A multi-scale detection model for tomato leaf diseases with small target detection head

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato diseases significantly reduce crop yield and quality.
• Detecting diseases on small tomato leaves presents a major challenge due to scale variations.
• Early disease detection is vital for timely intervention and effective crop management.

Purpose of the Study:

• To develop an advanced tomato disease detection method addressing multi-scale and small object challenges.
• To improve the accuracy and efficiency of identifying diseased tomato leaves.

Main Methods:

• Proposed TomatoLeafDet method integrating multi-scale feature processing and small object detection.
• Designed Cross Stage Partial -Serial Multi-kernel Feature Aggregation (CSP-SMKFA) for multi-scale feature extraction.
• Introduced Symmetrical Re-calibration Aggregation (SRCA) for enhanced feature fusion.
• Utilized a Re-Calibration Feature Pyramid Network with a small object detection head.

Main Results:

• TomatoLeafDet demonstrated superior performance compared to YOLOv9 and YOLOv10.
• Achieved significant improvements in mean Average Precision (mAP50) on the CCMT tomato dataset.
• Outperformed baseline models by 4.4%, YOLOv9s by 1.9%, and YOLOv10n by 2.3%.

Conclusions:

• The proposed TomatoLeafDet method effectively addresses the challenges of detecting diseased tomato leaves at various scales.
• This novel approach offers a promising solution for early and accurate disease identification in tomato cultivation.
• The method's enhanced feature processing capabilities contribute to improved detection efficacy and agricultural outcomes.