pith. sign in

arxiv: 2604.17652 · v1 · submitted 2026-04-19 · 💻 cs.CV

Self-Supervised Super-Resolution for Sentinel-5P Hyperspectral Images

Hyam Omar Ali , Antoine Crosnier , Romain Abraham , Baptiste Combelles , Fabrice J\'egou , Bruno Galerne This is my paper

Pith reviewed 2026-05-10 05:25 UTC · model grok-4.3

classification 💻 cs.CV
keywords self-supervised learningsuper-resolutionSentinel-5Phyperspectral imagingSUREequivariant imagingatmospheric monitoringremote sensing
0
0 comments X

The pith

Self-supervised super-resolution for Sentinel-5P matches supervised performance without any high-resolution ground truth.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a self-supervised framework for super-resolving Sentinel-5P hyperspectral images that trains directly on real low-resolution observations. It combines Stein's Unbiased Risk Estimator with an equivariant imaging constraint that incorporates the sensor's known degradation operator and noise statistics taken from SNR metadata, plus efficient depthwise separable convolution U-Nets. This removes the need for synthetic high-resolution pairs that supervised methods require. A reader would care because atmospheric monitoring data are inherently low-resolution, and methods that work without fabricated references can be applied immediately to actual satellite streams. Experiments show the self-supervised results stay comparable to supervised baselines while producing spatially sharper, physically consistent structures confirmed by independent EMIT validation.

Core claim

The central claim is that integrating Stein's Unbiased Risk Estimator with an equivariant imaging constraint, using the S5P degradation operator and SNR-derived noise statistics, produces a self-supervised hyperspectral super-resolution method whose performance on real and synthetic data is comparable to fully supervised baselines while guaranteeing that reconstructed fine-scale features remain physically meaningful, as verified by cross-validation against EMIT observations.

What carries the argument

The joint use of Stein's Unbiased Risk Estimator for risk estimation without targets and an equivariant imaging constraint that enforces consistency under the known S5P degradation operator, realized inside a depthwise separable U-Net that preserves spectral fidelity.

If this is right

  • Real S5P observations can be super-resolved directly without generating synthetic high-resolution pairs.
  • Super-resolved products maintain physical consistency suitable for atmospheric trace-gas analysis.
  • The same framework supports evaluation on data where no high-resolution reference exists at all.
  • Depthwise separable convolutions deliver the required efficiency while keeping spectral accuracy.
  • Qualitative spatial detail exceeds bicubic interpolation across multiple spectral bands.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could transfer to other atmospheric or Earth-observation sensors that lack paired high-resolution references.
  • Physical consistency validated against EMIT opens the possibility of fusing super-resolved S5P data with multi-sensor climate records.
  • Adding further physics-based priors such as radiative-transfer constraints inside the equivariant term might improve results without any supervision.
  • Wider adoption would lower the cost of preparing large labeled remote-sensing datasets.

Load-bearing premise

The degradation operator and noise statistics extracted from S5P SNR metadata accurately describe the real sensor behavior, and the equivariant imaging constraint is appropriate for atmospheric hyperspectral scenes.

What would settle it

Quantitative and structural mismatch between the method's super-resolved S5P output and co-located EMIT measurements on the same scenes, measured beyond the improvement already given by simple bicubic interpolation.

Figures

Figures reproduced from arXiv: 2604.17652 by Antoine Crosnier, Baptiste Combelles, Bruno Galerne, Fabrice J\'egou, Hyam Omar Ali, Romain Abraham.

Figure 1
Figure 1. Figure 1: Overview of the proposed Unet-S5P architecture and [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Relationship between SNR and bicubic PSNR across [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: GT-SHR qualitative evaluation through Unet-S5P-1M. Each spectral band is trained independently. The first and second [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative cross-sensor comparison between EMIT [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
read the original abstract

