Enhanced YOLOv8 algorithm for leaf disease detection with lightweight GOCR-ELAN module and loss function: WSIoU.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate crop leaf disease detection is vital for enhancing agricultural yield and quality.
• Existing methods struggle with diverse target sizes, occlusions, and complex environments.
• The YOLOv8 architecture offers a foundation for advanced detection capabilities.

Purpose of the Study:

• To enhance the YOLOv8 model for more accurate and efficient leaf disease detection.
• To address challenges like varying target sizes, occlusions, and detection errors.
• To develop a lightweight yet high-performing model for agricultural applications.

Main Methods:

• Replaced C2f modules with the GOCR-ELAN lightweight module to improve feature extraction and reduce parameters.
• Substituted CBS convolution with the ADown downsampling module to enhance feature selection and preservation in occluded scenes.
• Implemented the WSIoU loss function optimization algorithm to improve convergence speed and localization accuracy.

Main Results:

• Achieved a 28.7% reduction in model parameters and a 43.2% decrease in GFLOPs.
• Improved Mean Average Precision (MAP50) from 86% to 87.7% and MAP50-95 from 67% to 68.9%.
• Developed a highly competitive model with a file size of 4.55 MB, smaller than YOLOv5.

Conclusions:

• The enhanced YOLOv8 model offers a lightweight and efficient solution for crop leaf disease detection.
• The modifications significantly improve detection performance, particularly in challenging conditions.
• This approach contributes to advancements in precision agriculture through improved automated disease identification.