Deep Metric Learning-Based Strawberry Disease Detection With Unknowns.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Deep object detection models have advanced plant disease detection.
• Detecting unknown plant diseases remains a significant challenge in practical applications.
• Existing methods struggle with identifying novel or uncatalogued plant pathogens.

Purpose of the Study:

• To propose an effective strawberry disease detection scheme capable of identifying both known and unknown diseases.
• To develop a practical solution for real-world field conditions.
• To improve the accuracy and applicability of automated plant disease diagnosis.

Main Methods:

• A two-stage detection pipeline combining object detection with deep metric learning (DML).
• Stage 1: Object detection for known disease classes.
• Stage 2: DML-based post-filtering using softmax and K-nearest neighbor (K-NN) classifiers for known and unknown diseases.

Main Results:

• The DML-based post-filter significantly improved the mean Average Precision (mAP) for known disease detection.
• The DML-based K-NN classifier achieved 97.8% accuracy, demonstrating high recall and precision for both known and unknown diseases.
• The proposed scheme achieved a 93.7% mAP for known strawberry diseases in real-field data, with reasonable performance for unknowns.

Conclusions:

• The proposed two-stage scheme effectively detects known and unknown plant diseases, offering practical field applicability.
• Deep metric learning enhances the performance of plant disease detection systems.
• The methodology shows potential for identifying diverse plant diseases and disorders across various plant species.