Robust Plant Disease Diagnosis with Few Target-Domain Samples

TL;DR

Plant-disease diagnosis models suffer when test images come from unseen fields due to domain gaps. The authors introduce Target-Aware Metric Learning with Prioritized Sampling (TMPS), a metric-learning framework that exploits a small amount of labeled target-domain data (per class) by sampling target exemplars with a controlled probability $p$ to shape a robust embedding space. Across a large real-field leaf-disease dataset (223,073 images, 21 diseases, 3 crops), TMPS with as few as 10 target samples per disease significantly outperforms Baseline, conventional Metric Learning, All-Train, and Fine-Tuned baselines, with peak gains near $p=0.7$ and substantial macro-F1 improvements. The results highlight TMPS as a practical, adaptable approach for cross-field robustness in fine-grained plant-disease diagnostics and suggest potential applicability to other domains with large domain gaps and limited target data.

Abstract

Various deep learning-based systems have been proposed for accurate and convenient plant disease diagnosis, achieving impressive performance. However, recent studies show that these systems often fail to maintain diagnostic accuracy on images captured under different conditions from the training environment -- an essential criterion for model robustness. Many deep learning methods have shown high accuracy in plant disease diagnosis. However, they often struggle to generalize to images taken in conditions that differ from the training setting. This drop in performance stems from the subtle variability of disease symptoms and domain gaps -- differences in image context and environment. The root cause is the limited diversity of training data relative to task complexity, making even advanced models vulnerable in unseen domains. To tackle this challenge, we propose a simple yet highly adaptable learning framework called Target-Aware Metric Learning with Prioritized Sampling (TMPS), grounded in metric learning. TMPS operates under the assumption of access to a limited number of labeled samples from the target (deployment) domain and leverages these samples effectively to improve diagnostic robustness. We assess TMPS on a large-scale automated plant disease diagnostic task using a dataset comprising 223,073 leaf images sourced from 23 agricultural fields, spanning 21 diseases and healthy instances across three crop species. By incorporating just 10 target domain samples per disease into training, TMPS surpasses models trained using the same combined source and target samples, and those fine-tuned with these target samples after pre-training on source data. It achieves average macro F1 score improvements of 7.3 and 3.6 points, respectively, and a remarkable 18.7 and 17.1 point improvement over the baseline and conventional metric learning.

Figures (2)

• Figure 1: Example images illustrating the domain shift between the source and target datasets. The images show differences in leaf appearance, background complexity, and disease symptom expression.
• Figure 2: Change in performance with test field selection probability $p$. The solid line represents the average performance across five experiments, while error bars indicate the standard deviation.