Research on tobacco leaf grading algorithm based on transfer learning

In order to improve the accuracy and speed of tobacco leaf classification and the problems of insufficient tobacco leaf samples, this paper studied the classification algorithm of 41 kinds of tobacco leaf pictures based on deep learning method. In order to obtain better grading results even with a small number of samples, transfer learning was applied to the classic VGG16 network model, and the VGG16 network structure was fine-tuned. 1498 data sets were used. Finally, the experimental results showed that the number of trainable parameters decreased with the increase of the number of frozen layers. As for accuracy, when only the convolutional structure part is frozen, the model has the highest accuracy, reaching 91.26%, which is 1.25% higher than the original model.