Sentinel-5P (S5P) plays a critical role in atmospheric monitoring; however, its spatial resolution limits fine-scale analysis. Existing super-resolution (SR) approaches rely on supervised learning with synthetic low-resolution (LR) data, since true high-resolution (HR) data do not exist, limiting their applicability to real observations. We propose a self-supervised hyperspectral SR framework for S5P that enables training without HR ground truth. The method combines Stein's Unbiased Risk Estimator (SURE) with an equivariant imaging constraint, incorporating the S5P degradation operator and noise statistics derived from signal-to-noise ratio (SNR) metadata. We also introduce depthwise separable convolution U-Net architectures designed for efficiency and spectral fidelity. The framework is evaluated in two settings: (i) LR-HR, where synthetic LR data are used for direct comparison with supervised learning, and (ii) GT-SHR, where super-resolved images surpass the native spatial resolution without HR reference. Results across multiple bands show that self-supervised models achieve performance comparable to supervised methods while maintaining strong consistency. Qualitative analysis shows improved spatial detail over bicubic interpolation, and validation with EMIT data confirms that reconstructed structures are physically meaningful. Code is available at https://github.com/hyamomar/Sentinel-5P-Super-Resolution/tree/main/self_supervised

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a self-supervised super-resolution framework for Sentinel-5P hyperspectral images that combines Stein's Unbiased Risk Estimator (SURE) with an equivariant imaging constraint. It incorporates the S5P degradation operator and noise statistics derived from SNR metadata, introduces depthwise separable convolution U-Net architectures, and evaluates in two settings: synthetic LR-HR for direct comparison to supervised methods and GT-SHR for real-data super-resolution beyond native resolution. The central claim is that self-supervised models achieve performance comparable to supervised baselines while producing physically meaningful reconstructions, as confirmed by EMIT validation. Code is made available.

Significance. If the noise-model assumptions hold, this enables training on real S5P observations without unavailable HR ground truth, which would be a meaningful advance for atmospheric monitoring applications. The open code repository is a clear strength for reproducibility. The dual synthetic/real evaluation protocol and use of external EMIT data for physical validation add value beyond purely synthetic benchmarks.

major comments (2)
  1. [Methods (SURE loss and noise model)] The central claim of self-supervised parity with supervised performance rests on SURE providing an unbiased risk estimate. This requires the per-pixel SNR-derived noise covariance and forward operator to match the true S5P imaging statistics, including any unmodeled spatial/spectral correlations. No empirical validation of this match (e.g., noise histogram or covariance comparison on real granules) is shown, which is load-bearing for the unbiasedness assumption and thus for the reported comparability.
  2. [Results (quantitative comparisons)] Quantitative tables comparing self-supervised and supervised models report point estimates without error bars, standard deviations across runs, or statistical significance tests. This makes it impossible to determine whether observed differences are within noise, weakening support for the 'comparable performance' statement in the abstract and results.
minor comments (2)
  1. [Abstract] The acronym 'GT-SHR' is used in the abstract without expansion; define it at first use.
  2. [Methods (equivariant imaging constraint)] Clarify whether the equivariant imaging group (rotations/flips) was chosen based on domain knowledge for atmospheric scenes or tested for sensitivity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough and constructive review. The comments identify important areas for strengthening the manuscript, particularly around validation of modeling assumptions and statistical rigor in reporting. We address each major comment below and outline specific revisions.

read point-by-point responses

  1. Referee: [Methods (SURE loss and noise model)] The central claim of self-supervised parity with supervised performance rests on SURE providing an unbiased risk estimate. This requires the per-pixel SNR-derived noise covariance and forward operator to match the true S5P imaging statistics, including any unmodeled spatial/spectral correlations. No empirical validation of this match (e.g., noise histogram or covariance comparison on real granules) is shown, which is load-bearing for the unbiasedness assumption and thus for the reported comparability.

    Authors: We appreciate the referee's emphasis on the foundational assumptions of the SURE-based loss. The noise covariance is constructed from the per-pixel SNR metadata supplied in the official S5P Level-1B products, and the forward operator follows the documented sensor degradation model. We acknowledge that the original submission did not include direct empirical checks (such as residual histograms or covariance matrices computed on real granules) to verify the match with observed noise statistics. In the revised manuscript we will add a dedicated validation subsection that (i) extracts noise residuals from multiple real S5P granules, (ii) compares their empirical distributions and correlation structure against the assumed diagonal Gaussian model, and (iii) discusses any residual discrepancies and their potential impact on SURE unbiasedness. This addition will directly address the load-bearing concern. revision: yes

  2. Referee: [Results (quantitative comparisons)] Quantitative tables comparing self-supervised and supervised models report point estimates without error bars, standard deviations across runs, or statistical significance tests. This makes it impossible to determine whether observed differences are within noise, weakening support for the 'comparable performance' statement in the abstract and results.

