Apple leaf disease severity grading based on deep learning and the DRL-Watershed algorithm.

Area of Science:

• Agricultural Science
• Plant Pathology
• Computer Vision

Background:

• Apple leaf diseases reduce crop yield and quality.
• Current disease detection methods lack accuracy in segmentation and quantification, especially with complex backgrounds.
• Accurate assessment of disease severity is crucial for effective management.

Purpose of the Study:

• To develop an accurate and efficient method for segmenting apple leaf and diseased regions.
• To assess apple leaf disease severity by overcoming challenges posed by complex backgrounds and overlapping leaves.
• To improve upon existing automated disease detection techniques.

Main Methods:

• Utilized an improved HRNet_w32 backbone with a Normalization Attention Mechanism (NAM).
• Implemented a combined Dice Loss and Focal Loss for enhanced semantic segmentation of leaf and diseased areas.
• Applied the DRL-watershed algorithm for optimizing segmentation of overlapping leaf regions.

Main Results:

• The enhanced HRNet model achieved a mean intersection over union (mIoU) of 88.91% and mean pixel accuracy (mPA) of 94.13%.
• Significant improvements of 8.77% (mIoU) and 7.25% (mPA) were observed compared to the original HRNet.
• Disease severity assessment accuracy reached an impressive 97.65%.

Conclusions:

• The proposed method accurately segments apple leaves and diseased areas, even in complex environments.
• This approach effectively handles leaf overlap, providing a reliable basis for disease severity assessment.
• The findings offer a robust scientific foundation for developing targeted apple disease management strategies.