Pith. sign in

REVIEW 5 major objections 3 minor 62 references

COS2A: Conversion from Sentinel-2 to AVIRIS Hyperspectral Data Using Interpretable Algorithm With Spectral-Spatial Duality

T0 review · 5 major / 3 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 Sentinel-2 to AVIRIS: 12 bands in, 172 bands out.

desk verdict A genuinely new S2-to-AVIRIS conversion scheme whose algebra holds together, but whose empirical claims rest on a weak baseline and a deep solution that carries all the spectral burden. read the letter →

arxiv 2507.06575 v1 pith:7EKE5GNI submitted 2025-07-09 eess.IV eess.SP

Chia-Hsiang Lin , Jui-Ting Chen , Zi-Chao Leng , Jhao-Ting Lin This is my paper · ORCID
classification eess.IVeess.SP
keywords spectralsuper-resolutionhyperspectralimagingSentinel-2AVIRISdeepunfoldingconvexoptimizationcouplednonnegativematrixfactorizationspectral-spatialduality
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

The paper tries to establish that a 12-band, multi-resolution Sentinel-2 image can be computationally converted into a 172-band AVIRIS-grade hyperspectral image at a uniform 10-meter resolution, and that this conversion works on real satellite data for a variety of land cover types. If true, this would let the entire historical Sentinel-2 archive be reprocessed into high-standard hyperspectral data, making remote identification of materials, vegetation, and surface changes far more reliable than the original 12 bands allow. The proposed COS2A algorithm combines a lightweight deep-unfolding network with a convex optimization stage, and it reports a large quantitative gain over the divide-and-conquer baseline. The paper also claims that the conversion is interpretable because the hard spectral super-resolution problem is mathematically transformed into a well-studied spatial super-resolution problem.

What carries the argument

The machinery is the convex/deep (CODE) learning criterion together with a spectral-spatial duality theorem. The criterion regularizes the target hyperspectral image with a Q-quadratic norm centered at a rough deep solution, where Q is built from a blurring kernel that models the multi-resolution optical blur of Sentinel-2. The duality theorem rewrites the COS2A criterion as a coupled-NMF spatial super-resolution problem, allowing the author's algorithm to call a fast convex coupled-NMF solver and to use the estimated spectral response only for the four 10-meter bands.

What would settle it

Take a real Sentinel-2 scene containing a material not represented in the AVIRIS training set (for example, a unique mineral or artificial surface), run Algorithm 1, and compare the output spectrum at that pixel against a real AVIRIS overflight of the same scene; if the recovered spectrum diverges sharply from the measured AVIRIS reflectance, the claim that COS2A generalizes across diverse land cover types would fail.

Watch

Extended reading notes

Core claim

The central claim is that the COS2A problem, converting a multi-resolution 12-band Sentinel-2 image into a 172-band AVIRIS-level image, can be solved for the first time by an interpretable two-phase algorithm. A rough deep solution is obtained from a simple deep-unfolding network regularized by deep image prior, and this rough solution is then embedded into a convex criterion via a Q-quadratic-norm regularizer. The paper proves a spectral-spatial duality theorem showing the resulting optimization is equivalent to a coupled nonnegative matrix factorization spatial super-resolution problem, which is then solved by an existing convex algorithm. The output is the product of an endmember matrix and an abundance matrix, and experiments report an average PSNR of 35.0 dB versus 27.8 dB for the baseline, with substantially lower spectral angle error.

Load-bearing premise

The 172-band spectral information must already be present in the rough deep solution YDE, because the coupled-NMF stage only adds spatial detail and cannot create spectral bands that YDE missed.

Editorial extensions

If this is right

  • Historical Sentinel-2 acquisitions can, in principle, be reprocessed into AVIRIS-level hyperspectral data with uniform 10-meter spatial resolution.
  • The conversion handles the 10/20/60-meter resolution mismatch directly, using only the four 10-meter bands in the data-fidelity term and injecting spatial detail through the dual coupled-NMF stage.
  • The interpretable design means both the deep-unfolding network and the convex phase can be audited and trained with a relatively small paired dataset.
  • Quantitative results show a large margin over the divide-and-conquer baseline: average PSNR rises from 27.8 to 35.0 dB and average spectral angle falls from 7.71 to 2.50 degrees.
  • Real-data tests across coastline, mountain, farm, and city scenes show reconstructed spectra closely tracking real AVIRIS curves.

