Recent advances in plant disease severity assessment using convolutional neural networks

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Plant disease severity significantly impacts crop yield and quality.
• Accurate disease severity assessment is vital for effective agricultural monitoring and control.
• While deep learning, particularly Convolutional Neural Networks (CNNs), excels in plant disease identification, research on severity assessment remains limited.

Purpose of the Study:

• To establish criteria for grading plant disease severity based on expert consensus.
• To systematically review and analyze existing studies on CNN-based plant disease severity assessment.
• To identify challenges and propose future research directions for practical applications.

Main Methods:

• Literature review of prevailing views on disease severity grading.
• Categorization and analysis of 16 CNN-based studies, including classical CNNs, improved architectures, and segmentation networks.
• Investigation of dataset acquisition methods and performance evaluation metrics for CNN models.

Main Results:

• A comparative analysis of the advantages and disadvantages of various CNN architectures for disease severity assessment.
• Identification of common dataset acquisition strategies and evaluation metrics.
• Discussion of limitations and challenges in applying CNN-based methods in real-world agricultural settings.

Conclusions:

• CNNs show significant potential for plant disease severity assessment, complementing identification tasks.
• Standardized criteria and robust methodologies are needed for reliable severity grading.
• Addressing practical challenges requires further research into data acquisition, model generalization, and interpretability.