Multi-kernel inception aggregation diffusion network for tomato disease detection.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato leaf diseases like septoria leaf spot, leaf curl virus, verticillium wilt, and early blight significantly reduce crop yield and quality.
• Accurate and rapid detection of these diseases is challenging due to scale variations in affected leaves.
• Effective disease management requires timely identification to mitigate economic losses in tomato cultivation.

Purpose of the Study:

• To develop a real-time detection model for identifying tomato leaf diseases across different scales.
• To enhance the accuracy and efficiency of disease diagnosis in tomato plants.
• To provide an effective solution for improving the quality of tomato cultivation through advanced detection methods.

Main Methods:

• Proposed a Multi-kernel Inception Aggregation Diffusion Network (MIADN) for processing multi-scale features.
• Introduced the Multi-kernel Inception Module (MKIM) to extract and fuse multi-scale object features using diverse convolutional kernels.
• Integrated the FasterNet network for efficient feature extraction, preserving feature diversity and enhancing complex feature identification.

Main Results:

• The proposed MIADN model achieved a mean average precision (mAP50) of 96.6%.
• The method demonstrated a 4.1% improvement over the baseline model and a 2.0% improvement over the YOLOv9s model.
• Experimental results validated the model's effectiveness in detecting tomato leaf diseases at various scales.

Conclusions:

• The developed MIADN model offers a robust and accurate solution for real-time detection of tomato leaf diseases.
• The integration of MKIM and FasterNet significantly enhances feature processing and extraction capabilities.
• This approach contributes to high-quality tomato cultivation by enabling prompt and precise disease management.