Reading between the lines

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

  • The same spectral-spatial duality might transfer to other multispectral satellites, such as Landsat, provided a scene-adaptive spectral response can be estimated for the target hyperspectral sensor.
  • The ceiling of the method is set by the rough deep solution, since the coupled-NMF stage fuses spatial detail but cannot invent spectral bands missing from the deep solution; an ablation comparing the final output with YDE alone would quantify how much the duality stage actually adds.
  • The paper's claim about the whole historical Sentinel-2 archive presumes generalization beyond the August 2019 American scenes used for training; testing on other seasons, regions, or sensors would settle that assumption.
  • Because the dual problem is a spatial super-resolution, better spatial-fusion algorithms could be swapped into the convex stage without retraining the deep network, offering a modular path to further improvement.
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

5 major / 3 minor

Summary. The paper proposes COS2A, an algorithm to convert a 12-band multi-resolution Sentinel-2 image into a 172-band AVIRIS-like hyperspectral image at uniform 10 m resolution. The method first computes a rough deep solution YDE via a lightweight deep unfolding network (Algorithm 2), estimates a 4-band spectral response eD by ridge regression (Eq. (11)), and then solves a convex criterion (Eq. (4)) that combines data fidelity to the four 10-m Sentinel-2 bands, a Q-quadratic-norm regularizer pulling toward YDE, low-rank NMF structure, and volume/sparsity penalties. The main theoretical contribution is Theorem 1, a spectral-spatial duality that rewrites the criterion as a coupled-NMF spatial super-resolution problem, solvable by a fast convex algorithm. Experiments compare COS2A against a divide-and-conquer MST baseline (DAC) on simulated and real data, reporting higher PSNR/SSIM and lower SAM/RMSE.

Significance. The COS2A problem is new and practically relevant; the spectral-spatial duality (Theorem 1) is a clean, algebraically valid reduction, and the deep unfolding network is lightweight and interpretable. If the empirical claims are fully supported, the method could enable AVIRIS-level products from the Sentinel-2 archive. However, the current evidence is insufficient: the baseline is weak and non-representative, error bars are absent, the real-data validation is qualitative, and the spectral response for real data is inferred from the deep solution itself, creating a partial circularity. The central dependence on YDE's spectral content is not tested as a limitation.

major comments (5)
  1. [II-C, Theorem 1; Appendix A, Eq. (13)] The proof of Theorem 1 establishes ||AS−YDE||²_Q = ||YDEB−ASB||²_F. Thus the Q-regularizer in (4) only forces the block-averaged reconstruction ASB to match the block-averaged deep solution YDEB, while the data fidelity term uses only the four 10-m bands eYS. Consequently, all spectral information beyond these four bands must come from YDE; the coupled-NMF stage cannot repair spectral errors that YDE makes. This is the load-bearing assumption of the method, but the paper does not test it. Please add an experiment that systematically degrades YDE (e.g., removing the NIR bands from the deep network target) and show that the final output degrades correspondingly, and an out-of-distribution test on a scene type or acquisition period not represented in training.
  2. [II-D, Eq. (11); III-C] In the real-data experiment, eD is estimated by ridge regression using YDE as the surrogate for the true AVIRIS data, and the same eD is then plugged into criterion (4). The real-data evaluation therefore validates consistency with the deep solution rather than with the physical AVIRIS/Sentinel-2 spectral relationship. The per-pixel calibration scalar γ⋆ applied to the ground-truth AVIRIS pixels (Section III-C) further weakens the comparison by absorbing per-pixel gain errors. Please report quantitative real-data metrics against a globally or scene-wise calibrated reference, and evaluate the sensitivity of the reconstruction to eD by comparing eD estimated from YDE with eD estimated from held-out real AVIRIS/Sentinel-2 pairs.
  3. [III-B, Table II] The quantitative comparison uses a single baseline, DAC, which is a divide-and-conquer adaptation of MST, a 31-band visible-range method. This does not represent the current state of the art for remote-sensing spectral super-resolution; the paper itself cites SCDM [28] and US3RN [29] as methods that handle up to 102 bands. Additionally, Table II reports only averages over 10 images per land type without standard deviations, confidence intervals, or per-image values, so the consistency of the improvements cannot be assessed. Please add at least one recent remote-sensing baseline and report error bars or the full per-image results.
  4. [III-A and II-D (algorithm pipeline for real data)] The real-data pipeline is not fully specified. Section III-A states that D is known in simulation and 'should be estimated via ridge regression' for real data, but the only estimation criterion given, Eq. (11), estimates eD ∈ R^{4×172}, not the full 12×172 D required by Algorithm 2 and by the ADMM update (9). It is unclear how the deep solution YDE is computed for real Sentinel-2 images when the full D is unknown, and how the low/medium-resolution bands enter that computation. This missing step prevents reproduction of the central real-data experiment and must be clarified.
  5. [III-C] The real-data evaluation is qualitative: Figure 7 shows selected spectra and images, but no numerical error measure is reported for the real-data test. Moreover, all real-data pairs are from August 2019 sites in America (Table I), and the four land types match those in the simulated test, so the generalization claim stated in the abstract ('all historical Sentinel-2 data') is not demonstrated. Please include quantitative real-data results and test at least one scene type or acquisition date not present in training.
minor comments (3)
  1. [Section II-D, Eq. (10)] The vectorization convention used in the definition of ∥·∥_Q is not stated; since the proof of Theorem 1 relies on vec(CEF) = (F^T ⊗ C)vec(E), please specify column-major vectorization to avoid ambiguity.
  2. [Algorithm 2] The stopping criterion is referred to as 'predefined' but never defined; please state whether a fixed K stages are used or an epsilon-threshold.
  3. [Section III-C] The phrase 'randomly select five representative pixels with high spectral diversity' is self-contradictory; please describe the actual selection procedure (e.g., successive projection algorithm as used for Figure 5).

