Few-shot Metric Domain Adaptation: Practical Learning Strategies for an Automated Plant Disease Diagnosis

TL;DR

This paper introduces Few-shot Metric Domain Adaptation (FMDA), a practical method to improve plant disease diagnosis accuracy across different environments with limited target data, demonstrating significant performance gains in large-scale experiments.

Contribution

The paper proposes FMDA, a novel, efficient domain adaptation approach that minimizes feature space differences, enhancing diagnostic robustness with minimal target data in plant disease detection.

Findings

FMDA improved F1 scores by 11.1 to 29.3 points with only 10 target images.

FMDA outperformed fine-tuning methods with an average of 8.5 points higher.

Large-scale experiments involved over 223,000 leaf images across multiple crops.

Abstract

Numerous studies have explored image-based automated systems for plant disease diagnosis, demonstrating impressive diagnostic capabilities. However, recent large-scale analyses have revealed a critical limitation: that the diagnostic capability suffers significantly when validated on images captured in environments (domains) differing from those used during training. This shortfall stems from the inherently limited dataset size and the diverse manifestation of disease symptoms, combined with substantial variations in cultivation environments and imaging conditions, such as equipment and composition. These factors lead to insufficient variety in training data, ultimately constraining the system's robustness and generalization. To address these challenges, we propose Few-shot Metric Domain Adaptation (FMDA), a flexible and effective approach for enhancing diagnostic accuracy in practical systems, even when only limited target data is available. FMDA reduces domain discrepancies by introducing a constraint to the diagnostic model that minimizes the "distance" between feature spaces of source (training) data and target data with limited samples. FMDA is computationally efficient, requiring only basic feature distance calculations and backpropagation, and can be seamlessly integrated into any machine learning (ML) pipeline. In large-scale experiments, involving 223,015 leaf images across 20 fields and 3 crop species, FMDA achieved F1 score improvements of 11.1 to 29.3 points compared to cases without target data, using only 10 images per disease from the target domain. Moreover, FMDA consistently outperformed fine-tuning methods utilizing the same data, with an average improvement of 8.5 points.