    Authors: We agree that point estimates alone limit the interpretability of the quantitative comparisons. The original experiments were performed with single training runs per configuration. In the revision we will repeat all reported experiments with at least three independent random seeds, report mean performance together with standard deviations in the tables, and add statistical significance tests (paired Wilcoxon signed-rank tests) between self-supervised and supervised results. These changes will allow readers to assess whether differences fall within run-to-run variability and will strengthen the claim of comparable performance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation uses external metadata and standard SURE

full rationale

The framework applies Stein's Unbiased Risk Estimator (SURE) together with an equivariant imaging constraint, where the degradation operator and noise statistics are taken directly from external per-pixel SNR metadata rather than fitted to the target super-resolution output. Evaluation on synthetic LR-HR pairs and GT-SHR without ground truth follows standard protocols and does not reduce claimed performance parity to any self-referential quantity. No self-citations, ansatz smuggling, or fitted-input-as-prediction patterns appear in the derivation chain.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the unbiasedness of SURE under the modeled noise and the validity of the equivariant constraint given the known degradation operator; no new physical entities are postulated.

free parameters (1)
  • U-Net hyperparameters and loss balancing weights
    Standard deep-learning training choices that are not derived from first principles and must be selected or tuned.
axioms (2)
  • standard math Stein's Unbiased Risk Estimator provides an unbiased estimate of the true risk when the noise model matches the data
    Invoked to enable training without HR targets.
  • domain assumption The equivariant imaging constraint holds for Sentinel-5P hyperspectral observations under the stated degradation operator
    Core modeling choice that allows self-supervision.

pith-pipeline@v0.9.0 · 5558 in / 1458 out tokens · 63238 ms · 2026-05-10T05:25:34.685764+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references · 72 canonical work pages

  1. [1]

    Sentinel-5P Mission Overview, 2025

    Copernicus Program. Sentinel-5P Mission Overview, 2025. Accessed: 2025-12-23

    work page 2025

  2. [2]

    Sentinel-5P Applications Overview, 2025

    Copernicus Programm. Sentinel-5P Applications Overview, 2025. Ac- cessed: 2025-12-23

    work page 2025

  3. [3]

    Sentinel-5P Products Overview, 2025

    Copernicus Program. Sentinel-5P Products Overview, 2025. Accessed: 2025-12-23

    work page 2025

  4. [4]

    European Space Agency (ESA), 2018

    European Space Agency (ESA).Input/Output Data Specification for the TROPOMI L01b Data Processor. European Space Agency (ESA), 2018. Accessed: 2025-12-21. SUBMITTED 12

    work page 2018

  5. [5]

    Springer, 2011

    Gustavo Camps-Valls, Devis Tuia, Luis G ´omez-Chova, Sandra Jim´enez, and Jes ´us Malo.Remote sensing image processing. Springer, 2011

    work page 2011

  6. [6]

    Model-based super-resolution for sentinel-5p data.IEEE Transac- tions on Geoscience and Remote Sensing, 62:1–16, 2024

    Alessia Carbone, Rocco Restaino, Gemine Vivone, and Jocelyn Chanus- sot. Model-based super-resolution for sentinel-5p data.IEEE Transac- tions on Geoscience and Remote Sensing, 62:1–16, 2024

    work page 2024

  7. [7]

    Depth separable architecture for sentinel-5p super-resolution

    Hyam Omar Ali, Romain Abraham, and Bruno Galerne. Depth separable architecture for sentinel-5p super-resolution. InIGARSS 2025-2025 IEEE International Geoscience and Remote Sensing Symposium, pages 7524–7529. IEEE, 2025

    work page 2025

  8. [8]

    Super-resolution image reconstruction: a technical overview.IEEE signal processing magazine, 20(3):21–36, 2003

    Sung Cheol Park, Min Kyu Park, and Moon Gi Kang. Super-resolution image reconstruction: a technical overview.IEEE signal processing magazine, 20(3):21–36, 2003

    work page 2003

  9. [9]

    A review of image super-resolution approaches based on deep learning and applications in remote sensing.Remote Sensing, 14(21):5423, 2022

    Xuan Wang, Jinglei Yi, Jian Guo, Yongchao Song, Jun Lyu, Jindong Xu, Weiqing Yan, Jindong Zhao, Qing Cai, and Haigen Min. A review of image super-resolution approaches based on deep learning and applications in remote sensing.Remote Sensing, 14(21):5423, 2022

    work page 2022

  10. [10]

    Advancing image super-resolution techniques in remote sensing: A comprehensive survey.ISPRS Journal of Photogrammetry and Remote Sensing, 231:68–100, 2026

    Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, and Wen Nie. Advancing image super-resolution techniques in remote sensing: A comprehensive survey.ISPRS Journal of Photogrammetry and Remote Sensing, 231:68–100, 2026

    work page 2026

  11. [11]

    Deep learning for image super-resolution: A survey.IEEE transactions on pattern analysis and machine intelligence, 43(10):3365–3387, 2020

    Zhihao Wang, Jian Chen, and Steven CH Hoi. Deep learning for image super-resolution: A survey.IEEE transactions on pattern analysis and machine intelligence, 43(10):3365–3387, 2020

    work page 2020

  12. [12]

    Image super-resolution using deep convolutional networks.IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional networks.IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015

    work page 2015

  13. [13]

    Super-resolution of sentinel-2 images: Learning a globally applicable deep neural network.ISPRS Journal of Photogrammetry and Remote Sensing, 146:305–319, 2018

    Charis Lanaras, Jos ´e Bioucas-Dias, Silvano Galliani, Emmanuel Balt- savias, and Konrad Schindler. Super-resolution of sentinel-2 images: Learning a globally applicable deep neural network.ISPRS Journal of Photogrammetry and Remote Sensing, 146:305–319, 2018

    work page 2018

  14. [14]

    A review of hyperspectral image super-resolution based on deep learning.Remote Sensing, 15(11):2853, 2023

    Chi Chen, Yongcheng Wang, Ning Zhang, Yuxi Zhang, and Zhikang Zhao. A review of hyperspectral image super-resolution based on deep learning.Remote Sensing, 15(11):2853, 2023

    work page 2023

  15. [15]

    Advancements in deep learning-based super-resolution for remote sens- ing: A comprehensive review and future directions.Super-Resolution for Remote Sensing, pages 51–91, 2024

    Saba Hosseini Tabesh, Masoud Babadi Ataabadi, and Dongmei Chen. Advancements in deep learning-based super-resolution for remote sens- ing: A comprehensive review and future directions.Super-Resolution for Remote Sensing, pages 51–91, 2024

    work page 2024

  16. [16]

    Spectral super-resolution meets deep learning: Achievements and challenges.Information Fusion, 97:101812, 2023

    Jiang He, Qiangqiang Yuan, Jie Li, Yi Xiao, Denghong Liu, Huanfeng Shen, and Liangpei Zhang. Spectral super-resolution meets deep learning: Achievements and challenges.Information Fusion, 97:101812, 2023

    work page 2023

  17. [17]

    Nour Aburaed, Mohammed Q Alkhatib, Stephen Marshall, Jaime Za- balza, and Hussain Al Ahmad. A review of spatial enhancement of hyperspectral remote sensing imaging techniques.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 16:2275–2300, 2023

    work page 2023

  18. [18]

    Hyperspectral image super-resolution meets deep learning: A survey and perspective

    Xinya Wang, Qian Hu, Yingsong Cheng, and Jiayi Ma. Hyperspectral image super-resolution meets deep learning: A survey and perspective. IEEE/CAA Journal of Automatica Sinica, 10(8):1668–1691, 2023

    work page 2023

  19. [19]

    Single hyperspectral image super-resolution with grouped deep recursive residual network

    Yong Li, Lei Zhang, Chen Dingl, Wei Wei, and Yanning Zhang. Single hyperspectral image super-resolution with grouped deep recursive residual network. In2018 IEEE Fourth International Conference on Multimedia Big Data (BigMM), pages 1–4. IEEE, 2018

    work page 2018

  20. [20]

    Hyperspectral image super- resolution using generative adversarial network and residual learning

    Qian Huang, Wei Li, Ting Hu, and Ran Tao. Hyperspectral image super- resolution using generative adversarial network and residual learning. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 3012–3016. IEEE, 2019

    work page 2019

  21. [21]

    Yuchao Yang, Yulei Wang, Xin Xu, and Enyu Zhao. Butterfly residual network: A hybrid approach with spectral transformers and depth- wise convolutions for hyperspectral image super-resolution.IEEE Transactions on Neural Networks and Learning Systems, 2025

    work page 2025

  22. [22]

    Spatial-spectral deep residual network for hyperspectral image super-resolution.SN Computer Science, 4(4):424, 2023

    WeiFa Zheng and ZiXin Xie. Spatial-spectral deep residual network for hyperspectral image super-resolution.SN Computer Science, 4(4):424, 2023

    work page 2023

  23. [23]