Circularity Check

1 steps flagged · score 4.0 of 10

Output spectra are anchored to the deep solution YDE: eD is fitted from YDE (Eq. 11) and the Q-regularizer pulls AS toward YDE (Eq. 13), so the convex phase only refines YDE and cannot independently produce AVIRIS spectral content.

  1. fitted input called prediction [Section II-D, Eq. (11); Algorithm 1 Steps 3-4; Section II-C Eq. (4); Appendix A Eq. (13)]
    "we adopt the ridge regression [56], which promotes uniformly distributed solutions, thereby leading to the estimation criterion, i.e., eD := arg min_{D≥0} ∥DYDE − eYS∥²_F + η∥D∥²_F, where η := 0.0001 is an empirical setting, and YDE is used as a surrogate of the real A VIRIS data; note that in practice, we do not have real A VIRIS data in the COS2A problem."

    Algorithm 1 fits the spectral response eD from YDE via ridge regression (Eq. 11), then plugs this same eD into the data-fidelity term of the COS2A criterion (Eq. 4, ∥eYS − eDAS∥²_F). Theorem 1 (Eq. 13) further shows that the Q-regularizer is exactly ∥YDEB − ASB∥²_F. Hence the convex objective contains no independent reference to AVIRIS spectra except through YDE: the 20-m and 60-m Sentinel-2 bands influence the objective only through YDE, and the fidelity term only checks consistency with the fitted projection of YDE. The output Y_H⋆ = A⋆S⋆ is therefore statistically forced to stay near YDE, and any spectral band or material absent from YDE cannot be recovered by the subsequent coupled-NMF step.

full rationale

The proof of Theorem 1 is a legitimate algebraic equivalence: it manipulates the definitions Q = BB^T ⊗ I_M and B = I_{L/r^2} ⊗ (1_{r^2}/r^2) to show ∥YDEB − ASB∥²_F = ∥AS − YDE∥²_Q, so the spectral-spatial duality itself is not circular. The deep unfolding stage (Algorithm 2) is trained on real A VIRIS/Sentinel-2 pairs, which gives the central claim independent empirical content. The genuine self-referential step is the calibration of eD from the same YDE that the Q-regularizer treats as its reference, making the data-fidelity term a self-consistency constraint rather than an independent sensor constraint. This limits the convex phase to spatial refinement and makes the final output inherit YDE's spectral failure modes, a limitation that the paper itself acknowledges when it states that YDE is used as a surrogate because real A VIRIS data are unavailable in the COS2A problem. The authors' self-citations to the CODE framework [30] and the coupled-NMF solver [33] are load-bearing but are published, checkable algorithms with independent derivations, so they do not raise the circularity score beyond this partial self-reference.

Assumptions & free parameters 8 free parameters · 7 assumptions · 0 invented entities

The central claim rests on a handful of tuned constants and on the assumption that the rough deep solution YDE already contains the 172-band spectral information. No invented physical entities are introduced.

free parameters (8)
  • Deep unfolding network weights (~0.75M parameters) = trained on 646 AVIRIS images with simulated Sentinel-2 pairs
    Produces YDE, which carries the 172-band spectral content and is central to the final reconstruction.
  • Deep unfolding stages K = 4
    Chosen as a 'simple rough solution' to keep the network lightweight (Section II-D).
  • ADMM penalty rho = trainable in network
    Penalty parameter in the augmented Lagrangian (7), learned during training (Section II-D).
  • Q-regularizer blur factor r = 2
    Chosen because six 20-m bands dominate Sentinel-2, while the two 60-m bands have factor 6 (Section II-D).
  • Regularization coefficients lambda, alpha, beta = lambda=2, alpha=beta=0.002
    Empirically set in (4) to balance data fidelity, volume, and sparsity (Section II-C).
  • Ridge regularization eta = 0.0001
    Used in spectral response estimation (11) to spread response across bands (Section II-D).
  • Model order N = 10
    Rank of YH=AS for NMF, empirically set as 10 per [33] (Section III-A).
  • Real-data spectral response eD = estimated per scene via ridge regression from YDE
    No official AVIRIS/Sentinel-2 response is used; eD is inferred from the deep solution YDE and the four 10-m bands (11), so errors in YDE propagate into the observation model.
assumptions (7)
  • domain assumption Hyperspectral image is low-rank: YH=AS with N much smaller than 172 and L.
    Invoked in Section II-B to constrain the ill-posed COS2A problem.
  • standard math Craig's minimum-volume simplex criterion recovers true endmembers under mild conditions.
    Used in Section II-C to justify the volume penalty in (4), based on cited references [4], [45], [46].
  • domain assumption Abundance matrix S is sparse, justifying the l1 penalty.
    Used in Section II-C and in objective (4).
  • domain assumption Deep image prior (DIP) is an adequate implicit regularizer for the rough solution.
    Section II-D, criterion (5), relies on DIP theory rather than a large external dataset.
  • ad hoc to paper Uniform blurring with factor r=2 correctly models Sentinel-2 20-m band degradation in the Q regularizer.
    Section II-D chooses r=2 because six 20-m bands dominate, but actual mixed-resolution degradation is more complex.
  • ad hoc to paper The rough deep solution YDE contains sufficient AVIRIS-level spectral information for the final NMF refinement.
    Algorithm 1 uses YDEB as the hyperspectral input to the coupled-NMF step; without this, 172-band recovery is impossible.
  • domain assumption The 172 uncorrupted AVIRIS bands are sufficient for material identification.
    Used to define the target space in Section II-A and the experiments; standard in AVIRIS literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of COS2A: Conversion from Sentinel-2 to AVIRIS Hyperspectral Data Using Interpretable Algorithm With Spectral-Spatial Duality." pith.science (2026). https://pith.science/paper/7EKE5GNI

