An Efficient Model for Leafy Vegetable Disease Detection and Segmentation Based on Few-Shot Learning Framework and...

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Leafy vegetable diseases pose a significant threat to food security.
• Accurate and early disease detection is crucial for effective crop management.
• Existing methods struggle with complex backgrounds and limited disease sample data.

Purpose of the Study:

• To develop an advanced model for leafy vegetable disease detection and segmentation.
• To address the challenges of complex backgrounds and few-shot learning in agricultural disease identification.
• To enhance the robustness and fine-grained feature expression of disease detection models.

Main Methods:

• A few-shot learning framework combined with a prototype attention mechanism.
• Development of a prototype loss function to optimize sample-prototype distance relationships.
• Comparative analysis against traditional methods like YOLOv10 and TinySegformer.

Main Results:

• Achieved high performance in object detection (e.g., 0.93 precision, 0.91 mAP@50).
• Demonstrated excellent results in semantic segmentation (e.g., 0.95 precision, 0.92 mAP@50).
• Significantly outperformed traditional methods, validating the prototype attention mechanism's effectiveness.

Conclusions:

• The proposed model shows superior performance in leafy vegetable disease detection and segmentation, especially in few-shot and complex background scenarios.
• The prototype attention mechanism and loss function enhance model robustness and category discrimination.
• The method holds significant potential for practical applications in agricultural disease management.