    Dlra-net: Deep local residual attention network with contex- tual refinement for spectral super-resolution.International Journal of Computer Vision, 133(4):1499–1531, 2025

    Ahmed R El-gabri, Hussein A Aly, Tarek S Ghoniemy, and Mohamed A Elshafey. Dlra-net: Deep local residual attention network with contex- tual refinement for spectral super-resolution.International Journal of Computer Vision, 133(4):1499–1531, 2025

    work page 2025

  24. [24]

    Asymmetric dual-direction quasi-recursive network for single hyperspectral image super-resolution

    Heng Wang, Cong Wang, and Yuan Yuan. Asymmetric dual-direction quasi-recursive network for single hyperspectral image super-resolution. IEEE Transactions on Circuits and Systems for Video Technology, 33(11):6331–6346, 2023

    work page 2023

  25. [25]

    Recursive deep feature learning for hyperspectral image super-resolution.Applied Sciences, 16(2):1060, 2026

    Jiming Liu, Chen Yi, and Hehuan Li. Recursive deep feature learning for hyperspectral image super-resolution.Applied Sciences, 16(2):1060, 2026

    work page 2026

  26. [26]

    Dual-stage approach toward hyperspectral image super-resolution.IEEE Transactions on Image Processing, 31:7252–7263, 2022

    Qiang Li, Yuan Yuan, Xiuping Jia, and Qi Wang. Dual-stage approach toward hyperspectral image super-resolution.IEEE Transactions on Image Processing, 31:7252–7263, 2022

    work page 2022

  27. [27]

    Deep recursive network for hyperspectral image super-resolution.IEEE Transactions on Computational Imaging, 6:1233–1244, 2020

    Wei Wei, Jiangtao Nie, Yong Li, Lei Zhang, and Yanning Zhang. Deep recursive network for hyperspectral image super-resolution.IEEE Transactions on Computational Imaging, 6:1233–1244, 2020

    work page 2020

  28. [28]

    Jin-Fan Hu, Ting-Zhu Huang, Liang-Jian Deng, Tai-Xiang Jiang, Gem- ine Vivone, and Jocelyn Chanussot. Hyperspectral image super- resolution via deep spatiospectral attention convolutional neural net- works.IEEE Transactions on Neural Networks and Learning Systems, 33(12):7251–7265, 2021

    work page 2021

  29. [29]

    Hyperspectral image super-resolution by band attention through adversarial learning.IEEE Transactions on Geoscience and Remote Sensing, 58(6):4304–4318, 2020

    Jiaojiao Li, Ruxing Cui, Bo Li, Rui Song, Yunsong Li, Yuchao Dai, and Qian Du. Hyperspectral image super-resolution by band attention through adversarial learning.IEEE Transactions on Geoscience and Remote Sensing, 58(6):4304–4318, 2020

    work page 2020

  30. [30]

    Dual self-attention swin transformer for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    Yaqian Long, Xun Wang, Meng Xu, Shuyu Zhang, Shuguo Jiang, and Sen Jia. Dual self-attention swin transformer for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    work page 2023

  31. [31]

    Attention- driven dual feature guidance for hyperspectral super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–16, 2023

    Minghua Zhao, Jiawei Ning, Jing Hu, and Tingting Li. Attention- driven dual feature guidance for hyperspectral super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–16, 2023

    work page 2023

  32. [32]

    Attention-enhanced generative adversarial network for hyperspectral imagery spatial super-resolution.Remote Sensing, 15(14):3644, 2023

    Baorui Wang, Yifan Zhang, Yan Feng, Bobo Xie, and Shaohui Mei. Attention-enhanced generative adversarial network for hyperspectral imagery spatial super-resolution.Remote Sensing, 15(14):3644, 2023

    work page 2023

  33. [33]

    Interactformer: Interactive transformer and cnn for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 60:1–15, 2022

    Yaoting Liu, Jianwen Hu, Xudong Kang, Jing Luo, and Shaosheng Fan. Interactformer: Interactive transformer and cnn for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 60:1–15, 2022

    work page 2022

  34. [34]

    Essaformer: Efficient transformer for hyperspectral image super-resolution

    Mingjin Zhang, Chi Zhang, Qiming Zhang, Jie Guo, Xinbo Gao, and Jing Zhang. Essaformer: Efficient transformer for hyperspectral image super-resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 23073–23084, 2023

    work page 2023

  35. [35]