@misc{pith2026250706575,
  author       = {Pith},
  title        = {Pith review of: COS2A: Conversion from Sentinel-2 to AVIRIS Hyperspectral Data Using Interpretable Algorithm With Spectral-Spatial Duality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7EKE5GNI}},
  note         = {Machine review of arXiv:2507.06575}
}
read the original abstract

The Sentinel-2 satellite, launched by the European Space Agency (ESA), offers extensive spatial coverage and has become indispensable in a wide range of remote sensing applications. However, it just has 12 spectral bands, making substances/objects identification less effective, not mentioning the varying spatial resolutions (10/20/60 m) across the 12 bands. If such a multi-resolution 12-band image can be computationally converted into a hyperspectral image with uniformly high resolution (i.e., 10 m), it significantly facilitates remote identification tasks. Though there are some spectral super-resolution methods, they did not address the multi-resolution issue on one hand, and, more seriously, they mostly focused on the CAVE-level hyperspectral image reconstruction (involving only 31 visible bands) on the other hand, greatly limiting their applicability in real-world remote sensing scenarios. We ambitiously aim to convert Sentinel-2 data directly into NASA's AVIRIS-level hyperspectral image (encompassing up to 172 visible and near-infrared (NIR) bands, after ignoring those absorption/corruption ones). For the first time, this paper solves this specific super-resolution problem (highly ill-posed), allowing all historical Sentinel-2 data to have their corresponding high-standard AVIRIS counterparts. We achieve so by customizing a novel algorithm that introduces deep unfolding regularization and Q-quadratic-norm regularization into the so-called convex/deep (CODE) small-data learning criterion. Based on the derived spectral-spatial duality, the proposed interpretable COS2A algorithm demonstrates superior spectral super-resolution results across diverse land cover types, as validated through extensive experiments.

Figures

Figures reproduced from arXiv: 2507.06575 by the authors.

Figure 1
Figure 1. The proposed COS2A spectral super-resolution algorithm (i.e., [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Model 1 implements the proximal denoiser of (8) using the DIP [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. , and then be fed into the model 1 to implement (8). On the other hand, by the definition of (7), the update of YH actually amounts to an unconstrained convex quadratic problem having the following solution, i.e., Y k+1 H = arg min YH ∥YS − DYH∥ 2 F + ρ 2 ∥YH − Z k+1 − U k ∥ 2 F = (2DT D + ρI) −1 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 6
Figure 6. Figure 6: Spectral reconstruction results of (a) DAC and (b) the proposed [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Their corresponding COS2A-generated pixels in [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 7
Figure 7. Figure 7: The real data testing results over (a) coastline, (b) mountain, (c) farm, and (d) city. For each land type, the real AVIRIS image [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages

  1. [28]

    Spectral-cascaded diffusion model for remote sensing image spectral super-resolution,

    B. Chen, L. Liu, C. Liu, Z. Zou, and Z. Shi, “Spectral-cascaded diffusion model for remote sensing image spectral super-resolution,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–14, Aug. 2024

    work page 2024

  2. [29]

    Deep unfolding network for spatiospectral image super-resolution,

    Q. Ma, J. Jiang, X. Liu, and J. Ma, “Deep unfolding network for spatiospectral image super-resolution,” IEEE Transactions on Compu- tational Imaging , vol. 8, pp. 28–40, Dec. 2021

    work page 2021

  3. [1]

    Sentinel-2 data for land cover/use mapping: A review,

    D. Phiri, M. Simwanda, S. Salekin, V . R. Nyirenda, Y . Murayama, and M. Ranagalage, “Sentinel-2 data for land cover/use mapping: A review,” Remote sensing , vol. 12, no. 14, pp. 1–35, Jul. 2020

    work page 2020

  4. [2]

    Remote sensing for precision agriculture: Sentinel-2 improved features and applications,

