Scholarlink

Background <p>This study was based on the identification of drought-resistant wheat varieties from trials to assess drought levels accurately in wheat and respond promptly to the impact of drought stress on grain yield. Hyperspectral remote sensing data and canopy temperature parameters for different wheat varieties were obtained, and classification models for drought resistance were constructed using various machine learning algorithms.</p> Results <p>As the growth period progresses, the spectral reflectance in the near-infrared band first increases but then decreases in the following pattern: flowering &gt; heading &gt; booting &gt; jointing &gt; filling. The effective temperature range of the canopy temperature histogram for different drought-resistant varieties also gradually increased with growth stage, and during each growth stage, the temperature variation in strongly drought-resistant varieties was the smallest compared with that in extremely weak drought-resistant varieties. Vegetation indices can represent the differences in drought resistance among wheat varieties; under drought stress, as drought resistance decreases, the canopy temperature parameters increase. There are certain correlations between the vegetation index (VI), canopy temperature parameter (TP), and the drought index for wheat yield. Furthermore, among the classification models based on the VI, TP, and VI + TP, the random forest (RF) model has the highest accuracy rate. Among them, the accuracy rates of the random forest model for VI + TP are 89.47% for overall accuracy (OA) and 0.85 for Kappa, which are higher than those of VI (OA = 69.57%, Kappa = 0.57) and TP (OA = 76.19%, Kappa = 0.66). In the optimal classification model VI + TP-RF, TP contributes the most to the RF classification algorithm based on multisource data fusion.</p> Conclusions <p>This study confirms the feasibility of using multimodal data fusion for classifying drought-resistant wheat varieties and provides a new reference method for further clear evaluation of wheat drought grade.</p> Graphical abstract <p></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

Drought resistance classification of wheat varieties by multisource data fusion and machine learning via hyperspectral and thermal infrared imaging

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月18日

查看全文
  • Xiaomei Zhang,
  • Hanwen Guan,
  • Jiliang Zhao,
  • Liunan Suo,
  • Dongyu Li,
  • Zhiyao Ma,
  • Jianzhao Duan,
  • Li He,
  • Wandai Liu,
  • Wei Feng

摘要

Background

This study was based on the identification of drought-resistant wheat varieties from trials to assess drought levels accurately in wheat and respond promptly to the impact of drought stress on grain yield. Hyperspectral remote sensing data and canopy temperature parameters for different wheat varieties were obtained, and classification models for drought resistance were constructed using various machine learning algorithms.

Results

As the growth period progresses, the spectral reflectance in the near-infrared band first increases but then decreases in the following pattern: flowering > heading > booting > jointing > filling. The effective temperature range of the canopy temperature histogram for different drought-resistant varieties also gradually increased with growth stage, and during each growth stage, the temperature variation in strongly drought-resistant varieties was the smallest compared with that in extremely weak drought-resistant varieties. Vegetation indices can represent the differences in drought resistance among wheat varieties; under drought stress, as drought resistance decreases, the canopy temperature parameters increase. There are certain correlations between the vegetation index (VI), canopy temperature parameter (TP), and the drought index for wheat yield. Furthermore, among the classification models based on the VI, TP, and VI + TP, the random forest (RF) model has the highest accuracy rate. Among them, the accuracy rates of the random forest model for VI + TP are 89.47% for overall accuracy (OA) and 0.85 for Kappa, which are higher than those of VI (OA = 69.57%, Kappa = 0.57) and TP (OA = 76.19%, Kappa = 0.66). In the optimal classification model VI + TP-RF, TP contributes the most to the RF classification algorithm based on multisource data fusion.

Conclusions

This study confirms the feasibility of using multimodal data fusion for classifying drought-resistant wheat varieties and provides a new reference method for further clear evaluation of wheat drought grade.

Graphical abstract