Pith. sign in

REVIEW 3 major objections 7 minor 78 references

Super-Resolution of Sentinel-2 Images Using a Geometry-Guided Back-Projection Network with Self-Attention

T0 review · 3 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A geometry-guided back-projection network sharpens Sentinel-2's 20m bands to 10m, beating 13 fusion methods on every test metric.

desk verdict A well-ablated win on the simulated Wald task, but the real 20m-to-10m transfer is unproven and test-set tuning inflates the margins. read the letter →

arxiv 2508.04729 v1 pith:5WNYD7TK submitted 2025-08-05 eess.IV

Ivan Pereira-Sánchez , Daniel Torres , Francesc Alcover , Bartomeu Garau , Julia Navarro , Catalina Sbert , Joan Duran This is my paper · ORCID
classification eess.IV
keywords Sentinel-2super-resolutionimagefusionback-projectionmulti-headattentionnonlocalfilteringguidingWald'sprotocol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a geometry-guided, unfolded back-projection network can sharpen Sentinel-2's six 20m-resolution bands to 10m resolution more accurately than existing classical and deep-learning fusion methods. The proposed GINet+ builds a guiding image from the 10m bands through a cluster-based learned mapping, then feeds it into a back-projection loop whose fixed kernel is replaced by a residual network with a multi-head attention module. In experiments following Wald's protocol, GINet+ and its spectral-similarity sibling GINet take the first and second positions on every quality metric (ERGAS, PSNR, SSIM, SAM) on both validation and testing sets, and the ranking holds separately for urban, rural, and coastal test subsets. If the result transfers to real 20m-to-10m sharpening, it would give users the full 13-band Sentinel-2 spectral stack at 10m resolution without relying on a panchromatic band, improving vegetation, moisture, and water indexes that depend on the coarser bands.

What carries the argument

The load-bearing mechanism is an unfolded nonlocal back-projection iteration. The classical scheme $u^{k+1}=u^k+K*(DB u^k-f)\uparrow$ is rewritten with a nonlocal filter and then unfolded over $K=6$ stages, with each operator replaced by a learnable module: $DB$ becomes a convolution plus average pooling, bicubic upsampling becomes a transposed convolution, and the convergence kernel $K*_{NL}$ becomes a residual network (ResNL) with three residual blocks and a three-head multi-head attention (MHA) module. The MHA module computes similarity weights $\omega_{i,j}=\frac{1}{\Gamma_i}\exp(\theta(g_{Q_i})^t \phi(g_{Q_j}))$ over $3\times3$ patches within a $5\times5$ search window, using the upsampled error, the guiding image, and their concatenation as reference images; the nonlocal filter is applied uniformly to the upsampled error. The guiding image itself comes from one of two strategies: GINet uses the wavelength-nearest 10m band, while GINet+ uses a cluster-based module that assigns each 10m pixel to one of five clusters and applies a cluster-specific MLP to map pixels into the spectral space of the 20m bands. The whole pipeline is trained end-to-end with an $L1$ loss against the Wald-protocol reference, and all metrics are reported on that simulated task.

What would settle it

Compute the network's sharpened 20m bands on real (non-simulated) Sentinel-2 scenes with a known near-simultaneous 10m observation of the same area from another sensor, and measure the actual reflectance error at 10m scale; if the error is no better than bicubic upsampling, the transfer claim fails. Alternatively, retrain and test the network under Wald's protocol using a degradation kernel estimated from the sensor's measured modulation transfer function instead of simple average pooling; if the ranking over baselines changes or the margin collapses, the design's advantage is an artifact of the simulated protocol.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that replacing the linear back-projection kernel in an unfolded super-resolution scheme with a residual network that includes self-attention, and steering that network with a guiding image built from the finer 10m bands, yields the best reported sharpening of Sentinel-2 20m bands. The claim is supported by quantitative results: GINet+ achieves ERGAS 0.5546, PSNR 42.85 dB, SSIM 0.9809, and SAM 0.7932 on the full testing set, ahead of GINet, MMNet, DSen2, and ten other classical and learning-based methods, and the same top-two ordering repeats inside each of the Urban, Rural, and Coastal test sets. The paper further attributes the improvement to two design choices: a cluster-based module that maps 10m pixels into the spectral space of the 20m bands to produce the guiding image (the + variant), and a multi-head attention module that computes patch-based nonlocal similarity weights between the back-projected error, the guiding image, and their concatenation, which together reduce ringing artifacts while injecting 10m geometry into the reconstruction.

Load-bearing premise

The paper's gains are measured on simulated degradation: the 10m and 20m bands are downsampled by a factor of two, the network learns to invert that simulated 40m-to-20m degradation, and the authors assume the same learned mapping transfers to the real 20m-to-10m sharpening problem, which requires the true sensor degradation to match the downsampling model used in training.