    Fusformer: A transformer-based fusion network for hyperspectral image super-resolution.IEEE Geoscience and Remote Sensing Letters, 19:1–5, 2022

    Jin-Fan Hu, Ting-Zhu Huang, Liang-Jian Deng, Hong-Xia Dou, Danfeng Hong, and Gemine Vivone. Fusformer: A transformer-based fusion network for hyperspectral image super-resolution.IEEE Geoscience and Remote Sensing Letters, 19:1–5, 2022

    work page 2022

  36. [36]

    Msdformer: Multiscale deformable transformer for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–14, 2023

    Shi Chen, Lefei Zhang, and Liangpei Zhang. Msdformer: Multiscale deformable transformer for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 61:1–14, 2023

    work page 2023

  37. [37]

    A spectral and spatial transformer for hyperspectral remote sensing image super-resolution.International Journal of Digital Earth, 17(1):2313102, 2024

    Bingqian Wang, Jianhua Chen, Huajun Wang, Yipeng Tang, Jiongling Chen, and Ye Jiang. A spectral and spatial transformer for hyperspectral remote sensing image super-resolution.International Journal of Digital Earth, 17(1):2313102, 2024

    work page 2024

  38. [38]

    A multi-path neural network for hyperspectral image super-resolution

    Jing Zhang, Zekang Wan, Minhao Shao, and Yunsong Li. A multi-path neural network for hyperspectral image super-resolution. InInterna- tional Conference on Image and Graphics, pages 377–387. Springer, 2021

    work page 2021

  39. [39]

    Separable-spectral convolution and inception network for hyperspectral image super-resolution.International Journal of Machine Learning and Cybernetics, 10(10):2593–2607, 2019

    Ke Zheng, Lianru Gao, Qiong Ran, Ximin Cui, Bing Zhang, Wenzhi Liao, and Sen Jia. Separable-spectral convolution and inception network for hyperspectral image super-resolution.International Journal of Machine Learning and Cybernetics, 10(10):2593–2607, 2019

    work page 2019

  40. [40]

    A spectral diffusion prior for unsupervised hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 2024

    Jianjun Liu, Zebin Wu, and Liang Xiao. A spectral diffusion prior for unsupervised hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 2024

    work page 2024

  41. [41]

    Spectral-cascaded diffusion model for remote sensing image spectral super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    Bowen Chen, Liqin Liu, Chenyang Liu, Zhengxia Zou, and Zhenwei Shi. Spectral-cascaded diffusion model for remote sensing image spectral super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    work page 2024

  42. [42]

    Hsr-diff: Hyperspectral image super-resolution via condi- tional diffusion models

    Chanyue Wu, Dong Wang, Yunpeng Bai, Hanyu Mao, Ying Li, and Qiang Shen. Hsr-diff: Hyperspectral image super-resolution via condi- tional diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 7083–7093, 2023

    work page 2023

  43. [43]

    A comb concatenation diffusion model for hyperspectral image super-resolution.Engineering Applications of Artificial Intelligence, 160:111985, 2025

    Yinghao Xu, Hao Wang, Xin Sun, Qianlong Xie, Wenwen Zhang, Peng Ren, Fei Zhou, and Susanto Rahardja. A comb concatenation diffusion model for hyperspectral image super-resolution.Engineering Applications of Artificial Intelligence, 160:111985, 2025

    work page 2025

  44. [44]

    Ispdiff: Interpretable scale-propelled diffusion model for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    Wenqian Dong, Sen Liu, Song Xiao, Jiahui Qu, and Yunsong Li. Ispdiff: Interpretable scale-propelled diffusion model for hyperspectral image super-resolution.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    work page 2024

  45. [45]

    Efficient hyperspectral super-resolution of sentinel-5p data via dynamic multi- directional cascade fine-tuning.IEEE Geoscience and Remote Sensing Letters, 2024

    Alessia Carbone, Rocco Restaino, and Gemine Vivone. Efficient hyperspectral super-resolution of sentinel-5p data via dynamic multi- directional cascade fine-tuning.IEEE Geoscience and Remote Sensing Letters, 2024. SUBMITTED 13

    work page 2024

  46. [46]

    Xception: Deep learning with depthwise separable convolutions

    Franc ¸ois Chollet. Xception: Deep learning with depthwise separable convolutions. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1251–1258, 2017

    work page 2017

  47. [47]

