industrial classification of tobacco leaves was studied. Directed at the folded and wrinkled leaf morphologies, multimodal fusion of multi-angle imaging (reflected light, transmitted light, and localized reflected imaging) and origin information was performed. The binary-classification sorting model was applied to an automatic production line for industrial classification of raw tobacco, and comparison of the proposed model with different models and production verification were conducted. The results showed that: 1) The tobacco leaf algorithm model based on the multimodal fusion distinctively outperformed single-feature models and general convolutional neural network (CNN) models based on three evaluation metrics: sorting accuracy, precision, and recall rate. 2) Under the experimental conditions of this study, the grade qualification rate of the sorted tobacco leaves exceeded 85.2% with a relative standard deviation (RSD) of 1.4%, indicating high accuracy and stability that could meet the sorting requirements of production. 3) A comparative analysis between the machine sorting and manual sorting revealed that the efficiency of the machine sorting was 50 times as much as that of manual sorting, and the grade qualification rate was 7 percentage points higher than manual sorting. Additionally, the machine sorting could effectively eliminate the impact of environmental factors and human fatigue in manual sorting. The multimodal fusion model for binary-classification sorting of tobacco leaves established in this study therefore demonstrates satisfactory practical value for industrial application.