A precision grading method for walnut leaf brown spot disease integrating hierarchical feature selection and dynamic...

Area of Science:

• Plant Pathology
• Computer Vision
• Smart Agriculture

Background:

• Walnut leaf brown spot, caused by *Ophiognomonia leptostyla*, is a major threat to walnut cultivation.
• Accurate grading of plant diseases is crucial for smart agriculture but challenged by blurred lesion edges and complex feature extraction.

Purpose of the Study:

• To develop an advanced model for precise grading of walnut leaf brown spot disease.
• To address limitations in current disease grading methods, particularly concerning blurred lesion edges and feature extraction efficiency.

Main Methods:

• Proposed a novel disease grading method integrating hierarchical feature selection and adaptive multi-scale dilated convolution.
• Developed the CogFuse-MobileViT model with three key modules: Hierarchical Feature Screening Module (HFSM), Edge Feature Focus Module (ECFM), and Adaptive Multi-Scale Dilated Convolution Fusion Module (AMSDIDCM).

Main Results:

• The CogFuse-MobileViT model achieved 86.61% accuracy on the test set.
• Demonstrated a 7.8 percentage point improvement over the standard MobileViTv3 model.
• Significantly outperformed other mainstream disease grading models in experimental evaluations.

Conclusions:

• The CogFuse-MobileViT model effectively resolves challenges of blurred edges and inefficient feature extraction in walnut leaf brown spot disease grading.
• Provides a reliable technical solution for precision grading and holds practical value for intelligent plant disease diagnosis in smart agriculture.