ORIGINAL RESEARCH article
Front. Plant Sci.
Sec. Plant Pathogen Interactions
Volume 16 - 2025 | doi: 10.3389/fpls.2025.1599877
This article is part of the Research TopicInnovative Field Diagnostics for Real-Time Plant Pathogen Detection and ManagementView all 3 articles
Vertical Stratification-enabled Early Monitoring of Cotton Verticillium Wilt Using in-situ Leaf Spectroscopy via Machine Learning Models
Provisionally accepted- 1Aerospace Information Research Institute, Chinese Academy of Sciences (CAS), Beijing, China
- 2University of Chinese Academy of Sciences, Beijing, Beijing, China
- 3The Key Laboratory of the Oasis Ecological Agriculture, College of Agriculture, Shihezi University, Shihezi, Xinjiang Uyghur Region, China
- 4Xinjiang Academy of Agricultural and Reclamation Sciences (XAARS), Shihezi, Xinjiang Uyghur Region, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Early monitoring of cotton Verticillium wilt (VW) is crucial for preventing significant yield losses and quality deterioration. Current hyperspectral approaches often overlook the bottomup disease progression and the impact of leaf stratification on VW detection. To address this, vertical spectral traits were examined to improve early diagnosis. A total of 551 in-situ leaf spectra were averaged from thousands of measurements, alongside corresponding RGB images from top, middle, and bottom leaf layers. Five severity levels (SL=0-4) were classified based on lesion coverage.Various vegetation indices and signal features were extracted for VW identification. Three feature selection methods, Relief-F, Lasso, and Random Forest (RF), were integrated with five machine learning models, including LightGBM, ANN, XGBoost, RF, and SVM. Results showed that spectral reflectance varied significantly by severity and layer, with the most pronounced variations in the bottom layer's visible spectrum. LightGBM with RF-selected features achieved the best performance and fastest training, with accuracies of 0.82, 0.81, and 0.91 for the top, middle, and bottom leaf layers, respectively. Early-stage detection (SL=0-2) was most effective in the lowest layer, showing 38% and 34% higher precision (SL=1) than the upper two. Critical spectral features varied with vertical leaf layers and disease severity, with blue and red-edge bands identified as most important. For assessing five disease severity levels, the most informative features for the top, middle, and bottom layers were AntGitelson, Blue Index (B), and PRI570. For detecting early symptoms (SL=1), the blue band was particularly effective, followed by water-related bands. At the initial infection stage, the most significant indicators for top, middle, and bottom layers were Blue/red index (BRI), B, and WSCT, respectively. This study deepens understanding of vertical leaf spectral dynamics and enables rapid, non-destructive in vivo detection of cotton Verticillium wilt, enhancing the applicability of portable hyperspectral devices and informing leaf-layer-aware precision disease management strategies.
Keywords: Cotton verticillium wilt, vertical leaf layer, Hyperspectral reflectance, machine learning, Disease Severity
Received: 25 Mar 2025; Accepted: 30 May 2025.
Copyright: © 2025 Gao, Huang, Zhang, Zhang and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Changping Huang, Aerospace Information Research Institute, Chinese Academy of Sciences (CAS), Beijing, China
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.