Multi-Scale Feature Fusion Based RT-DETR for Tomato Leaf Disease Detection in Complex Backgrounds.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate detection of plant diseases is crucial for crop yield and food security.
• Existing methods for plant disease detection face challenges with complex backgrounds and imbalanced datasets.

Purpose of the Study:

• To develop an efficient and accurate model for detecting tomato leaf diseases.
• To enhance feature extraction and model learning performance for improved disease identification.

Main Methods:

• Proposed a multi-scale feature fusion network based on an improved RT-DETR model.
• Incorporated multi-scale extended residual modules and a multi-scale feature pyramid network for enhanced feature extraction.
• Introduced an adaptive focal loss (AFL) function to address overfitting and improve learning on imbalanced datasets.

Main Results:

• Achieved an AP@0.50 of 97.9% for tomato disease detection.
• Demonstrated high detection accuracy of 85.4% for other crop diseases.
• The adaptive focal loss (AFL) showed improved performance on imbalanced datasets.

Conclusions:

• The developed multi-scale feature fusion network with AFL offers a robust solution for plant disease detection.
• The model's high accuracy provides a valuable reference for practical agricultural applications.
• The approach effectively handles complex backgrounds and dataset imbalances, paving the way for improved crop monitoring.