DM-YOLO: improved YOLOv9 model for tomato leaf disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato leaf disease detection in natural settings is challenging due to variable lighting, overlapping symptoms, small lesions, and leaf occlusion.
• Accurate and early disease detection is crucial for effective crop management and yield preservation in agriculture.

Purpose of the Study:

• To propose an improved tomato leaf disease detection method, DM-YOLO, enhancing accuracy and robustness in complex natural environments.
• To address limitations of existing methods in detecting small, overlapping, or occluded disease lesions on tomato leaves.

Main Methods:

• Developed DM-YOLO, an enhanced detection model based on the YOLOv9 algorithm.
• Incorporated lightweight dynamic up-sampling (DySample) for improved small lesion feature extraction and background noise suppression.
• Utilized the MPDIoU loss function to refine the localization of overlapping lesion margins.

Main Results:

• DM-YOLO demonstrated superior performance compared to mainstream improved models, with precision increases ranging from 1.7% to 2.3%.
• On a tomato leaf disease dataset, DM-YOLO achieved 92.5% precision, 95.1% average precision (AP), and 86.4% mean average precision (mAP).
• Compared to the baseline YOLOv9, DM-YOLO showed improvements of 3% in precision, 1.7% in AP, and 1.4% in mAP.

Conclusions:

• The proposed DM-YOLO method offers significant improvements in tomato leaf disease detection accuracy and localization.
• This advanced detection capability provides strong support for the development of smart agriculture and efficient plant disease control strategies.
• DM-YOLO shows considerable potential for real-world applications in agricultural monitoring and management.