A Multi-Modal Open Object Detection Model for Tomato Leaf Diseases with Strong Generalization Performance Using...

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Precision agriculture demands accurate disease detection for crop health and yield.
• Current methods struggle with new diseases, requiring costly manual retraining.
• Open-vocabulary object detection (OVD) offers a potential solution for adaptable disease identification.

Purpose of the Study:

• To develop an innovative multimodal model (PDC-VLD) for accurate detection of new tomato leaf diseases.
• To overcome limitations of fixed-category detection methods and reduce manual annotation costs.
• To leverage OVD technology for efficient and adaptable plant disease identification.

Main Methods:

• Developed a progressive visual transformer-convolutional pyramid module (PVT-C) with self-supervised learning (DINO) for feature extraction.
• Implemented a context feature guided module (CFG) to enhance model adaptability in data-scarce scenarios.
• Utilized an open-vocabulary object detection (OVD) approach within the VLDet framework.

Main Results:

• The PDC-VLD model demonstrated superior performance in identifying tomato leaf diseases, including new categories.
• Achieved 61.2% on the primary evaluation metric $$ and 56.4% on $$.
• Showcased strong performance with 87.7% on $$, 81.0% on $$, and 45.5% average recall.

Conclusions:

• The PDC-VLD model offers an efficient and accurate solution for detecting novel plant diseases.
• Significantly reduces the need for extensive manual annotation in agricultural disease surveillance.
• Provides critical technical support for agricultural workers in managing emerging crop diseases.