Scholarlink

<p>Super-resolution techniques are essential for enhancing spatial detail in images, especially hyperspectral data, which possess rich spectral information but commonly suffer from low spatial resolution. Despite advancements in imaging hardware, acquiring high-resolution hyperspectral images remains challenging due to the vast data volume and acquisition limitations. Graph Signal Processing (GSP), and particularly the Spectral Graph Wavelet Transform (SGWT), offers an effective framework for processing signals on irregular domains by capturing the intrinsic relationships between spatial and spectral components through graph models. In this paper, we introduce a novel super-resolution method for hyperspectral images that leverages SGWT to extract wavelet coefficients from the data. An embedding network converts the low-resolution input into discriminative feature maps, which are then used to predict corresponding wavelet coefficient images. These predicted coefficients enable the reconstruction of a high-resolution hyperspectral image via the inverse SGWT. By jointly exploiting spatial and spectral information embedded within the hyperspectral cube, the proposed approach enhances image quality effectively. Experimental results validate the effectiveness of the proposed method.</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

Hyperspectral image super-resolution based on spectral graph wavelet transform

  • 期刊论文

  • 发表日期：

    2026年1月31日

查看全文
  • Katia Omani,
  • Mourad Lahdir,
  • Nadia Zikiou,
  • David Helbert

摘要

Super-resolution techniques are essential for enhancing spatial detail in images, especially hyperspectral data, which possess rich spectral information but commonly suffer from low spatial resolution. Despite advancements in imaging hardware, acquiring high-resolution hyperspectral images remains challenging due to the vast data volume and acquisition limitations. Graph Signal Processing (GSP), and particularly the Spectral Graph Wavelet Transform (SGWT), offers an effective framework for processing signals on irregular domains by capturing the intrinsic relationships between spatial and spectral components through graph models. In this paper, we introduce a novel super-resolution method for hyperspectral images that leverages SGWT to extract wavelet coefficients from the data. An embedding network converts the low-resolution input into discriminative feature maps, which are then used to predict corresponding wavelet coefficient images. These predicted coefficients enable the reconstruction of a high-resolution hyperspectral image via the inverse SGWT. By jointly exploiting spatial and spectral information embedded within the hyperspectral cube, the proposed approach enhances image quality effectively. Experimental results validate the effectiveness of the proposed method.