    J. Segarra, M. L. Buchaillot, J. L. Araus, and S. C. Kefauver, “Remote sensing for precision agriculture: Sentinel-2 improved features and applications,” Agronomy, vol. 10, no. 5, pp. 1–18, May 2020

    work page 2020

  5. [3]

    Nonnegative blind source separation for ill-conditioned mixtures via John ellipsoid,

    C.-H. Lin and J. M. Bioucas-Dias, “Nonnegative blind source separation for ill-conditioned mixtures via John ellipsoid,” IEEE Transactions on Neural Networks and Learning Systems , vol. 32, no. 5, pp. 2209–2223, Jul. 2020. 13

    work page 2020

  6. [4]

    Identifiability of the simplex volume minimization criterion for blind hyperspectral unmixing: The no-pure-pixel case,

    C.-H. Lin, W.-K. Ma, W.-C. Li, C.-Y . Chi, and A. Ambikapathi, “Identifiability of the simplex volume minimization criterion for blind hyperspectral unmixing: The no-pure-pixel case,” IEEE Transactions on Geoscience and Remote Sensing , vol. 53, no. 10, pp. 5530–5546, May 2015

    work page 2015

  7. [5]

    Hyperspectral change detection: An ex- perimental comparative study,

    M. Hasanlou and S. T. Seydi, “Hyperspectral change detection: An ex- perimental comparative study,”International Journal of Remote Sensing , vol. 39, no. 20, pp. 7029–7083, Apr. 2018

    work page 2018

  8. [6]

    A subspace-based change detection method for hyperspectral images,

    C. Wu, B. Du, and L. Zhang, “A subspace-based change detection method for hyperspectral images,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 6, no. 2, pp. 815– 830, Feb. 2013

    work page 2013

Show all 62 references
  1. [7]

    SuperRPCA: A collaborative superpixel representation prior-aided RPCA for hyperspectral anomaly detection,

    J.-T. Lin and C.-H. Lin, “SuperRPCA: A collaborative superpixel representation prior-aided RPCA for hyperspectral anomaly detection,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1– 16, Sep. 2024

    2024

  2. [8]

    Unsupervised abundance matrix reconstruction transformer-guided fractional attention mechanism for hyperspectral anomaly detection,

    S.-S. Young, C.-H. Lin, and Z.-C. Leng, “Unsupervised abundance matrix reconstruction transformer-guided fractional attention mechanism for hyperspectral anomaly detection,” IEEE Transactions on Neural Networks and Learning Systems , vol. 36, no. 5, pp. 9150–9164, May 2025

    2025

  3. [9]

    Learned spectral super-resolution,

    S. Galliani, C. Lanaras, D. Marmanis, E. Baltsavias, and K. Schindler, “Learned spectral super-resolution,” arXiv preprint arXiv:1703.09470 , Mar. 2017. [Online]. Available: https://arxiv.org/abs/1703.09470

    2017 arXiv

  4. [10]

    Spectral super-resolution meets deep learning: Achievements and challenges,

    J. He, Q. Yuan, J. Li, Y . Xiao, D. Liu, H. Shen, and L. Zhang, “Spectral super-resolution meets deep learning: Achievements and challenges,” Information Fusion , vol. 97, pp. 1–22, Sep. 2023

    2023

  5. [11]

    Spectral super-resolution via deep low- rank tensor representation,

    R. Dian, Y . Liu, and S. Li, “Spectral super-resolution via deep low- rank tensor representation,” IEEE Transactions on Neural Networks and Learning Systems , vol. 36, no. 3, pp. 5140–5150, Mar. 2025

    2025

  6. [12]

    Generalized assorted pixel camera: Postcapture control of resolution, dynamic range, and spectrum,

    F. Yasuma, T. Mitsunaga, D. Iso, and S. K. Nayar, “Generalized assorted pixel camera: Postcapture control of resolution, dynamic range, and spectrum,” IEEE Transactions on Image Processing , vol. 19, no. 9, pp. 2241–2253, Sep. 2010

    2010

  7. [13]

    Significant remote sensing vegetation indices: A review of developments and applications,

    J. Xue and B. Su, “Significant remote sensing vegetation indices: A review of developments and applications,” Journal of Sensors, vol. 2017, no. 1, pp. 1–17, May 2017

    2017

  8. [14]

    CODE-MM: Convex deep mangrove mapping algorithm based on optical satellite images,

    C.-H. Lin, M.-C. Chu, and P.-W. Tang, “CODE-MM: Convex deep mangrove mapping algorithm based on optical satellite images,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–19, Sep. 2023

    2023

  9. [15]

    A VIRIS Data Portal,

    “A VIRIS Data Portal,” [Online]. Available: https://aviris.jpl.nasa.gov/ dataportal/

  10. [16]

    Metasurface- empowered snapshot hyperspectral imaging with convex/deep (CODE) small-data learning theory,

