Leaf disease image retrieval with object detection and deep metric learning

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate plant disease identification is crucial for effective crop management and food security.
• Deep learning models, particularly for plant leaf image classification, are popular but often require extensive data and retraining for new disease types.
• Existing methods lack flexibility in adapting to novel or emerging plant diseases.

Purpose of the Study:

• To develop a novel image retrieval system for automated detection, localization, and identification of plant leaf diseases.
• To enable flexible recognition of new disease types without the need for complete model retraining.
• To improve the accuracy and efficiency of plant disease surveillance systems.

Main Methods:

• Optimization of the YOLOv5 algorithm to enhance recognition of small objects, improving leaf object extraction.
• Integration of classification recognition with metric learning to jointly learn image categorization and similarity measurement.
• Construction of an efficient image retrieval system for rapid leaf disease type determination.

Main Results:

• The optimized YOLOv5 algorithm demonstrated improved accuracy in extracting leaf objects.
• The integrated approach effectively combined classification and metric learning for robust disease identification.
• The developed image retrieval system proved efficient and nimble in determining leaf disease types across multiple datasets.

Conclusions:

• The proposed system offers a flexible and accurate solution for automated plant disease identification in open settings.
• This work provides a foundation for advanced plant disease surveillance, supporting intelligent agriculture and crop research.
• The system's ability to identify new diseases without retraining represents a significant advancement in the field.