DINOV2-FCS: a model for fruit leaf disease classification and severity prediction

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate fruit disease severity assessment is vital for optimizing fruit production.
• Current machine learning methods for disease prediction face challenges in accuracy and generalizability.
• Large vision model technology offers potential for improved agricultural applications.

Purpose of the Study:

• To develop an advanced model for fruit leaf disease classification and severity prediction.
• To leverage the DINOV2 visual large vision model for enhanced feature extraction.
• To address limitations in current models regarding accuracy and generalization.

Main Methods:

• Constructed the DINOV2-Fruit Leaf Classification and Segmentation Model (DINOV2-FCS) using the DINOV2 visual large vision model backbone.
• Proposed the Class-Patch Feature Fusion Module (C-PFFM) to integrate local and global features for improved classification of similar leaf spots.
• Introduced Explicit Feature Fusion Architecture (EFFA) and Alterable Kernel Atrous Spatial Pyramid Pooling (AKASPP) to enhance segmentation of fine disease spots.

Main Results:

• Achieved 99.67% accuracy in disease classification and 95.68% accuracy in disease severity classification on a five-fruit dataset.
• Demonstrated strong generalizability with 83.95% mIoU and 95.24% accuracy in disease severity grading across four datasets.
• Outperformed existing state-of-the-art models in both accuracy and generalization capabilities.

Conclusions:

• The DINOV2-FCS model offers a significant advancement in fruit leaf disease classification and severity prediction.
• The model exhibits robust performance and strong generalization, making it suitable for diverse fruit types.
• This research provides a valuable new tool for agricultural disease management and research.