    C.-H. Lin, S.-H. Huang, T.-H. Lin, and P.-C. Wu, “Metasurface- empowered snapshot hyperspectral imaging with convex/deep (CODE) small-data learning theory,” Nature Communications, vol. 14, no. 1, pp. 1–10, Nov. 2023

    2023

  11. [17]

    A fea- sibility study for signal-in-space design for LEO-PNT solutions with miniaturized satellites,

    R. M. Ferre, J. Praks, G. Seco-Granados, and E. S. Lohan, “A fea- sibility study for signal-in-space design for LEO-PNT solutions with miniaturized satellites,” IEEE Journal on Miniaturization for Air and Space Systems , vol. 3, no. 4, pp. 171–183, Sep. 2022

    2022

  12. [18]

    All-addition hyperspectral compressed sensinging for metasurface-driven miniaturized satellite,

    C.-H. Lin and T.-H. Lin, “All-addition hyperspectral compressed sensinging for metasurface-driven miniaturized satellite,” IEEE Trans- actions on Geoscience and Remote Sensinging , vol. 60, pp. 1–15, Mar. 2021

    2021

  13. [19]

    Advances in information processing and biological imaging using flat optics,

    X. Wang, H. Hao, X. He, P. Xie, J. Liu, J. Tan, H. Li, H. Wang, P. Genevet, Y . Luo, X. Ding, and G. Hu, “Advances in information processing and biological imaging using flat optics,” Nature Reviews Electrical Engineering , vol. 1, no. 6, pp. 391–411, May 2024

    2024

  14. [20]

    HyperQUEEN: Hyperspectral quantum deep network for image restoration,

    C.-H. Lin and Y .-Y . Chen, “HyperQUEEN: Hyperspectral quantum deep network for image restoration,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–20, May 2023

    2023

  15. [21]

    Quantum information- empowered graph neural network for hyperspectral change detection,

    C.-H. Lin, T.-H. Lin, and J. Chanussot, “Quantum information- empowered graph neural network for hyperspectral change detection,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1– 15, Nov. 2024

    2024

  16. [22]

    Gaussian processes for under- determined source separation,

    A. Liutkus, R. Badeau, and G. Richard, “Gaussian processes for under- determined source separation,” IEEE Transactions on Signal Processing , vol. 59, no. 7, pp. 3155–3167, Feb. 2011

    2011

  17. [23]

    PRIME: Blind multispectral unmixing using virtual quantum prism and convex geometry,

    C.-H. Lin and J.-T. Lin, “PRIME: Blind multispectral unmixing using virtual quantum prism and convex geometry,” IEEE Transactions on Geoscience and Remote Sensing , vol. 63, pp. 1–15, Feb. 2025

    2025

  18. [24]

    MST++: Multi-stage spectral-wise transformer for effi- cient spectral reconstruction,

    Y . Cai, J. Lin, Z. Lin, H. Wang, Y . Zhang, H. Pfister, R. Timofte, and L. Van-Gool, “MST++: Multi-stage spectral-wise transformer for effi- cient spectral reconstruction,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition , New Orleans, LA, USA, 18–24 Jun. ...

    2022

  19. [25]

    Multistage spatial-spectral fusion network for spectral super-resolution,

    Y . Wu, R. Dian, and S. Li, “Multistage spatial-spectral fusion network for spectral super-resolution,” IEEE Transactions on Neural Networks and Learning Systems , pp. 1–11, Oct. 2024

    2024

  20. [26]

    RepCPSI: Coordinate- preserving proximity spectral interaction network with reparameteriza- tion for lightweight spectral super-resolution,

    C. Wu, J. Li, R. Song, Y . Li, and Q. Du, “RepCPSI: Coordinate- preserving proximity spectral interaction network with reparameteriza- tion for lightweight spectral super-resolution,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–13, Apr. 2023

    2023

  21. [27]

    Fast reconstruction of hyperspectral image from its RGB counterpart using ADMM-Adam theory,

    C.-H. Lin, T.-H. Lin, T.-H. Lin, and T.-H. Lin, “Fast reconstruction of hyperspectral image from its RGB counterpart using ADMM-Adam theory,” in Proc. Workshop on Hyperspectral Imaging and Signal Pro- cessing: Evolution in Remote Sensing , Rome, Italy, 13–16 Sep. 2022, pp. 1–5

    2022

  22. [30]

    ADMM-ADAM: A new inverse imaging framework blending the advantages of convex optimization and deep learning,

    C.-H. Lin, Y .-C. Lin, and P.-W. Tang, “ADMM-ADAM: A new inverse imaging framework blending the advantages of convex optimization and deep learning,” IEEE Transactions on Geoscience and Remote Sensinging, vol. 60, pp. 1–16, Sep. 2021

    2021

  23. [31]

    Inverse problem transform: Solving hy- perspectral inpainting via deterministic compressed sensing,

    C.-H. Lin and P.-W. Tang, “Inverse problem transform: Solving hy- perspectral inpainting via deterministic compressed sensing,” in Proc. Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing , Amsterdam, Netherlands, 24–26 Mar. 2021, pp. 1–5