Editorial extensions

If this is right

  • Sentinel-2 analysis chains that currently use 20m bands (red edge, SWIR) could ingest 10m sharpened versions, improving the spatial fidelity of indexes like NDMI and NDWI and of false-color urban composites.
  • The unfolded back-projection architecture with a learned nonlocal kernel gives a template for sharpening multi-resolution satellites that lack a panchromatic band, such as applying the same scheme to the 60m bands using the 10m and 20m bands as guidance.
  • Because GINet+ kept its top-two ranking across urban, rural, and coastal subsets without landscape-specific retraining, the learned guidance and attention appear to be scene-agnostic rather than tuned to one land-cover type.
  • The cluster-based guiding image, which learns a mapping from 10m reflectance to the 20m spectral space per land-cover cluster, could replace hand-crafted spectral-similarity rules in other band-ratio and compositing tasks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported margins (roughly 1 dB PSNR over MMNet) are measured only under Wald's protocol, where the degradation is a downsampling the network is trained to invert; on real 20m-to-10m sharpening, where the sensor's point-spread function and registration errors enter, the margins could shrink, and a sensor-realistic validation is the natural next test.
  • The paper does not ablate the cluster-based guiding image against a simple learned linear regression from the 10m bands; such an ablation would show whether the cluster structure (L=5) is carrying the gain or merely acting as a piecewise-linear regressor.
  • Because the MHA module restricts patch interactions to a 5x5 window, the 'nonlocal' component is local in practice; replacing it with a true nonlocal search or an approximate fast nonlocal filter would test whether the attention mechanism's benefit comes from the windowed patch similarity or from the extra parameters.
  • The method's applicability to other sensors (e.g., hyperspectral missions without a pan band) is plausible, but the cluster-based guiding image assumes the fine bands and coarse bands have a stable per-cluster spectral relationship, which may not hold across sensors with different band sets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper proposes a supervised deep-learning method for sharpening Sentinel-2 20m bands to 10m using the 10m bands as guidance. The architecture is an unfolded back-projection scheme whose correction kernel is replaced by a residual network with multi-head attention; a cluster-based module generates a guiding image from the 10m bands. Two variants are tested: GINet (spectral-similarity guide) and GINet+ (cluster-learned guide). The authors construct a dataset following Wald's protocol, with 500 training, 100 validation, and 900 test crops across urban/rural/coastal areas, and compare against 12 classical and deep baselines using ERGAS, PSNR, SSIM, and SAM. Tables 1-4 report GINet+ and GINet as the top two methods on all metrics. Ablations examine the residual kernel variant, number of stages, patch/window sizes, and loss functions.

Significance. The empirical study is carefully structured: the test set is separate from training and validation, covers three landscape types, and the comparison includes both classical and recent deep-learning methods with metrics for spatial and spectral fidelity. The ablations give support to the role of the guiding image and the attention module. If the improvements transfer to real 20m-to-10m sharpening, the method would be practically valuable for Sentinel-2 applications. However, the central claim is established only on a simulated degradation, the downsampling kernel is unspecified, and the guiding image resolution differs between training and deployment. In addition, two core modules are drawn from unpublished preprints, which limits reproducibility. These concerns are substantial enough that the current evidence does not yet support the unqualified claim in the abstract.

major comments (3)
  1. [Section 4.1, Eq. (2)] The Wald-protocol experiments use a synthetic degradation whose blur kernel is never specified. Since the network replaces DB with a learned convolution plus average pooling, the reported margins in Tables 1-4 may reflect specialization to that particular degradation. Please state the exact downsampling operator, use a Sentinel-2 MTF-based degradation, and add an evaluation with a different degradation kernel or on native 20m data to separate model performance from simulator artifacts.
  2. [Section 3.2 and Section 4.1] In training and testing under the Wald protocol, the guiding image is generated from the downsampled 10m bands, i.e., at 20m resolution. At real deployment, the guide would be built from the original 10m bands, at 10m resolution. Because patch windows and attention are defined in pixel units and the cluster module is trained on degraded 10m bands, the learned behavior may not transfer across this resolution gap. The assumption that 40-to-20m and 20-to-10m mappings are equivalent (Section 2.1) is not validated. Please test the model on real 20m inputs (using, e.g., spectral indices or visual inspection) or adapt the training to match deployment resolution.
  3. [Section 3.1, Section 3.2, References] The two signature components are based on Pereira-Sánchez et al. (2025) and Pereira-Sánchez et al. (2024), references that are incomplete or listed as submitted. This prevents readers from assessing the novelty of the architecture or reproducing the method. Please provide full method details (or an appendix), release code and pretrained models, or give stable preprint identifiers.
minor comments (7)
  1. [Section 4.1] The Wald protocol is cited to Hallada and Cox (1983); the correct reference is Wald et al. (1997), which is already in the bibliography.
  2. [Section 3.3, Eq. (4)] The loss formula notation (L0 1,2, L0 2,2, Lα 1,2) is confusing; please define the exponent and subscripts explicitly and state that the L1 loss is a special case with α=0 and i=1.
  3. [Section 6, Table 7] The patch/window ablation reports only PSNR; for consistency with the rest of the paper, report ERGAS, SSIM, and SAM as well.
  4. [Section 3.2] The choice of L=5 clusters is not ablated; please justify it or include a sensitivity study.
  5. [Section 5.1] Tables 1-4 report point estimates from a single evaluation; state whether multiple seeds were used and report standard deviations if available.
  6. [References] Several reference entries contain placeholder text (e.g., the entries for Pereira-Sánchez et al. 2024 and 2025); these need completion.
  7. [Section 5.2] The qualitative comparison uses one image per landscape and non-calibrated color/error maps; consider reporting quantitative zoom statistics or using a common color scale for the error maps.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is benchmarked against external methods on held-out data, and the cited self-works are fully described in the paper rather than used as load-bearing evidence.

