Predicting yellow mosaic disease severity in yardlong bean using visible imaging coupled with machine learning model.

Abstract

Accurate estimation of plant disease severity is pivotal for effective management and decision-making. Field experiments were conducted to understand the correlation and predict the yellow mosaic disease severity in yard-long beans using visible image indices. A total of 45 visible / Red Green Blue (RGB) indices were derived from the RGB images and correlated with disease severity, and also used as inputs for predicting disease severity using nine machine learning (ML) models. Out of 143 genotypes screened based on final disease severity 3, 18, 18, 17, 34 and 53 genotypes were grouped in immune, resistant, moderately resistant, moderately susceptible, susceptible and highly susceptible categories, respectively. Model performances was evaluated using R 2 , d-index, mean bias error, and normalized Root Mean Square Error (n-RMSE) metrics. Results revealed that 34 indices exhibited significant correlations (p 2 and d-index values exceeding 0.92 and 0.98, respectively, in calibration, and 0.88 and 0.96 in validation, underscoring their effectiveness in predicting YMD severity using RGB images only. Random Forest (RF), Cubist, XGBoost (XGB), K-Nearest Neighbors (KNN), and Gradient Boosting Machine (GBM) emerged as the five top-performing models for predicting YMD severity using visible indices in yard-long beans. These findings hold practical implications for timely disease management strategies, expediting breeding programs, and aiding policy planners and farmers in making well-informed decisions.

Plant phenotyping relevance

RGB画像から可視画像指標を抽出し、機械学習で植物病害の重症度を推定・検証する手法が研究の中心であり、植物状態の定量的フェノタイピングに該当する。

Code and data availability

The paper's phenotype dataset (RGB image-derived visible indices and disease severity scores for 100 yardlong bean genotypes) and ML analysis are not publicly deposited; the Data availability statement says datasets are available from the corresponding author on reasonable request. No public code, images, or trained模型的

No evidence-backed public reproduction asset is currently recorded.