Scholarlink

<p>Traditional compression techniques struggle to achieve high compression ratios while maintaining low reconstruction errors for hyperspectral images (HSIs). Deep learning-based methods, particularly those utilizing implicit neural representations (INRs), have shown promise in this domain. However, existing methods often overlook the spectral mixing and continuity inherent in HSI pixels, resulting in compromised reconstruction quality. To address this, we propose CUINR, a novel HSI compression method that integrates spectral unmixing with INRs. By establishing a continuous mapping from channel indices to spectral images, CUINR leverages the spectral continuity of HSIs. Additionally, CUINR introduces an unmixing reconstruction module and unmixing constraints to model the spectral mixing process, enhancing the physical significance and quality of reconstructed images. Experimental results indicate that CUINR outperforms baseline algorithms on multiple datasets, demonstrating the effectiveness of our approach in utilizing spectral characteristics to enhance reconstruction quality. The source code is available at: <a href="https://github.com/wgs1210/CUINR.git">https://github.com/wgs1210/CUINR.git</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

CUINR: combining unmixing with implicit neural representation for enhanced hyperspectral image compression

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Guisong Wang,
  • Yifan He,
  • Xiaofeng Du,
  • Xiaozhu Xie,
  • Wang Man,
  • Qin Nie

摘要

Traditional compression techniques struggle to achieve high compression ratios while maintaining low reconstruction errors for hyperspectral images (HSIs). Deep learning-based methods, particularly those utilizing implicit neural representations (INRs), have shown promise in this domain. However, existing methods often overlook the spectral mixing and continuity inherent in HSI pixels, resulting in compromised reconstruction quality. To address this, we propose CUINR, a novel HSI compression method that integrates spectral unmixing with INRs. By establishing a continuous mapping from channel indices to spectral images, CUINR leverages the spectral continuity of HSIs. Additionally, CUINR introduces an unmixing reconstruction module and unmixing constraints to model the spectral mixing process, enhancing the physical significance and quality of reconstructed images. Experimental results indicate that CUINR outperforms baseline algorithms on multiple datasets, demonstrating the effectiveness of our approach in utilizing spectral characteristics to enhance reconstruction quality. The source code is available at: https://github.com/wgs1210/CUINR.git.