full rationale

The paper's central claim is that GINet and GINet+ outperform classical and deep learning-based super-resolution and fusion techniques. This is supported by comparisons against external methods (DSen2, ResSen2, FusionNet, SRPPNN, MMNet, AWFLN, UTeRM, LGCT, WINet, GSA, CNMF) on held-out Sentinel-2 crops from nine areas not used in training, with results reported on validation, full test, and three landscape subsets (Tables 1-4). The proposed architecture's two signature modules are attributed to the authors' own prior or preprint works (Pereira-Sanchez et al., 2024 and 2025), but the paper fully specifies both modules: the residual MHA network with three parallel attention heads is described in Section 3.1 and Figure 2, and the cluster-based guiding image procedure is described in Section 3.2 and Figure 3. These self-citations are provenance statements, not arguments that reduce a prediction to the cited works. No equation is defined in terms of the target result: Eq. (1) is the standard observation model, Eq. (2) is the standard nonlocal back-projection iteration, and the learned replacements (convolution plus pooling for DB, transposed convolution for upsampling, and ResNL for the back-projection kernel) are trained with an L1 loss against ground truth. The Wald-protocol evaluation on simulated 40m-to-20m data is the same protocol applied to all compared methods, so a win on that benchmark is not forced by construction; the absence of native-resolution 20m-to-10m validation is a generalization risk, not a circular reduction. The manuscript also does not specify the exact downsampling kernel used to create LR inputs, but this is an experimental-reporting limitation rather than circularity. Accordingly, no circular step meets the evidentiary bar of the analysis rules.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim leans on six hand-chosen hyperparameters and four domain assumptions. The most consequential free choices are the cluster count (L = 5), the stage count (K = 6), and the attention patch/window sizes; the last three were selected by searching on the full testing set rather than a held-out validation split, which mildly self-confirms the headline results. The domain assumptions are the standard observation model, the scale-invariance of the degradation under Wald's protocol, the self-similarity of satellite scenes, and the cluster-to-spectral mapping structure. No new physical entities are postulated; the guiding image is a learned intermediate representation, not an invented physical quantity.

free parameters (6)
  • Number of clusters L = 5
    Chosen by hand without an ablation; Section 3.2 states 'we choose L = 5'. The cluster partition determines the structure of the guiding image.
  • Number of back-projection stages K = 6
    Selected by comparing 3, 6, and 9 stages; Table 6 reports the comparison on the full testing set, and 6 stages wins.
  • Attention patch size = 3x3
    Selected from a 3x3/5x5/7x7 search in Table 7, evaluated on the full testing set rather than a held-out split.
  • Attention window size = 5x5
    Selected from the same Table 7 search on the full testing set.
  • Loss function and stage weight alpha = L1, (alpha = 0)
    Table 8 compares L1, MSE, and stage-weighted losses with alpha = 0.1 and 0.5 on the testing set; plain L1 was chosen.
  • Learning rate and epoch count = 1e-4, 1500 epochs
    Fixed in Section 3.3 with no sensitivity study; standard Adam settings.
assumptions (4)
  • domain assumption Observation model f = DBu + eta with known low-pass blur B and decimation D (Eq. 1)
    Section 3.1 anchors the unfolding scheme; if the blur/decimation model mismatches the real sensor, the back-projection correction is misaligned.
  • domain assumption Wald protocol validity: the simulated 2x degradation (40m from 20m, 20m from 10m) equals the real 20m-to-10m relationship
    Section 4.1 builds the whole benchmark on this; gains are measured on simulated data and assumed to transfer to real acquisitions.
  • domain assumption Self-similarity of satellite imagery: every patch has similar patches elsewhere in the image
    The MHA nonlocal module (Section 3.1) assumes this holds for S2 scenes; ablations show it helps but cannot establish the assumption for all landscapes.
  • domain assumption Pixel clusters in 10m-band space map per cluster, via an MLP, to the spectral space of the 20m bands
    The SpecUp module (Section 3.2, Fig. 3) postulates this structure for the guiding image; the cluster-specific MLP is learned end-to-end with the SR loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Super-Resolution of Sentinel-2 Images Using a Geometry-Guided Back-Projection Network with Self-Attention." pith.science (2026). https://pith.science/paper/5WNYD7TK

