Citrus Disease Detection Based on Dilated Reparam Feature Enhancement and Shared Parameter Head.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Accurate citrus disease identification is crucial for effective pesticide application in orchards.
• Existing models face challenges with accuracy and efficiency due to varied lesion patterns and complex environments.

Purpose of the Study:

• To develop an improved, lightweight YOLOv8-based model (YOLOv8n-DE) for enhanced citrus disease detection.
• To increase the accuracy and efficiency of disease identification in real-world orchard settings.

Main Methods:

• Introduced the DR module for feature enhancement and Detect_Shared architecture for parameter efficiency within the YOLOv8 framework.
• Evaluated the YOLOv8n-DE model on both public and custom orchard-collected datasets.

Main Results:

• YOLOv8n-DE achieved 97.6% classification accuracy, 91.8% recall, and 97.3% mAP.
• Demonstrated a 48.17% reduction in parameters, 59.26% decrease in computational load, and 41.94% smaller model size compared to YOLOv8.
• Achieved 90.4% mAP for challenging citrus diseases, with reduced classification/regression errors and false positives/negatives.

Conclusions:

• YOLOv8n-DE offers superior performance and lightweight advantages for citrus disease detection.
• The model supports the advancement of precision agriculture through efficient and accurate disease identification in orchards.