HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution

The core challenge of hyperspectral compressive imaging is to reconstruct the three-dimensional hyperspectral image from two-dimensional compressed measurements. While recent deep learning-based methods have demonsetrated outstanding performance, they often lack robust theoretical interpretability....

Full description

Saved in:
Bibliographic Details
Main Authors: Le Sun, Xihan Ma, Xinyu Wang, Qiao Chen, Zebin Wu
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/11022735/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850122086375227392
author Le Sun
Xihan Ma
Xinyu Wang
Qiao Chen
Zebin Wu
author_facet Le Sun
Xihan Ma
Xinyu Wang
Qiao Chen
Zebin Wu
author_sort Le Sun
collection DOAJ
description The core challenge of hyperspectral compressive imaging is to reconstruct the three-dimensional hyperspectral image from two-dimensional compressed measurements. While recent deep learning-based methods have demonsetrated outstanding performance, they often lack robust theoretical interpretability. Conversely, traditional iterative optimization algorithms are built upon sound mathematical derivations. To combine the advantages of both approaches, we propose a spectral transformer network, termed STTODNet, which integrates deep tensor decomposition and omni-dimensional dynamic convolution (ODConv). Specifically, we incorporate a deep Tucker decomposition module within the self-attention mechanism to effectively extract low-rank prior features inherent in the hyperspectral image. Moreover, we replace the conventional linear projection layer with ODConv to substantially improve feature extraction capabilities. A three-scale U-Net network structure is designed as the approximate operator for solving the prior within our deep unfolding network architecture. Extensive experimental results demonstrate that STTODNet achieves superior results in terms of reconstruction quality, interpretability, and computational efficiency when compared to state-of-the-art methods.
format Article
id doaj-art-d829eb9fc87d43209ada7f59f98f153c
institution OA Journals
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-d829eb9fc87d43209ada7f59f98f153c2025-08-20T02:34:55ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118149881500010.1109/JSTARS.2025.357617911022735HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic ConvolutionLe Sun0https://orcid.org/0000-0001-6465-8678Xihan Ma1Xinyu Wang2Qiao Chen3https://orcid.org/0000-0002-6458-8742Zebin Wu4https://orcid.org/0000-0002-7162-0202School of Computer Science, Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology, Nanjing University of Information Science and Technology, Nanjing, ChinaSchool of Computer Science, Nanjing University of Information Science and Technology, Nanjing, ChinaSchool of Computer Science, Nanjing University of Information Science and Technology, Nanjing, ChinaInstitute of Forest Resource Information Techniques, Chinese Academy of Forestry (CAF), Beijing, ChinaSchool of Computer Engineering, Nanjing University of Science and Technology, Nanjing, ChinaThe core challenge of hyperspectral compressive imaging is to reconstruct the three-dimensional hyperspectral image from two-dimensional compressed measurements. While recent deep learning-based methods have demonsetrated outstanding performance, they often lack robust theoretical interpretability. Conversely, traditional iterative optimization algorithms are built upon sound mathematical derivations. To combine the advantages of both approaches, we propose a spectral transformer network, termed STTODNet, which integrates deep tensor decomposition and omni-dimensional dynamic convolution (ODConv). Specifically, we incorporate a deep Tucker decomposition module within the self-attention mechanism to effectively extract low-rank prior features inherent in the hyperspectral image. Moreover, we replace the conventional linear projection layer with ODConv to substantially improve feature extraction capabilities. A three-scale U-Net network structure is designed as the approximate operator for solving the prior within our deep unfolding network architecture. Extensive experimental results demonstrate that STTODNet achieves superior results in terms of reconstruction quality, interpretability, and computational efficiency when compared to state-of-the-art methods.https://ieeexplore.ieee.org/document/11022735/Compressive imagingdeep Tucker decomposition (DTD)omni-dimensional dynamic convolution (ODConv)transformer
spellingShingle Le Sun
Xihan Ma
Xinyu Wang
Qiao Chen
Zebin Wu
HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Compressive imaging
deep Tucker decomposition (DTD)
omni-dimensional dynamic convolution (ODConv)
transformer
title HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
title_full HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
title_fullStr HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
title_full_unstemmed HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
title_short HSI Reconstruction: A Spectral Transformer With Tensor Decomposition and Dynamic Convolution
title_sort hsi reconstruction a spectral transformer with tensor decomposition and dynamic convolution
topic Compressive imaging
deep Tucker decomposition (DTD)
omni-dimensional dynamic convolution (ODConv)
transformer
url https://ieeexplore.ieee.org/document/11022735/
work_keys_str_mv AT lesun hsireconstructionaspectraltransformerwithtensordecompositionanddynamicconvolution
AT xihanma hsireconstructionaspectraltransformerwithtensordecompositionanddynamicconvolution
AT xinyuwang hsireconstructionaspectraltransformerwithtensordecompositionanddynamicconvolution
AT qiaochen hsireconstructionaspectraltransformerwithtensordecompositionanddynamicconvolution
AT zebinwu hsireconstructionaspectraltransformerwithtensordecompositionanddynamicconvolution