Scholarlink

<p>Hyperspectral images (HSI) have been widely used in various fields because of their rich spectral information. In order to make up for its inherent low spatial resolution disadvantage, it is an effective way to choose the fusion of HSI and multispectral images (MSI) in the same scene. In recent years, various deep learning-based networks have been employed to achieve HSI and MSI fusion. However, most fusion networks only use a limited spatial range of image pixel information, and lack the capability to extract and utilize the global features of an image. In this paper, we design a new hybrid attention module to fully retain the structural information of the source images and guide more image pixels to participate in the fusion process. In addition, the residual dense fusion module is designed to gradually achieve the fusion of two images, which can effectively retain the spectral information of HSI and increase the rich spatial background structure of MSI. Furthermore, we introduce Swin Transformer to highlight the spatial structure of images. The experimental results demonstrate that the proposed network has advantages in terms of both subjective and objective results when compared with the state-of-the-art methods.</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 and Multispectral Image Fusion Based on Residual Dense Fusion Network

  • 期刊论文

  • 发表日期：

    2025年2月11日

查看全文
  • Yuyuan Luo,
  • Jiawei Deng,
  • Bin Yang

摘要

Hyperspectral images (HSI) have been widely used in various fields because of their rich spectral information. In order to make up for its inherent low spatial resolution disadvantage, it is an effective way to choose the fusion of HSI and multispectral images (MSI) in the same scene. In recent years, various deep learning-based networks have been employed to achieve HSI and MSI fusion. However, most fusion networks only use a limited spatial range of image pixel information, and lack the capability to extract and utilize the global features of an image. In this paper, we design a new hybrid attention module to fully retain the structural information of the source images and guide more image pixels to participate in the fusion process. In addition, the residual dense fusion module is designed to gradually achieve the fusion of two images, which can effectively retain the spectral information of HSI and increase the rich spatial background structure of MSI. Furthermore, we introduce Swin Transformer to highlight the spatial structure of images. The experimental results demonstrate that the proposed network has advantages in terms of both subjective and objective results when compared with the state-of-the-art methods.