@misc{pith2026250804729,
  author       = {Pith},
  title        = {Pith review of: Super-Resolution of Sentinel-2 Images Using a Geometry-Guided Back-Projection Network with Self-Attention},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5WNYD7TK}},
  note         = {Machine review of arXiv:2508.04729}
}
read the original abstract

The Sentinel-2 mission provides multispectral imagery with 13 bands at resolutions of 10m, 20m, and 60m. In particular, the 10m bands offer fine structural detail, while the 20m bands capture richer spectral information. In this paper, we propose a geometry-guided super-resolution model for fusing the 10m and 20m bands. Our approach introduces a cluster-based learning procedure to generate a geometry-rich guiding image from the 10m bands. This image is integrated into an unfolded back-projection architecture that leverages image self-similarities through a multi-head attention mechanism, which models nonlocal patch-based interactions across spatial and spectral dimensions. We also generate a dataset for evaluation, comprising three testing sets that include urban, rural, and coastal landscapes. Experimental results demonstrate that our method outperforms both classical and deep learning-based super-resolution and fusion techniques.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

78 extracted references · 74 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    , author Focareta, M

    author Addabbo, P. , author Focareta, M. , author Marcuccio, S. , author Votto, C. , author Ullo, S.L. , et al., year 2016 . title Contribution of sentinel-2 data for applications in vegetation monitoring . journal Acta Imeko volume 5 , pages 44--54

    work page 2016

  3. [3]

    , author Alparone, L

    author Aiazzi, B. , author Alparone, L. , author Baronti, S. , author Garzelli, A. , author Selva, M. , year 2003 . title An mtf-based spectral distortion minimizing model for pan-sharpening of very high resolution multispectral images of urban areas , in: booktitle 2003 2nd GRSS/ISPRS Joint Workshop on Remote Sensing and Data Fusion over Urban Areas , or...

    work page 2003

  4. [4]

    , author Aksu, K

    author Aldo g an, C. , author Aksu, K. , author Demirel, H. , year 2024 . title Enhancement of sentinel-2a images for ship detection via real-esrgan model . journal Applied Sciences volume 14 , pages 11988

    work page 2024

  5. [5]

    , author Arienzo, A

    author Alparone, L. , author Arienzo, A. , author Garzelli, A. , year 2024 . title Spatial resolution enhancement of satellite hyperspectral data via nested hypersharpening with sentinel-2 multispectral data . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 17 , pages 10956--10966

    work page 2024

  6. [6]

    , author Baronti, S

    author Alparone, L. , author Baronti, S. , author Garzelli, A. , author Nencini, F. , year 2004 . title A global quality measurement of pan-sharpened multispectral imagery . journal IEEE Geoscience and Remote Sensing Letters volume 1 , pages 313--317

    work page 2004

  7. [7]

    , author Caselles, V

    author Ballester, C. , author Caselles, V. , author Igual, L. , author Verdera, J. , author Roug \'e , B. , year 2006 . title A variational model for p+ xs image fusion . journal International Journal of Computer Vision volume 69 , pages 43--58

    work page 2006

  8. [8]

    , author Huang, B

    author Cai, J. , author Huang, B. , year 2021 . title Super-resolution-guided progressive pansharpening based on a deep convolutional neural network . journal IEEE Transactions on Geoscience and Remote Sensing volume 59 , pages 5206--5220

    work page 2021

Show all 78 references
  1. [9]

    , author Han, M

    author Dai, S. , author Han, M. , author Wu, Y. , author Gong, Y. , year 2007 . title Bilateral back-projection for single image super resolution , in: booktitle 2007 IEEE International Conference on Multimedia and Expo , pp. pages 1039--1042

    2007

  2. [10]

    , author Vivone, G

    author Deng, L.J. , author Vivone, G. , author Jin, C. , author Chanussot, J. , year 2020 . title Detail injection-based deep convolutional neural networks for pansharpening . journal IEEE Transactions on Geoscience and Remote Sensing volume 59 , pages 6995--7010

    2020

  3. [11]

    , author Kalaitzis, A

    author Deudon, M. , author Kalaitzis, A. , author Goytom, I. , author Arefin, M. , author Lin, Z. , author Sankaran, K. , author Michalski, V. , author Kahou, S. , author Cornebise, J. , author Bengio, Y. , year 2020 . title Highres-net: Recursive fusion for multi-frame super-...

    2020 arXiv

  4. [12]

    , author Loy, C

    author Dong, C. , author Loy, C. , author He, K. , author Tang, X. , year 2014 . title Learning a deep convolutional network for image super-resolution , in: booktitle Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings,...

