Tomato Leaf Disease Detection Method Based on Multi-Scale Feature Fusion.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato yield and quality are significantly impacted by early and accurate leaf disease detection.
• Manual diagnosis methods are labor-intensive and prone to subjective bias, necessitating automated solutions.
• Existing automated methods struggle with accuracy under complex environmental conditions.

Purpose of the Study:

• To develop an enhanced YOLOv11n-based detection framework for improved tomato leaf disease identification.
• To address the limitations of current disease detection methods in real-world agricultural settings.
• To enhance the robustness and practical applicability of automated disease monitoring systems.

Main Methods:

• An enhanced YOLOv11n model was developed, incorporating an EfficientMSF module for multi-scale feature extraction.
• A C2CU module was integrated to improve global contextual representation, and a CAFMFusion module for efficient feature fusion.
• The model was trained and evaluated on a self-constructed dataset comprising nine tomato leaf categories (eight diseases and healthy samples).

Main Results:

• The proposed model achieved an average Recall of 71.0%, mAP@0.5 of 76.5%, and mAP@0.5-0.95 of 60.5%.
• Performance improvements over the baseline YOLOv11n were 3.4% for Recall, 1.3% for mAP@0.5, and 2.0% for mAP@0.5-0.95.
• Notably, mAP@0.5 for the challenging Leaf Mold class improved by 3.4%.

Conclusions:

• The enhanced YOLOv11n framework demonstrates strong robustness and practical applicability in complex field conditions.
• The developed model offers an effective solution for intelligent tomato disease monitoring.
• This approach supports precision agricultural management through accurate and early disease detection.