    2021

  24. [32]

    Transformer-driven inverse problem transform for fast blind hyperspectral image dehazing,

    P.-W. Tang, C.-H. Lin, and Y . Liu, “Transformer-driven inverse problem transform for fast blind hyperspectral image dehazing,” IEEE Transac- tions on Geoscience and Remote Sensing , vol. 62, pp. 1–14, Jan. 2024

    2024

  25. [33]

    A convex optimization- based coupled nonnegative matrix factorization algorithm for hyperspec- tral and multispectral data fusion,

    C.-H. Lin, F. Ma, C.-Y . Chi, and C.-H. Hsieh, “A convex optimization- based coupled nonnegative matrix factorization algorithm for hyperspec- tral and multispectral data fusion,” IEEE Transactions on Geoscience and Remote Sensing , vol. 56, no. 3, pp. 1652–1667, Nov. 2017

    2017

  26. [34]

    An explicit and scene-adapted definition of convex self-similarity prior with application to unsupervised Sentinel-2 super-resolution,

    C.-H. Lin and J. M. Bioucas-Dias, “An explicit and scene-adapted definition of convex self-similarity prior with application to unsupervised Sentinel-2 super-resolution,” IEEE Transactions on Geoscience and Remote Sensinging , vol. 58, no. 5, pp. 3352–3365, Dec. 2019

    2019

  27. [35]

    HyperKING: Quantum-classical generative adversarial networks for hyperspectral image restoration,

    C.-H. Lin and S.-S. Young, “HyperKING: Quantum-classical generative adversarial networks for hyperspectral image restoration,” IEEE Trans- actions on Geoscience and Remote Sensing , vol. 63, pp. 1–19, Apr. 2025

    2025

  28. [36]

    Sentinel-2 Google Earth Engine,

    “Sentinel-2 Google Earth Engine,” [Online]. Available: https://developers.google.com/earth-engine/datasets/catalog/ COPERNICUS S2 SR HARMONIZED

  29. [37]

    Hyperspectral change detection using semi- supervised graph neural network and convex deep learning,

    T.-H. Lin and C.-H. Lin, “Hyperspectral change detection using semi- supervised graph neural network and convex deep learning,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–18, Jun. 2023

    2023

  30. [38]

    Chi, W.-C

    C.-Y . Chi, W.-C. Li, and C.-H. Lin, Convex Optimization for Signal Processing and Communications: From Fundamentals to Applications . Boca Raton, FL, USA: CRC Press, 2017

    2017

  31. [39]

    CODE-IF: A convex/deep image fusion algorithm for efficient hyperspectral super-resolution,

    C.-H. Lin, C.-Y . Hsieh, and J.-T. Lin, “CODE-IF: A convex/deep image fusion algorithm for efficient hyperspectral super-resolution,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–18, Apr. 2024

    2024

  32. [40]

    Low- rank and sparse representation for hyperspectral image processing: A review,

    J. Peng, W. Sun, H.-C. Li, W. Li, X. Meng, C. Ge, and Q. Du, “Low- rank and sparse representation for hyperspectral image processing: A review,”IEEE Geoscience and Remote Sensing Magazine , vol. 10, no. 1, pp. 10–43, Jun. 2021

    2021

  33. [41]

    Denoising of hyperspectral images using nonconvex low rank matrix approximation,

    Y . Chen, Y . Guo, Y . Wang, D. Wang, C. Peng, and G. He, “Denoising of hyperspectral images using nonconvex low rank matrix approximation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 55, no. 9, pp. 5366–5380, Jun. 2017

    2017

  34. [42]

    Signal subspace identification for incom- plete hyperspectral image with applications to various inverse problems,

    C.-H. Lin and S.-S. Young, “Signal subspace identification for incom- plete hyperspectral image with applications to various inverse problems,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1– 16, Mar. 2024

    2024

  35. [43]

    A fast hyperplane- based minimum-volume enclosing simplex algorithm for blind hyper- spectral unmixing,

    C.-H. Lin, C.-Y . Chi, Y .-H. Wang, and T.-H. Chan, “A fast hyperplane- based minimum-volume enclosing simplex algorithm for blind hyper- spectral unmixing,” IEEE Transactions on Signal Processing , vol. 64, no. 8, pp. 1946–1961, Dec. 2015

    1946

  36. [44]

    Simultaneously sparse and low-rank abundance matrix estimation for hyperspectral image unmixing,

    P. V . Giampouras, K. E. Themelis, A. A. Rontogiannis, and K. D. Koutroumbas, “Simultaneously sparse and low-rank abundance matrix estimation for hyperspectral image unmixing,” IEEE Transactions on Geoscience and Remote Sensing , vol. 54, no. 8, pp. 4775–4789, Apr. 2016. 14

    2016

  37. [45]

    Minimum-volume transforms for remotely sensed data,

    M. D. Craig, “Minimum-volume transforms for remotely sensed data,” IEEE Transactions on Geoscience and Remote Sensing , vol. 32, no. 3, pp. 542–552, Aug. 2002

