Research on the Classification Method of Fresh Tobacco Leaf Maturity Based on Transfer Learning and Multi-Feature Fusion

Addressing the challenges of diversity, variability, and insufficient generalization in identifying the maturity of fresh tobacco leaves in field conditions, this study proposes a classification method based on transfer learning and multi-feature fusion. Initially, the EfficientNetB0 model is employed for transfer learning, efficiently extracting deep features of fresh tobacco leaves by freezing certain layers and fine-tuning the higher-level network. Subsequently, traditional features such as color, texture, and pixel characteristics are extracted, and features significantly correlated with maturity are selected through correlation analysis. Finally, a dynamic feature weighting mechanism is introduced to adaptively fuse deep features with traditional features, optimizing weight allocation to enhance model performance. The results demonstrate that the model achieves a 91.87% accuracy in fresh tobacco leaf maturity recognition, significantly improving the accuracy and generalization ability of maturity identification. This method provides an efficient and robust solution for recognizing tobacco leaf maturity in complex field environments and explores its application in automatic harvesting systems, promoting the advancement of intelligent agricultural technology.