    2014

  5. [13]

    , author Zhang, L

    author Dong, W. , author Zhang, L. , author Shi, G. , author Wu, X. , year 2009 . title Nonlocal back-projection for adaptive image enlargement , in: booktitle 2009 16th IEEE International Conference on Image Processing (ICIP) , pp. pages 349--352

    2009

  6. [14]

    , author Beyer, L

    author Dosovitskiy, A. , author Beyer, L. , author Kolesnikov, A. , author Weissenborn, D. , author Zhai, X. , author Unterthiner, T. , author Dehghani, M. , author Minderer, M. , author Heigold, G. , author Gelly, S. , et al., year 2020 . title An image is worth 16x16 words: ...

    2020 arXiv

  7. [15]

    , author Zhang, Y

    author Du, Y. , author Zhang, Y. , author Ling, F. , author Wang, Q. , author Li, W. , author Li, X. , year 2016 . title Water bodies’ mapping from sentinel-2 imagery with modified normalized difference water index at 10-m spatial resolution produced by sharpening the swir ban...

    2016

  8. [16]

    , author Buades, A

    author Duran, J. , author Buades, A. , author Coll, B. , author Sbert, C. , year 2014 . title A nonlocal variational model for pansharpening image fusion . journal SIAM Journal on Imaging Sciences volume 7 , pages 761--796

    2014

  9. [17]

    title Sentinel-2 — sentiwiki

    author European Space Agency , . title Sentinel-2 — sentiwiki . howpublished https://sentiwiki.copernicus.eu/web/sentinel-2 . note Accessed: 2025-05-16

    2025

  10. [18]

    , author Haut, J

    author Fernandez-Beltran, R. , author Haut, J. , author Paoletti, M. , author Plaza, J. , author Plaza, A. , author Pla, F. , year 2018 . title Multimodal probabilistic latent semantic analysis for sentinel-1 and sentinel-2 image fusion . journal IEEE Geoscience and Remote Sen...

    2018

  11. [19]

    , author Sesma, R

    author Galar, M. , author Sesma, R. , author Ayala, C. , author Aranda, C. , year 2019 . title Super-resolution for sentinel-2 images . journal The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences volume 42 , pages 95--102

    2019

  12. [20]

    , year 1996

    author Gao, B. , year 1996 . title Ndwi—a normalized difference water index for remote sensing of vegetation liquid water from space . journal Remote Sensing of Environment volume 58 , pages 257--266

    1996

  13. [21]

    , author Bouzinac, C

    author Gascon, F. , author Bouzinac, C. , author Th \'e paut, O. , author Jung, M. , author Francesconi, B. , author Louis, J. , author Lonjou, V. , author Lafrance, B. , author Massera, S. , author Gaudel-Vacaresse, A. , et al., year 2017 . title Copernicus sentinel-2a calibr...

    2017

  14. [22]

    , author Jogun, T

    author Ga s parovi \'c , M. , author Jogun, T. , year 2018 . title The effect of fusing sentinel-2 bands on land-cover classification . journal International journal of remote sensing volume 39 , pages 822--841

    2018

  15. [23]

    chromaticity

    author Gillespie, A. , author Kahle, A. , author Walker, R. , year 1987 . title Color enhancement of highly correlated images. ii. channel ratio and “chromaticity” transformation techniques . journal Remote Sensing of Environment volume 22 , pages 343--365

    1987

  16. [24]

    , author Cox, S

    author Hallada, W. , author Cox, S. , year 1983 . title Image sharpening for mixed spatial and spectral resolution satellite systems , in: booktitle International Symposium on Remote Sensing of Environment , pp. pages --

    1983

  17. [25]

    , author Shakhnarovich, G

    author Haris, M. , author Shakhnarovich, G. , author Ukita, N. , year 2018 . title Deep back-projection networks for super-resolution , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 1664--1673

    2018

  18. [26]

    , author Fu, X

    author He, W. , author Fu, X. , author Li, N. , author Ren, Q. , author Jia, S. , year 2024 . title Lgct: Local-global collaborative transformer for fusion of hyperspectral and multispectral images . journal IEEE Transactions on Geoscience and Remote Sensing volume 62 , pages 1--14

    2024

  19. [27]

    , author Peleg, S

    author Irani, M. , author Peleg, S. , year 1991 . title Improving resolution by image registration . journal CVGIP: Graphical Models and Image Processing volume 53 , pages 231--239

    1991

  20. [28]

    , author Peleg, S

    author Irani, M. , author Peleg, S. , year 1993 . title Motion analysis for image enhancement: Resolution, occlusion, and transparency . journal Journal of Visual Communication and Image Representation volume 4 , pages 324--335

    1993

  21. [29]

    , year 2018

    author Kaplan, G. , year 2018 . title Sentinel-2 pan sharpening—comparative analysis . journal Proceedings volume 2

    2018

  22. [30]

    , author Miyamoto, H

    author Kato, S. , author Miyamoto, H. , author Amici, S. , author Oda, A. , author Matsushita, H. , author Nakamura, R. , year 2021 . title Automated classification of heat sources detected using swir remote sensing . journal International Journal of Applied Earth Observation ...

