Scholarlink

Purpose <p>High-altitude unmanned aerial vehicle (UAV) remote sensing enables efficient large-scale crop monitoring, but reduced spatial resolution can obscure small disease symptoms and limit severity diagnosis. This study aimed to develop a task-oriented super-resolution (SR) framework for maize rust mapping and to determine whether recovered pathological details can improve downstream disease severity classification under operational high-altitude imaging conditions.</p> Methods <p>We proposed the Enhanced Residual Spatial-Channel Attention Generative Adversarial Network (ERSCA-WGAN), a pathology-aware SR model designed to reconstruct diagnostically relevant features, including rust pustules and subtle lesion boundaries. The model was trained using paired synthetic low- and high-resolution UAV images and validated on independent real-world UAV imagery acquired at 100-m and 25-m flight altitudes. Model effectiveness was evaluated mainly through its contribution to downstream disease severity classification.</p> Results <p>On the real 100-m UAV dataset, SR-enhanced imagery achieved an overall disease severity classification accuracy of 93.13%, improving the original low-resolution baseline by 3.27 percentage points. The performance gap between the enhanced 100-m imagery and the true 25-m high-resolution upper bound was reduced to 2.56%. ERSCA-WGAN also reduced misclassification between the visually ambiguous Slight and Moderate severity stages by 41.67%.</p> Conclusion <p>The proposed ERSCA-WGAN framework effectively bridges high-altitude UAV surveys and precise maize rust diagnosis. These results indicate that pathology-oriented SR can improve scalable crop disease monitoring while preserving operational flight efficiency.High-altitude unmanned aerial vehicle (UAV) remote sensing enables efficient large-scale crop monitoring, but the consequent reduction in spatial resolution often obscures critical disease symptoms, limiting the accuracy of severity diagnosis. This study aimed to develop a task-oriented super-resolution (SR) framework for maize rust mapping to determine whether recovering pathological details can significantly improve downstream disease severity classification under operational high-altitude imaging conditions.</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

Super-resolution-enhanced UAV imagery for high accuracy maize rust mapping at operational altitudes

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • Tao Liu,
  • Wang Liu,
  • Tiezhu Shi,
  • Zhongwen Hu,
  • Huan Zhang

摘要

Purpose

High-altitude unmanned aerial vehicle (UAV) remote sensing enables efficient large-scale crop monitoring, but reduced spatial resolution can obscure small disease symptoms and limit severity diagnosis. This study aimed to develop a task-oriented super-resolution (SR) framework for maize rust mapping and to determine whether recovered pathological details can improve downstream disease severity classification under operational high-altitude imaging conditions.

Methods

We proposed the Enhanced Residual Spatial-Channel Attention Generative Adversarial Network (ERSCA-WGAN), a pathology-aware SR model designed to reconstruct diagnostically relevant features, including rust pustules and subtle lesion boundaries. The model was trained using paired synthetic low- and high-resolution UAV images and validated on independent real-world UAV imagery acquired at 100-m and 25-m flight altitudes. Model effectiveness was evaluated mainly through its contribution to downstream disease severity classification.

Results

On the real 100-m UAV dataset, SR-enhanced imagery achieved an overall disease severity classification accuracy of 93.13%, improving the original low-resolution baseline by 3.27 percentage points. The performance gap between the enhanced 100-m imagery and the true 25-m high-resolution upper bound was reduced to 2.56%. ERSCA-WGAN also reduced misclassification between the visually ambiguous Slight and Moderate severity stages by 41.67%.

Conclusion

The proposed ERSCA-WGAN framework effectively bridges high-altitude UAV surveys and precise maize rust diagnosis. These results indicate that pathology-oriented SR can improve scalable crop disease monitoring while preserving operational flight efficiency.High-altitude unmanned aerial vehicle (UAV) remote sensing enables efficient large-scale crop monitoring, but the consequent reduction in spatial resolution often obscures critical disease symptoms, limiting the accuracy of severity diagnosis. This study aimed to develop a task-oriented super-resolution (SR) framework for maize rust mapping to determine whether recovering pathological details can significantly improve downstream disease severity classification under operational high-altitude imaging conditions.