Scholarlink

<p>The copyright protection of remote sensing images still face challenges. The existing protection methods are vulnerable to image processing attacks. Therefore, we propose a new protection framework, specifically, through remote sensing image feature extraction and identification, achieving copyright protection. To extract features with strong anti-interference ability, we propose a Convolutional block attention module-based Transformer Remote Sensing U-Net (CTRSU-Net) model and a SegNet + model. With these models, the remote sensing images can be segmented to provide robust semantic features. We also propose an ensemble learning to further fuse these features. The fused features are then used to create perceptual hash codes for similarity detection, achieving final remote sensing image identification. Experiments demonstrate that the proposed CTRSU-Net achieves an overall accuracy of 91.2%, increases by 3.6% compared to the baseline model U-Net on a remote sensing image dataset from China computer federation big data competition. After combining the CTRSU-Net and SegNet+, the accuracy has reached 92.1%, increases by 2.1% and 0.9% compared to single SegNet + and single CTRSU-Net.</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

Remote sensing image protection using CTRSU-Net, SegNet + and ensemble learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月21日

查看全文
  • De Li,
  • Chao Song,
  • Xun Jin

摘要

The copyright protection of remote sensing images still face challenges. The existing protection methods are vulnerable to image processing attacks. Therefore, we propose a new protection framework, specifically, through remote sensing image feature extraction and identification, achieving copyright protection. To extract features with strong anti-interference ability, we propose a Convolutional block attention module-based Transformer Remote Sensing U-Net (CTRSU-Net) model and a SegNet + model. With these models, the remote sensing images can be segmented to provide robust semantic features. We also propose an ensemble learning to further fuse these features. The fused features are then used to create perceptual hash codes for similarity detection, achieving final remote sensing image identification. Experiments demonstrate that the proposed CTRSU-Net achieves an overall accuracy of 91.2%, increases by 3.6% compared to the baseline model U-Net on a remote sensing image dataset from China computer federation big data competition. After combining the CTRSU-Net and SegNet+, the accuracy has reached 92.1%, increases by 2.1% and 0.9% compared to single SegNet + and single CTRSU-Net.