    2021

  23. [31]

    , author Chanussot, J

    author Khan, M. , author Chanussot, J. , author Condat, L. , author Montanvert, A. , year 2008 . title Indusion: Fusion of multispectral and panchromatic images using the induction scaling technique . journal IEEE Geoscience and Remote Sensing Letters volume 5 , pages 98--102

    2008

  24. [32]

    , year 2014

    author Kingma, D.P. , year 2014 . title Adam: A method for stochastic optimization . journal arXiv preprint arXiv:1412.6980

    2014 arXiv

  25. [33]

    , author Kristollari, V

    author Kremezi, M. , author Kristollari, V. , author Karathanassi, V. , author Topouzelis, K. , author Kolokoussis, P. , author Taggio, N. , author Aiello, A. , author Ceriola, G. , author Barbone, E. , author Corradi, P. , year 2022 . title Increasing the sentinel-2 potential...

    2022

  26. [34]

    , author Bioucas-Dias, J

    author Lanaras, C. , author Bioucas-Dias, J. , author Baltsavias, E. , author Schindler, K. , year 2017 . title Super-resolution of multispectral multiresolution images from a single sensor , in: booktitle Proceedings of the IEEE Conference on Computer Vision and Pattern Recog...

    2017

  27. [35]

    , author Bioucas-Dias, J

    author Lanaras, C. , author Bioucas-Dias, J. , author Galliani, S. , author Baltsavias, E. , author Schindler, K. , year 2018 . title Super-resolution of sentinel-2 images: Learning a globally applicable deep neural network . journal ISPRS Journal of Photogrammetry and Remote ...

    2018

  28. [36]

    , author K \"o rner, M

    author Liebel, L. , author K \"o rner, M. , year 2016 . title Single-image super resolution for multispectral remote sensing data using convolutional neural networks . journal The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences vol...

    2016

  29. [37]

    , author Son, S

    author Lim, B. , author Son, S. , author Kim, H. , author Nah, S. , author Mu Lee, K. , year 2017 . title Enhanced deep residual networks for single image super-resolution , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition workshops ,...

    2017

  30. [38]

    , author Bioucas-Dias, J

    author Lin, C.H. , author Bioucas-Dias, J. , year 2020 . title An explicit and scene-adapted definition of convex self-similarity prior with application to unsupervised sentinel-2 super-resolution . journal IEEE Transactions on Geoscience and Remote Sensing volume 58 , pages 3...

    2020

  31. [39]

    , author Yang, Y

    author Lu, H. , author Yang, Y. , author Huang, S. , author Chen, X. , author Chi, B. , author Liu, A. , author Tu, W. , year 2023 . title Awfln: An adaptive weighted feature learning network for pansharpening . journal IEEE Transactions on Geoscience and Remote Sensing volume...

    2023

  32. [40]

    , author Lam, E

    author Mai, T. , author Lam, E. , author Lee, C. , year 2024 . title Deep unfolding tensor rank minimization with generalized detail injection for pansharpening . journal IEEE Transactions on Geoscience and Remote Sensing

    2024

  33. [41]

    , author Rott, H

    author Malenovsk \`y , Z. , author Rott, H. , author Cihlar, J. , author Schaepman, M. , author Garc \' a-Santos, G. , author Fernandes, R. , author Berger, M. , year 2012 . title Sentinels for science: Potential of sentinel-1,-2, and-3 missions for scientific observations of ...

    2012

  34. [42]

    , year 1996

    author McFeeters, S.K. , year 1996 . title The use of the normalized difference water index (ndwi) in the delineation of open water features . journal International Journal of Remote Sensing volume 17 , pages 1425--1432

    1996

  35. [43]

    , author Brody, W.R

    author Medoff, B.P. , author Brody, W.R. , author Nassi, M. , author Macovski, A. , year 1983 . title Iterative convolution backprojection algorithms for image reconstruction from limited data . journal JOSA volume 73 , pages 1493--1500

    1983

  36. [44]

    , author Ulfarsson, M

    author Nguyen, H. , author Ulfarsson, M. , author Sveinsson, J. , author Dalla Mura, M. , year 2021 . title Sentinel-2 sharpening using a single unsupervised convolutional neural network with mtf-based degradation model . journal IEEE Journal of Selected Topics in Applied Eart...

    2021

  37. [45]

    , author Ulfarsson, M

    author Nguyen, H. , author Ulfarsson, M. , author Sveinsson, J. , author Mura, M. , year 2023 . title Unsupervised sentinel-2 image fusion using a deep unrolling method . journal IEEE Geoscience and Remote Sensing Letters volume 20 , pages 1--5

    2023

  38. [46]

    , author Sveinsson, J

    author Palsson, F. , author Sveinsson, J. , author Ulfarsson, M. , year 2018 . title Sentinel-2 image fusion using a deep residual network . journal Remote Sensing volume 10

    2018

  39. [47]

    , author Choi, J

    author Park, H. , author Choi, J. , author Park, N. , author Choi, S. , year 2017 . title Sharpening the vnir and swir bands of sentinel-2a imagery through modified selected and synthesized band schemes . journal Remote Sensing volume 9

    2017

  40. [48]

    , author Sans, E

    author Pereira-S \'a nchez, I. , author Sans, E. , author Navarro, J. , author Duran, J. , year 2024 . title A comprehensive overview of satellite image fusion: From classical model-based to cutting-edge deep learning approaches . journal Super-Resolution for Remote Sensing , ...

