An efficient deep learning model for tomato disease detection.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato cultivation faces significant economic losses due to diseases impacting quality and yield.
• Accurate detection of tomato diseases is challenging due to intricate backgrounds and environmental interferences.
• Existing methods struggle with small lesion localization, false positives, and false negatives in real-world scenarios.

Purpose of the Study:

• To develop an advanced tomato disease detection system (TomatoDet) for intricate agricultural settings.
• To improve the accuracy and efficiency of identifying four major tomato diseases: late blight, gray leaf spot, brown rot, and leaf mold.
• To address challenges including background noise, small target detection, and false detection rates.

Main Methods:

• A novel feature extraction backbone network integrating Swin-DDETR's self-attention mechanism.
• Incorporation of the dynamic activation function Meta-ACON to enhance disease feature representation.
• An enhanced bidirectional weighted feature pyramid network (IBiFPN) for effective multi-scale feature fusion.

Main Results:

• Achieved a mean Average Precision (mAP) of 92.3% on a curated dataset, an 8.7% improvement over the baseline.
• Demonstrated a detection speed of 46.6 frames per second (FPS), suitable for agricultural applications.
• Successfully mitigated false positives and negatives caused by overlapping and occluded disease targets.

Conclusions:

• TomatoDet offers a robust solution for precise tomato disease detection in challenging environments.
• The proposed method significantly enhances detection accuracy and speed, meeting practical agricultural demands.
• This advancement contributes to improved tomato crop management and yield protection.