    Deep image prior

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Deep image prior. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 9446–9454, 2018

    work page 2018

  48. [48]

    zero-shot

    Assaf Shocher, Nadav Cohen, and Michal Irani. “zero-shot” super- resolution using deep internal learning. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3118– 3126, 2018

    work page 2018

  49. [49]

    Blind super- resolution kernel estimation using an internal-gan.Advances in neural information processing systems, 32, 2019

    Sefi Bell-Kligler, Assaf Shocher, and Michal Irani. Blind super- resolution kernel estimation using an internal-gan.Advances in neural information processing systems, 32, 2019

    work page 2019

  50. [50]

    Blind super- resolution with iterative kernel correction

    Jinjin Gu, Hannan Lu, Wangmeng Zuo, and Chao Dong. Blind super- resolution with iterative kernel correction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1604–1613, 2019

    work page 2019

  51. [51]

    Scale- equivariant imaging: Self-supervised learning for image super-resolution and deblurring.IEEE Transactions on Computational Imaging, 2026

    J ´er´emy Scanvic, Mike Davies, Patrice Abry, and Juli ´an Tachella. Scale- equivariant imaging: Self-supervised learning for image super-resolution and deblurring.IEEE Transactions on Computational Imaging, 2026

    work page 2026

  52. [52]

    Real-time image super-resolution using recursive depthwise separable convolution net- work.IEEE Access, 7:99804–99816, 2019

    Kwok-Wai Hung, Zhikai Zhang, and Jianmin Jiang. Real-time image super-resolution using recursive depthwise separable convolution net- work.IEEE Access, 7:99804–99816, 2019

    work page 2019

  53. [53]

    Depth separable-cnn for improved spectral super-resolution.IEEE Access, 11:23063–23072, 2023

    Sadia Hussain and Brejesh Lall. Depth separable-cnn for improved spectral super-resolution.IEEE Access, 11:23063–23072, 2023

    work page 2023

  54. [54]

    Enhanced feature refinement network based on depthwise separable convolution for lightweight image super-resolution.Symmetry, 16(11):1406, 2024

    Weizhe Sun, Ran Ke, Zhen Liu, Haoran Lu, Dong Li, Fei Yang, and Lei Zhang. Enhanced feature refinement network based on depthwise separable convolution for lightweight image super-resolution.Symmetry, 16(11):1406, 2024

    work page 2024

  55. [55]

    Multi-scale xception based depthwise separable convolution for single image super- resolution.Plos one, 16(8):e0249278, 2021

    Wazir Muhammad, Supavadee Aramvith, and Takao Onoye. Multi-scale xception based depthwise separable convolution for single image super- resolution.Plos one, 16(8):e0249278, 2021

    work page 2021

  56. [56]

    Single image super- resolution: Depthwise separable convolution super-resolution generative adversarial network.Applied Sciences, 10(1):375, 2020

    Zetao Jiang, Yongsong Huang, and Lirui Hu. Single image super- resolution: Depthwise separable convolution super-resolution generative adversarial network.Applied Sciences, 10(1):375, 2020

    work page 2020

  57. [57]

    Enhancing change detection in hyperspectral images: A semi-supervised approach with u-net and attention mechanism

    I Bidari, S Chickerur, and S Kadam. Enhancing change detection in hyperspectral images: A semi-supervised approach with u-net and attention mechanism. InComputer Science Engineering, pages 71–80. CRC Press, 2024

    work page 2024

  58. [58]

    Hyperspectral change detection based on modification of unet neural networks.Journal of Applied Remote Sensing, 15(2):028505– 028505, 2021

    Marwa S Moustafa, Sayed A Mohamed, Sayed Ahmed, and Ayman H Nasr. Hyperspectral change detection based on modification of unet neural networks.Journal of Applied Remote Sensing, 15(2):028505– 028505, 2021

    work page 2021

  59. [59]

    Dca- unet: Enhancing small object segmentation in hyperspectral images with dual channel attention unet.Journal of the Franklin Institute, 362(4):107532, 2025

    Kunbo Han, Mingjin Chen, Chongzhi Gao, and Chunmei Qing. Dca- unet: Enhancing small object segmentation in hyperspectral images with dual channel attention unet.Journal of the Franklin Institute, 362(4):107532, 2025

    work page 2025

  60. [60]

    Semantic segmentation of hyperspectral remote sensing images based on pse-unet model.Sensors, 22(24):9678, 2022