    2024

  41. [49]

    , author Navarro, J

    author Pereira-Sánchez, I. , author Navarro, J. , author Petro, A. , author Duran, J. , year 2025 . title Model-guided network with cluster-based operators for spatio-spectral super-resolution . http://arxiv.org/abs/2505.24605 arXiv:2505.24605

    2025 arXiv

  42. [50]

    , author Sans, E

    author Pereira-Sánchez, I. , author Sans, E. , author Navarro, J. , author Duran, J. , year 2024 . title Multi-head attention residual unfolded network for model-based pansharpening . journal Submitted http://arxiv.org/abs/2409.02675 arXiv:2409.02675

    2024 arXiv

  43. [51]

    , year 1981

    author Peters, T.M. , year 1981 . title Algorithms for fast back- and re-projection in computed tomography . journal IEEE Transactions on Nuclear Science volume 28 , pages 3641--3647

    1981

  44. [52]

    , author Wu, Z

    author Qin, X. , author Wu, Z. , author Luo, X. , author andD.Dineng Zhao, J.S. , author Zhou, J. , author Cui, J. , author Wan, H. , author Xu, G. , year 2024 . title Musrfm: Multiple scale resolution fusion based precise and robust satellite derived bathymetry model for isla...

    2024

  45. [53]

    , author Jones, P

    author Rabbani, M. , author Jones, P. , year 1991 . title Digital Image Compression Techniques . edition 1st ed., publisher Society of Photo-Optical Instrumentation Engineers (SPIE) , address USA

    1991

  46. [54]

    , author Wald, L

    author Ranchin, T. , author Wald, L. , year 2000 . title Fusion of high spatial and spectral resolution images: The arsis concept and its implementation . journal Photogrammetric engineering and remote sensing volume 66 , pages 49--61

    2000

  47. [55]

    , author Mateo-Garc \' a, G

    author Razzak, M. , author Mateo-Garc \' a, G. , author Lecuyer, G. , author G \'o mez-Chova, L. , author Gal, Y. , author Kalaitzis, F. , year 2023 . title Multi-spectral multi-image super-resolution of sentinel-2 with radiometric consistency losses and its effect on building...

    2023

  48. [56]

    , author Marcello, J

    author Salgueiro Romero, L. , author Marcello, J. , author Vilaplana, V. , year 2020 . title Super-resolution of sentinel-2 imagery using generative adversarial networks . journal Remote Sensing volume 12 , pages 2424

    2020

  49. [57]

    , author Turner, E

    author Seager, S. , author Turner, E. , author Schafer, J. , author Ford, E. , year 2005 . title Vegetation’s red edge: A possible spectroscopic biosignature of extraterrestrial plants . journal Astrobiology volume 5 , pages 372–390

    2005

  50. [58]

    , author Buchaillot, M

    author Segarra, J. , author Buchaillot, M. , author Araus, J. , author Kefauver, S. , year 2020 . title Remote sensing for precision agriculture: Sentinel-2 improved features and applications . journal Agronomy volume 10 , pages 641

    2020

  51. [59]

    , author Aiazzi, B

    author Selva, M. , author Aiazzi, B. , author Butera, F. , author Chiarantini, L. , author Baronti, S. , year 2015 . title Hyper-sharpening: A first approach on sim-ga data . journal IEEE Journal of selected topics in applied earth observations and remote sensing volume 8 , pa...

    2015

  52. [60]

    , year 1992

    author Shettigara, V. , year 1992 . title A generalized component substitution technique for spatial enhancement of multispectral images using a higher resolution data set . journal Photogrammetric Engineering and remote sensing volume 58 , pages 561--567

    1992

  53. [61]

    , author Nalepa, J

    author Tarasiewicz, T. , author Nalepa, J. , author Farrugia, R. , author Valentino, G. , author Chen, M. , author Briffa, J. , author Kawulok, M. , year 2023 . title Multitemporal and multispectral data fusion for super-resolution of sentinel-2 images . journal IEEE Transacti...

