Scholarlink

<p>The integration of machine learning with digital phenotyping offers a&#xa0;promising approach for rapid and objective assessment of plant stress responses. This study aimed to predict herbicide-induced phytotoxicity in common bean genotypes using spectral indices derived from RGB images processed by a&#xa0;Multilayer Perceptron (MLP) neural network. Spectral indices associated with chlorophyll content, leaf greenness, and photosynthetic activity were extracted from RGB images and used as predictor variables. Model interpretation was performed using SHAP (SHapley Additive exPlanations) to identify the spectral indices contributing most to the predictions. Model robustness was evaluated through Leave-One-Out cross-validation and 30 independent random seed initializations. The MLP achieved satisfactory predictive performance (R<sup>2</sup> = 0.797), accurately reproducing the principal experimental effects observed by conventional phenotyping, with no significant differences between observed and predicted phytotoxicity values across the evaluated genotypes. SHAP analysis identified the Modified Chlorophyll Absorption Ratio Index, Chlorophyll Vegetation Index, Normalized Difference Vegetation Index, and Color Index of Vegetation Extraction as the most influential predictors, indicating that the model primarily relied on spectral changes associated with chlorophyll degradation and reduced photosynthetic activity. These findings demonstrate that RGB-based digital phenotyping combined with interpretable machine learning provides an objective, reliable, low-cost, and scalable approach for herbicide phytotoxicity assessment, supporting high-throughput phenotyping and genotype selection in common bean breeding programs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of Herbicide Induced Phytotoxicity in Common Bean Genotypes Using Machine Learning and RGB Imaging

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Luan Tiago dos Santos Carbonari,
  • Carlos Zacarias Joaquin Junior,
  • Mauro Bitencourt de Souza,
  • Altamir Frederico Guidolin,
  • Jefferson Luís Meirelles Coimbra

摘要

The integration of machine learning with digital phenotyping offers a promising approach for rapid and objective assessment of plant stress responses. This study aimed to predict herbicide-induced phytotoxicity in common bean genotypes using spectral indices derived from RGB images processed by a Multilayer Perceptron (MLP) neural network. Spectral indices associated with chlorophyll content, leaf greenness, and photosynthetic activity were extracted from RGB images and used as predictor variables. Model interpretation was performed using SHAP (SHapley Additive exPlanations) to identify the spectral indices contributing most to the predictions. Model robustness was evaluated through Leave-One-Out cross-validation and 30 independent random seed initializations. The MLP achieved satisfactory predictive performance (R2 = 0.797), accurately reproducing the principal experimental effects observed by conventional phenotyping, with no significant differences between observed and predicted phytotoxicity values across the evaluated genotypes. SHAP analysis identified the Modified Chlorophyll Absorption Ratio Index, Chlorophyll Vegetation Index, Normalized Difference Vegetation Index, and Color Index of Vegetation Extraction as the most influential predictors, indicating that the model primarily relied on spectral changes associated with chlorophyll degradation and reduced photosynthetic activity. These findings demonstrate that RGB-based digital phenotyping combined with interpretable machine learning provides an objective, reliable, low-cost, and scalable approach for herbicide phytotoxicity assessment, supporting high-throughput phenotyping and genotype selection in common bean breeding programs.