Scholarlink

<p>Enhancing satellite imagery is critical for accurate weather analysis and meteorological research. This paper proposes an efficient GAN-based approach for cloud removal that integrates clustering and thresholding during preprocessing to ensure precise segmentation of clouds and shadows. A conditional image generation model replaces missing regions while preserving contextual information. The end-to-end pipeline enhances efficiency by streamlining preprocessing and generation tasks, while the discriminator utilizes multiscale feature comparison and Global Average Pooling for legitimate assessment. The superiority of the proposed approach is validated through comprehensive comparisons with ground truth data and existing methods. The model is evaluated on RICE2 and SEN12MS-CR datasets, achieving PSNR/SSIM scores of 41.24/0.983 and 39.15/0.962, respectively. It outperforms the next best existing model with an 8.2% PSNR and 0.4% SSIM gain on RICE2, and a 23% PSNR and 6.6% SSIM improvement on SEN12MS-CR. These results underscore the model’s effectiveness and contribution to advancing meteorological analysis.</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

Satellite cloud removal in meteorology: an efficient generative adversarial network based approach

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Sanjukta Mishra,
  • Jayanta Aich,
  • Samarjit Kar,
  • Parag Kumar Guhathakurta

摘要

Enhancing satellite imagery is critical for accurate weather analysis and meteorological research. This paper proposes an efficient GAN-based approach for cloud removal that integrates clustering and thresholding during preprocessing to ensure precise segmentation of clouds and shadows. A conditional image generation model replaces missing regions while preserving contextual information. The end-to-end pipeline enhances efficiency by streamlining preprocessing and generation tasks, while the discriminator utilizes multiscale feature comparison and Global Average Pooling for legitimate assessment. The superiority of the proposed approach is validated through comprehensive comparisons with ground truth data and existing methods. The model is evaluated on RICE2 and SEN12MS-CR datasets, achieving PSNR/SSIM scores of 41.24/0.983 and 39.15/0.962, respectively. It outperforms the next best existing model with an 8.2% PSNR and 0.4% SSIM gain on RICE2, and a 23% PSNR and 6.6% SSIM improvement on SEN12MS-CR. These results underscore the model’s effectiveness and contribution to advancing meteorological analysis.