    2023

  54. [62]

    , author Reinartz, P

    author Traganos, D. , author Reinartz, P. , year 2018 . title Mapping mediterranean seagrasses with sentinel-2 imagery . journal Marine pollution bulletin volume 134 , pages 197--209

    2018

  55. [63]

    , author Palsson, F

    author Ulfarsson, M. , author Palsson, F. , author Dalla Mura, M. , author Sveinsson, J. , year 2019 . title Sentinel-2 sharpening using a reduced-rank method . journal IEEE Transactions on Geoscience and Remote Sensing volume 57 , pages 6408--6420

    2019

  56. [64]

    , author Fonseca, J

    author Vaqueiro, M. , author Fonseca, J. , author Oliveira, H. , author Mora, A. , year 2021 . title Multi-image super-resolution algorithm supported on sentinel-2 satellite images geolocation error , in: booktitle 2021 International Young Engineers Forum (YEF-ECE) , organizat...

    2021

  57. [65]

    , author Ranchin, T

    author Wald, L. , author Ranchin, T. , author Mangolini, M. , year 1997 . title Fusion of satellite images of different spatial resolutions: Assessing the quality of resulting images . journal Photogrammetric Engineering and Remote Sensing volume 63 , pages 691--699

    1997

  58. [66]

    , author Huang, B

    author Wang, J. , author Huang, B. , author Zhang, H. , author Ma, P. , year 2019 . title Sentinel-2a image fusion using a machine learning approach . journal IEEE Transactions on Geoscience and Remote Sensing volume 57 , pages 9589--9601

    2019

  59. [67]

    , author Shi, W

    author Wang, Q. , author Shi, W. , author Atkinson, P. , author Pardo-Ig \'u zquiza, E. , year 2015 . title A new geostatistical solution to remote sensing image downscaling . journal IEEE transactions on geoscience and remote sensing volume 54 , pages 386--396

    2015

  60. [68]

    , author Shi, W

    author Wang, Q. , author Shi, W. , author Li, Z. , author Atkinson, P. , year 2016 . title Fusion of sentinel-2 images . journal Remote Sensing of Environment volume 187 , pages 241--252

    2016

  61. [69]

    , author Bovik, A

    author Wang, Z. , author Bovik, A. , year 2002 . title A universal image quality index . journal IEEE Signal Processing Letters volume 9 , pages 81--84

    2002

  62. [70]

    , author Lin, L

    author Wu, J. , author Lin, L. , author Zhang, C. , author Li, T. , author Cheng, X. , author Nan, F. , year 2023 . title Generating sentinel-2 all-band 10-m data by sharpening 20/60-m bands: A hierarchical fusion network . journal ISPRS Journal of Photogrammetry and Remote Se...

    2023

  63. [71]

    , author Zhou, M

    author Yan, K. , author Zhou, M. , author Zhang, L. , author Xie, C. , year 2022 . title Memory-augmented model-driven network for pansharpening , in: editor Avidan, S. , editor Brostow, G. , editor Ciss \'e , M. , editor Farinella, G. , editor Hassner, T. (Eds.), booktitle Co...

    2022

  64. [72]

    , author Fu, X

    author Yang, J. , author Fu, X. , author Hu, Y. , author Huang, Y. , author Ding, X. , author Paisley, J. , year 2017 . title Pannet: A deep network architecture for pan-sharpening , in: booktitle Proceedings of the IEEE international conference on computer vision , pp. pages ...

    2017

  65. [73]

    , author Yairi, T

    author Yokoya, N. , author Yairi, T. , author Iwasaki, A. , year 2011 . title Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion . journal IEEE Transactions on Geoscience and Remote Sensing volume 50 , pages 528--537

    2011

  66. [74]

    , author He, X

    author Zhang, J. , author He, X. , author Yan, K. , author Cao, K. , author Li, R. , author Xie, C. , author Zhou, M. , author Hong, D. , year 2024 . title Pan-sharpening with wavelet-enhanced high-frequency information . journal IEEE Transactions on Geoscience and Remote Sens...

    2024

  67. [75]

    , author Wang, A

    author Zhang, K. , author Wang, A. , author Zhang, F. , author Wan, W. , author Sun, J. , author Bruzzone, L. , year 2023 . title Spatial-spectral dual back-projection network for pansharpening . journal IEEE Transactions on Geoscience and Remote Sensing volume 61 , pages 1--16

    2023

  68. [76]

    , author Su, W

    author Zhang, M. , author Su, W. , author Fu, Y. , author Zhu, D. , author Xue, J.H. , author Huang, J. , author Wang, W. , author Wu, J. , author Yao, C. , year 2019 . title Super-resolution enhancement of sentinel-2 image for retrieving lai and chlorophyll content of summer ...

    2019

  69. [77]

    , author Liu, Q

    author Zhou, H. , author Liu, Q. , author Wang, Y. , year 2022 . title Panformer: A transformer based model for pan-sharpening , in: booktitle 2022 IEEE International Conference on Multimedia and Expo (ICME) , organization IEEE . pp. pages 1--6

    2022

  70. [78]

    , author Rav-Acha, A

    author Zomet, A. , author Rav-Acha, A. , author Peleg, S. , year 2001 . title Robust super-resolution , in: booktitle Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001 , pp. pages I--I

    2001

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.