Scholarlink

<p>This paper presents a filling method to restore the SLC-off images of the Landsat 7 satellite without using reference images. Firstly, the preprocessing stage provides processing for gap localization, unsupervised classification, and local average greyscale calculation. Secondly, we conduct regression analysis to obtain the grey-level slope of the pixels on the two sides of the gap in vertical, 45 º and 135 º directions, and combine them with unsupervised classification results to decide the most likely filling direction for current missing pixel. We use a random strategy and cubic spline interpolation to fill edge pixels and non-edge pixels of the gaps. Finally, we utilize an adaptive filter to filter the gaps. The experimental results indicate that this method significantly outperforms other methods in terms of <i>PSNR</i>, <i>SSIM</i>, <i>MAPE</i> and <i>FID</i>. Our code will be available in GitHub depository at <a href="https://github.com/cheenyanjun/GapfillingLandsat-7norefer">https://github.com/cheenyanjun/GapfillingLandsat-7norefer</a>.</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

No-reference-image gap filling for Landsat-7 ETM + SLC-off images under unsupervised classification criteria

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Yanjun Chen,
  • Dazhi Zhou,
  • Jing Niu,
  • Xinke Li

摘要

This paper presents a filling method to restore the SLC-off images of the Landsat 7 satellite without using reference images. Firstly, the preprocessing stage provides processing for gap localization, unsupervised classification, and local average greyscale calculation. Secondly, we conduct regression analysis to obtain the grey-level slope of the pixels on the two sides of the gap in vertical, 45 º and 135 º directions, and combine them with unsupervised classification results to decide the most likely filling direction for current missing pixel. We use a random strategy and cubic spline interpolation to fill edge pixels and non-edge pixels of the gaps. Finally, we utilize an adaptive filter to filter the gaps. The experimental results indicate that this method significantly outperforms other methods in terms of PSNR, SSIM, MAPE and FID. Our code will be available in GitHub depository at https://github.com/cheenyanjun/GapfillingLandsat-7norefer.