Scholarlink

<p>In this paper, ESRGAN-RRSIC-MVO, a robust remote sensing image classification framework based on deep learning technique is proposed. It combines physics-based degradation simulation, ESRGAN super-resolution, MobileNetV3 feature extraction, Multiverse Optimization (MVO) and dual SoftMax/SVM classification. Results from experiments on the UC Merced Land Use dataset yield 96.20% accuracy (Cohen’s Kappa: 0.9601), with only a 3.48% decrease when the atmosphere is in severe degradation, significantly outperforming baseline methods (13–16% degradation). The developed algorithm has been validated with the outcomes, showing the effectiveness of the combination of super-resolution and degradation-aware training for operational remote sensing applications.</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

ESRGAN-RRSIC-MVO: A Physics-Aware Deep Optimization Framework for Atmospheric-Robust Satellite Image Classification

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • N. Prajwal Hegde,
  • K. Prasanna Kumar,
  • Ancy Stephen,
  • K. D. V. Prasad,
  • M. Kameswara Rao,
  • P. Neelaveni

摘要

In this paper, ESRGAN-RRSIC-MVO, a robust remote sensing image classification framework based on deep learning technique is proposed. It combines physics-based degradation simulation, ESRGAN super-resolution, MobileNetV3 feature extraction, Multiverse Optimization (MVO) and dual SoftMax/SVM classification. Results from experiments on the UC Merced Land Use dataset yield 96.20% accuracy (Cohen’s Kappa: 0.9601), with only a 3.48% decrease when the atmosphere is in severe degradation, significantly outperforming baseline methods (13–16% degradation). The developed algorithm has been validated with the outcomes, showing the effectiveness of the combination of super-resolution and degradation-aware training for operational remote sensing applications.