Scholarlink

Purpose <p>Tobacco mosaic virus (TMV) poses a significant threat to global tobacco production, often leading to substantial reductions in yield and quality. Early, asymptomatic detection is critical for timely intervention, yet current diagnostic methods are labor-intensive and unsuitable for large-scale monitoring. In this study, we present a novel approach for early asymptomatic TMV diagnosis in field conditions by integrating unmanned aerial vehicle (UAV)-based hyperspectral imaging with advanced machine learning techniques. </p> Methods <p>Field-grown tobacco plants were artificially inoculated with TMV, and hyperspectral data were collected at both asymptomatic and symptomatic stages. A series of spectral preprocessing techniques and machine learning models, including PLS-DA, SVC, and XGBoost, were evaluated. Feature band selection algorithms such as CARS and SVM-RFE were applied to identify key wavelengths, resulting in the retention of three optimal bands (527.33&#xa0;nm, 544.37&#xa0;nm, and 705.37&#xa0;nm). A TMV‑specific vegetation index (TMVI) was constructed using logistic regression.</p> Results <p>Models based on the three selected wavelengths achieved external‑validation accuracies of 90.75% for asymptomatic plants and 91.32% for symptomatic plants. The Tobacco Mosaic Virus Index (TMVI) achieved diagnostic accuracies of 82.52% and 82.92%, providing a simplified alternative to machine‑learning models. Visualization using the three wavelengths enabled clear differentiation of infected plants even before visible symptoms appeared.</p> Conclusion <p>This study represents the first demonstration of drone-based hyperspectral remote sensing for early TMV diagnosis in field conditions and provides a practical solution for large-scale, rapid screening. The TMVI offers a simplified, cost-effective diagnostic alternative, facilitating early intervention and effective disease management in tobacco cultivation.</p> Highlights <p>For the first time, hyperspectral imaging technology has been employed for the early asymptomatic detection of TMV in the field, achieving a diagnostic accuracy exceeding 90%. Through two rounds of spectral wavelength selection, three key bands for the early diagnosis of TMV were identified. A specialized vegetation index for the early diagnosis of TMV was developed, achieving an accuracy of 82%.</p> Impact <p>Through two rounds of band selection, three key spectral bands were identified for the field diagnosis of asymptomatic TMV infection. A specialized vegetation index was subsequently developed based on these bands, enabling early, large-area, real-time monitoring of TMV. This approach can be deployed using customized multispectral drone systems, offering an efficient tool for precise field management of TMV.</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

Early asymptomatic diagnosis of tobacco mosaic virus in fields utilizing hyperspectral imaging technology from unmanned aerial vehicle

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月6日

查看全文
  • Jianjun Huang,
  • Wei Kuang,
  • Qianjun Tang,
  • Can Wang,
  • Yansong Xiao,
  • Tianbo Liu,
  • Jiaying Li,
  • Kai Teng,
  • Hailin Cai,
  • Zhipeng Xiao,
  • Hong Zhou,
  • Xiangping Zhou,
  • Weiai Zeng,
  • Jianwu Li,
  • Zheming Yuan,
  • Shaolong Wu,
  • Yuan Chen

摘要

Purpose

Tobacco mosaic virus (TMV) poses a significant threat to global tobacco production, often leading to substantial reductions in yield and quality. Early, asymptomatic detection is critical for timely intervention, yet current diagnostic methods are labor-intensive and unsuitable for large-scale monitoring. In this study, we present a novel approach for early asymptomatic TMV diagnosis in field conditions by integrating unmanned aerial vehicle (UAV)-based hyperspectral imaging with advanced machine learning techniques.

Methods

Field-grown tobacco plants were artificially inoculated with TMV, and hyperspectral data were collected at both asymptomatic and symptomatic stages. A series of spectral preprocessing techniques and machine learning models, including PLS-DA, SVC, and XGBoost, were evaluated. Feature band selection algorithms such as CARS and SVM-RFE were applied to identify key wavelengths, resulting in the retention of three optimal bands (527.33 nm, 544.37 nm, and 705.37 nm). A TMV‑specific vegetation index (TMVI) was constructed using logistic regression.

Results

Models based on the three selected wavelengths achieved external‑validation accuracies of 90.75% for asymptomatic plants and 91.32% for symptomatic plants. The Tobacco Mosaic Virus Index (TMVI) achieved diagnostic accuracies of 82.52% and 82.92%, providing a simplified alternative to machine‑learning models. Visualization using the three wavelengths enabled clear differentiation of infected plants even before visible symptoms appeared.

Conclusion

This study represents the first demonstration of drone-based hyperspectral remote sensing for early TMV diagnosis in field conditions and provides a practical solution for large-scale, rapid screening. The TMVI offers a simplified, cost-effective diagnostic alternative, facilitating early intervention and effective disease management in tobacco cultivation.

Highlights

For the first time, hyperspectral imaging technology has been employed for the early asymptomatic detection of TMV in the field, achieving a diagnostic accuracy exceeding 90%. Through two rounds of spectral wavelength selection, three key bands for the early diagnosis of TMV were identified. A specialized vegetation index for the early diagnosis of TMV was developed, achieving an accuracy of 82%.

Impact

Through two rounds of band selection, three key spectral bands were identified for the field diagnosis of asymptomatic TMV infection. A specialized vegetation index was subsequently developed based on these bands, enabling early, large-area, real-time monitoring of TMV. This approach can be deployed using customized multispectral drone systems, offering an efficient tool for precise field management of TMV.