    Jiaju Li, Hefeng Wang, Anbing Zhang, and Yuliang Liu. Semantic segmentation of hyperspectral remote sensing images based on pse-unet model.Sensors, 22(24):9678, 2022

    work page 2022

  61. [61]

    A high-level feature channel attention unet network for cholangiocar- cinoma segmentation from microscopy hyperspectral images.Machine Vision and Applications, 34(5):72, 2023

    Hongmin Gao, Mengran Yang, Xueying Cao, Qin Liu, and Peipei Xu. A high-level feature channel attention unet network for cholangiocar- cinoma segmentation from microscopy hyperspectral images.Machine Vision and Applications, 34(5):72, 2023

    work page 2023

  62. [62]

    Robust equivariant imaging: a fully unsupervised framework for learning to image from noisy and partial measurements

    Dongdong Chen, Juli ´an Tachella, and Mike E Davies. Robust equivariant imaging: a fully unsupervised framework for learning to image from noisy and partial measurements. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5647– 5656, 2022

    work page 2022

  63. [63]

    U-net: Convo- lutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convo- lutional networks for biomedical image segmentation. InInternational Conference on Medical image computing and computer-assisted inter- vention, pages 234–241. Springer, 2015

    work page 2015

  64. [64]

    Copernicus Data Space Ecosystem, 2025

    Copernicus Data Space. Copernicus Data Space Ecosystem, 2025. Accessed: 2025-12-23

    work page 2025

  65. [65]

    Orientation-cue invariant population responses to contrast-modulated and phase-reversed contour stimuli in macaque v1 and v2.PLoS One, 9(9):e106753, 2014

    Xu An, Hongliang Gong, Jiapeng Yin, Xiaochun Wang, Yanxia Pan, Xian Zhang, Yiliang Lu, Yupeng Yang, Zoltan Toth, Ingo Schiessl, et al. Orientation-cue invariant population responses to contrast-modulated and phase-reversed contour stimuli in macaque v1 and v2.PLoS One, 9(9):e106753, 2014

    work page 2014

  66. [66]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004

    work page 2004

  67. [67]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018

    work page 2018

  68. [68]

    Thompson, Robert O

    David R. Thompson, Robert O. Green, Christine Bradley, Philip G. Brodrick, Natalie Mahowald, Eyal Ben Dor, Matthew Bennett, Michael Bernas, Nimrod Carmon, K. Dana Chadwick, Roger N. Clark, Red Wil- low Coleman, Evan Cox, Ernesto Diaz, Michael L. Eastwood, Regina Eckert, Bethany L. Ehlmann, Paul Ginoux, Mar ´ıa Gonc ¸alves Ageitos, Kathleen Grant, Luis Gua...

    work page 2024

  69. [69]

    Orbital navigation utilities for earth remote sensing

    Space Science and University of Wisconsin-Madison Engineering Cen- ter. Orbital navigation utilities for earth remote sensing. Last accessed 30 January 2026 at http://www.ssec.wisc.edu/

    work page 2026

  70. [70]

    An empirical radiometric intercompari- son methodology based on global simultaneous nadir overpasses applied to landsat 8 and sentinel-2.Remote Sensing, 12(17), 2020

    Jorge Gil, Juan Fernando Rodrigo, Pablo Salvador, Diego G ´omez, Julia Sanz, and Jose Luis Casanova. An empirical radiometric intercompari- son methodology based on global simultaneous nadir overpasses applied to landsat 8 and sentinel-2.Remote Sensing, 12(17), 2020. APPENDIXA CHANNELCONFIGURATIONS OFUNET-S5P VARIANTS This section provides detailed channe...

    work page 2020

  71. [71]

    •BD2–BD6:497→63→8→1 •BD7–BD8:480→60→8→1 The decoder follows a shared expansion method for all the bands:1→8→64→512

    Unet-S5P-800k: the encoder consists of three levels of channel compression. •BD2–BD6:497→63→8→1 •BD7–BD8:480→60→8→1 The decoder follows a shared expansion method for all the bands:1→8→64→512

    work page

  72. [72]

    •BD2–BD6:497→180→65→24→9 •BD7–BD8:480→173→63→23→9 The decoder structure is shared across all bands:9→ 25→70→195→542

    Unet-S5P-1M: the encoder consists of four levels. •BD2–BD6:497→180→65→24→9 •BD7–BD8:480→173→63→23→9 The decoder structure is shared across all bands:9→ 25→70→195→542

    work page