Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution

The superresolution (SR) technique is a leading solution for achieving high spatial–spectral resolution in hyperspectral (HS) images, which current sensors struggle to provide due to cost and physical constraints. This study presents a multistage optimization framework that leverages high...

Full description

Saved in:
Bibliographic Details
Main Authors: Zeinab Dehghan, Jingxiang Yang, Abdolraheem Khader, Jian Fang, Liang Xiao
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11066299/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849714400281231360
author Zeinab Dehghan
Jingxiang Yang
Abdolraheem Khader
Jian Fang
Liang Xiao
author_facet Zeinab Dehghan
Jingxiang Yang
Abdolraheem Khader
Jian Fang
Liang Xiao
author_sort Zeinab Dehghan
collection DOAJ
description The superresolution (SR) technique is a leading solution for achieving high spatial–spectral resolution in hyperspectral (HS) images, which current sensors struggle to provide due to cost and physical constraints. This study presents a multistage optimization framework that leverages high- and low-frequency components, along with a quadratic splitting method, to address the SR problem. Traditional model-based approaches often use shallow architectures with limited generalization. To overcome this, we integrated our model into a deep convolutional neural network enhanced by a Transformer module for regularization. Although the Transformer’s capabilities are noteworthy, it can improve in capturing local self-similarity and spectral correlations. Furthermore, these models frequently overlook the importance of multiscale and short-range information. Therefore, we introduce a multiscale architecture that allows Transformers to better capture short- and long-range dependencies. By implementing multiscale spatial-aware and multidepth channel-aware modules, we generate comprehensive deep spatial–spectral prior feature maps. The spatial branch focuses on using local–global prior features for HS image reconstruction, while the spectral branch emphasizes the most informative channels and their correlations. Experiments demonstrate that our method significantly outperforms state-of-the-art fusion-based SR techniques in terms of efficiency.
format Article
id doaj-art-ea28da419bf14c879588cadc27f6cecc
institution DOAJ
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-ea28da419bf14c879588cadc27f6cecc2025-08-20T03:13:43ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118179261794310.1109/JSTARS.2025.358575111066299Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image SuperresolutionZeinab Dehghan0https://orcid.org/0009-0001-5656-5321Jingxiang Yang1https://orcid.org/0000-0002-1234-0614Abdolraheem Khader2https://orcid.org/0000-0002-1164-3103Jian Fang3https://orcid.org/0000-0002-0781-0183Liang Xiao4https://orcid.org/0000-0003-0178-9384School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, ChinaSchool of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, ChinaSchool of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, ChinaSchool of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, ChinaSchool of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, ChinaThe superresolution (SR) technique is a leading solution for achieving high spatial–spectral resolution in hyperspectral (HS) images, which current sensors struggle to provide due to cost and physical constraints. This study presents a multistage optimization framework that leverages high- and low-frequency components, along with a quadratic splitting method, to address the SR problem. Traditional model-based approaches often use shallow architectures with limited generalization. To overcome this, we integrated our model into a deep convolutional neural network enhanced by a Transformer module for regularization. Although the Transformer’s capabilities are noteworthy, it can improve in capturing local self-similarity and spectral correlations. Furthermore, these models frequently overlook the importance of multiscale and short-range information. Therefore, we introduce a multiscale architecture that allows Transformers to better capture short- and long-range dependencies. By implementing multiscale spatial-aware and multidepth channel-aware modules, we generate comprehensive deep spatial–spectral prior feature maps. The spatial branch focuses on using local–global prior features for HS image reconstruction, while the spectral branch emphasizes the most informative channels and their correlations. Experiments demonstrate that our method significantly outperforms state-of-the-art fusion-based SR techniques in terms of efficiency.https://ieeexplore.ieee.org/document/11066299/Hyperspectral (HS)image fusionmultispectral (MS)neural networkpanchromatic (PAN)pansharpening
spellingShingle Zeinab Dehghan
Jingxiang Yang
Abdolraheem Khader
Jian Fang
Liang Xiao
Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Hyperspectral (HS)
image fusion
multispectral (MS)
neural network
panchromatic (PAN)
pansharpening
title Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
title_full Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
title_fullStr Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
title_full_unstemmed Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
title_short Learning Spectral–Spatial-Former Deep Prior for Hyperspectral Image Superresolution
title_sort learning spectral x2013 spatial former deep prior for hyperspectral image superresolution
topic Hyperspectral (HS)
image fusion
multispectral (MS)
neural network
panchromatic (PAN)
pansharpening
url https://ieeexplore.ieee.org/document/11066299/
work_keys_str_mv AT zeinabdehghan learningspectralx2013spatialformerdeeppriorforhyperspectralimagesuperresolution
AT jingxiangyang learningspectralx2013spatialformerdeeppriorforhyperspectralimagesuperresolution
AT abdolraheemkhader learningspectralx2013spatialformerdeeppriorforhyperspectralimagesuperresolution
AT jianfang learningspectralx2013spatialformerdeeppriorforhyperspectralimagesuperresolution
AT liangxiao learningspectralx2013spatialformerdeeppriorforhyperspectralimagesuperresolution