    2002

  38. [46]

    Hyperspectral unmixing based on mixtures of Dirichlet components,

    J. M. Nascimento and J. M. Bioucas-Dias, “Hyperspectral unmixing based on mixtures of Dirichlet components,” IEEE Transactions on Geoscience and Remote Sensing , vol. 50, no. 3, pp. 863–878, Sep. 2011

    2011

  39. [47]

    On the end- member identifiability of Craig’s criterion for hyperspectral unmixing: A statistical analysis for three-source case,

    C.-H. Lin, A. Ambikapathi, W.-C. Li, and C.-Y . Chi, “On the end- member identifiability of Craig’s criterion for hyperspectral unmixing: A statistical analysis for three-source case,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing , Vancouve...

    2013

  40. [48]

    Regularization parameter selection in minimum volume hyperspectral unmixing,

    L. Zhuang, C.-H. Lin, M. A. T. Figueiredo, and J. M. Bioucas-Dias, “Regularization parameter selection in minimum volume hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing , vol. 57, no. 12, pp. 9858–9877, Aug. 2019

    2019

  41. [49]

    Sparse unmixing of hyperspectral data,

    M.-D. Iordache, J. M. Bioucas-Dias, and A. Plaza, “Sparse unmixing of hyperspectral data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 6, pp. 2014–2039, Jan. 2011

    2014

  42. [50]

    Super-resolution image re- construction: A technical overview,

    S. C. Park, M. K. Park, and M. G. Kang, “Super-resolution image re- construction: A technical overview,” IEEE Signal Processing Magazine , vol. 20, no. 3, pp. 21–36, May 2003

    2003

  43. [51]

    Learning the parts of objects by non- negative matrix factorization,

    D. D. Lee and H. S. Seung, “Learning the parts of objects by non- negative matrix factorization,” Nature, vol. 401, no. 6755, pp. 788–791, Oct. 1999

    1999

  44. [52]

    Deep image prior,

    D. Ulyanov, A. Vedaldi, and V . Lempitsky, “Deep image prior,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition , Salt Lake City, UT, USA, 18–23 Jun. 2018, pp. 9446–9454

    2018

  45. [53]

    Proximal algorithms,

    N. Parikh and S. Boyd, “Proximal algorithms,” F oundations and Trends® in Optimization , vol. 1, no. 3, pp. 127–239, Jan. 2014

    2014

  46. [54]

    Plug- and-play image restoration with deep denoiser prior,

    K. Zhang, Y . Li, W. Zuo, L. Zhang, L. Van-Gool, and R. Timofte, “Plug- and-play image restoration with deep denoiser prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 44, no. 10, pp. 6360– 6376, Jun. 2021

    2021

  47. [55]

    Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising,

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, “Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising,” IEEE Transactions on Image Processing , vol. 26, no. 7, pp. 3142–3155, Feb. 2017

    2017

  48. [56]

    Ridge regression,

    G. C. McDonald, “Ridge regression,” Wiley Interdisciplinary Reviews: Computational Statistics , vol. 1, no. 1, pp. 93–100, Jul. 2009

    2009

  49. [57]

    DCSN: Deep com- pressed sensinging network for efficient hyperspectral data transmission of miniaturized satellite,

    C.-C. Hsu, C.-H. Lin, C.-H. Kao, and Y .-C. Lin, “DCSN: Deep com- pressed sensinging network for efficient hyperspectral data transmission of miniaturized satellite,” IEEE Transactions on Geoscience and Remote Sensinging, vol. 59, no. 9, pp. 7773–7789, 2021

    2021

  50. [58]

    ENVI 5.2,

    “ENVI 5.2,” [Online]. Available: https://www.nv5geospatialsoftware. com/

  51. [59]

    Sedgewick and K

    R. Sedgewick and K. Wayne, Algorithms. Boston, MA, USA: Addison- Wesley Professional, 2011

    2011

  52. [60]

    NTIRE 2022 spectral recovery challenge and data set,

    B. Arad et al. , “NTIRE 2022 spectral recovery challenge and data set,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition , New Orleans, LA, USA, 19-20 Jun. 2022, pp. 863–881

    2022

  53. [61]

    A practical algorithm for topic modeling with provable guarantees,

    S. Arora, R. Ge, Y . Halpern, D. Mimno, A. Moitra, D. Sontag, Y . Wu, and M. Zhu, “A practical algorithm for topic modeling with provable guarantees,” arXiv preprint arXiv:1212.4777 , Dec. 2012. [Online]. Available: https://arxiv.org/abs/1212.4777

    2012 arXiv

  54. [62]

    Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion,

    N. Yokoya, T. Yairi, and A. Iwasaki, “Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion,” IEEE Transactions on Geoscience and Remote Sensing , vol. 50, no. 2, pp. 528–537, Aug. 2011. Chia-Hsiang Lin (S’10-M’18-SM’24) received the...

    2011

Pith tools

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