Accurate grading of flue-cured tobacco leaves is crucial for tobacco quality control and industrial applications. However, traditional manual grading is subjective, inefficient, and labor-intensive, while existing deep learning-based methods often fail to capture multi-scale complementary features, leading to limited grading precision. To address these issues, this paper proposes a novel deep learning framework that integrates multiple pre-trained architectures with hierarchical feature fusion for robust tobacco leaf classification. Specifically, the framework leverages three complementary backbone networks, incorporates a convolutional block attention module to enhance feature discriminability via channel and spatial attention mechanisms, and designs a hierarchical feature fusion module with learnable attention weights to adaptively combine low-level, mid-level, and high-level features. Experimental results confirm that the proposed method achieves superior performance in flue-cured tobacco leaf grading, boasting a remarkable accuracy of 99.95% and effectively capturing the subtle visual characteristics essential for tobacco leaf quality assessment. In conclusion, the proposed architecture provides a comprehensive and reliable solution for flue-cured tobacco leaf grading, with potential applications extended to other fine-grained visual recognition tasks in agricultural product classification.
Pang et al. (Thu,) studied this question.