结合重叠光谱块嵌入和深度卷积的高光谱解混网络
DOI:
CSTR:
作者:
作者单位:

沈阳理工大学

作者简介:

通讯作者:

中图分类号:

基金项目:

辽宁省属本科高校基本科研业务费专项资金项目(LJ212410144018),2024年度辽宁省教育厅基本科研计划项目(SYLUGXRC202258)


Hyperspectral unmixing network combined with overlapping spectral patch embedding and deep convolution
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    基于Transformer的高光谱解混方法将高光谱图像划分为大小固定且不重叠的小块,将分好的块输入给Transformer编码器后,再通过解码器进行解混。然而,不重叠的块划分会导致相邻像素之间相似信息被割裂,同时Transformer前馈层中会丢失一些局部信息,进而影响解混效果。本文采用改进的重叠光谱块划分以及在Transformer全连接层中加入深度卷积的高光谱解混方法,经过在多个数据集上进行实验,结果表明改进后的方法相比原始方法能够取得较为显著的解混效果。

    Abstract:

    Hyperspectral unmixing method based on Transformer divides the hyperspectral image into small patches of fixed size without overlapping, then put the divided patches input to the Transformer encoder and next unmixed by the decoder. However, the non-overlapping patch division will lead to the similar information between neighboring pixels being cut off. What’s more, some local information will be lost in the Transformer feedforward layer, which will affect the unmixing effect. In this paper, we adopt the improved hyperspectral unmixing method with overlapping spectral patch division and deep convolution in the fully connected layer of the Transformer. Hyperspectral image experiments show that the improved method is able to achieve a more significant unmixing effect compared with the original method.

    参考文献
    相似文献
    引证文献
引用本文
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-02-19
  • 最后修改日期:2025-04-04
  • 录用日期:2025-05-08
  • 在线发布日期:
  • 出版日期:
文章二维码