Pith. sign in

REVIEW 4 major objections 5 minor 53 references

Multitemporal Latent Dynamical Framework for Hyperspectral Images Unmixing

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A latent neural ODE models how material abundances evolve across multitemporal hyperspectral images, with convergence and stability claims.

desk verdict A creative neural-ODE model for multitemporal unmixing whose advertised theoretical guarantees do not survive careful reading; empirical gains are mixed. read the letter →

arxiv 2505.20902 v1 pith:5AR3HC4N submitted 2025-05-27 eess.IV cs.CV

Ruiying Li , Bin Pan , Lan Ma , Xia Xu , Zhenwei Shi This is my paper
classification eess.IVcs.CV
keywords multitemporalhyperspectralunmixingneuralordinarydifferentialequationslatentdynamicsdynamicaldiscretizationabundanceendmembervariabilityconsistencyconvergencestability
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 argues that multitemporal hyperspectral unmixing should treat the evolution of material abundances as a continuous dynamical process rather than a sequence of independent per-image estimates. It proposes MiLD, a framework in which abundances are latent variables whose time derivative is given by a neural ordinary differential equation, while endmembers vary through a perturbed linear mixing model. The paper converts that ODE into a computable recurrence with a multistep, bidirectional discretization, then states consistency, convergence, and stability theorems for it. A sympathetic reading is that, if the theorems hold, MiLD gives a principled way to estimate time-varying abundances from image sequences and justifies the network architecture as a numerical scheme.

What carries the argument

The load-bearing object is the latent ODE for abundances, $\frac{dA_t}{dt}=G_1(Y_t)+G_2(A_t)$, in which $G_1$ lets the observed image drive the hidden state and $G_2$ describes autonomous material evolution. The companion piece is the dynamical discretization: a trapezoidal-rule and Taylor-expansion derivation that turns the ODE into the recurrence $A_t = F_{t-K}(z_{t-K}) + \cdots + z_t$, averaged over $K$ steps and over left-to-right and right-to-left directions. The consistency, convergence, and stability proofs all hang on this discretization relating the continuous model to the discrete image sequence.

What would settle it

On the synthetic datasets with known abundances, record both the intermediate pseudo-abundance $z_t$ from the per-time encoders and the final estimate $A_t$. If $\max_t \|A_t - z_t\|$ stays large while the loss is low, or if replacing the neighboring pseudo-abundances $z_{t'}$ ($t'\neq t$) with random values while holding $Y_t$ fixed does not change the output $A_t$, then the approximation that carries the convergence proof is not load-bearing for the reported results.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that multitemporal unmixing can be formulated as a latent initial-value problem: $\frac{dA_t}{dt}=G_1(Y_t)+G_2(A_t)$, where $A_t$ are the true abundances, $Y_t$ the observed images, and $G_1,G_2$ learned maps. To make this computable, MiLD first obtains pseudo-abundances $z_t$ from per-time autoencoders, then builds the abundance at time $t$ from a multistep and bidirectional average of contributions $F_{t-k}(z_{t-k}) + z_t$ over $k=1,\dots,K$, with neural networks $F$ parameterizing the discretized dynamics. The authors claim the discretization is consistent with the ODE, that $z_t$ converges to $A_t$ as the step size $\Delta\to 0$, and that the latent dynamical system is stable to perturbations in the initial estimates. These three properties are presented as the theoretical support that earlier ODE-based unmixing approaches lacked.

Load-bearing premise

The load-bearing bet is that the independent per-time estimates $z_t$ are already close enough to the true abundances $A_t$ that the dynamical correction can be treated as a small perturbation, even though the deployed step size is not the small-$\Delta$ regime the proof covers.

Editorial extensions

If this is right

  • Irregularly spaced acquisitions can be handled by choosing the number of steps $K$ in the multistep discretization, preserving the flexibility of latent ODEs for non-uniform time gaps.
  • Each abundance map is estimated using both past and future frames through the bidirectional average, so any time step can be corrected by information from the whole sequence.
  • Under the stated theorems, the recurrence is a numerically faithful scheme for the latent ODE in the small-step limit, giving a principled justification for the neural architecture.
  • Because the system is stable, poor initial unmixing at one time should not amplify into later time estimates during joint training.

Reading between the lines

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

  • A direct check of the theory would quantify $\|z_t - A_t\|$ on synthetic data under the deployed hyperparameters; the convergence theorem only controls this gap as $\Delta\to 0$.
  • The local-in-time recurrence suggests an online variant in which each new acquisition updates the latent dynamics without reprocessing the full sequence, a natural next step the paper does not discuss.
  • The same latent-ODE-plus-discretization recipe could be transferred to other per-pixel time-series inverse problems such as dynamic image fusion or change detection, where abundance-like hidden states evolve continuously.
  • Bidirectional averaging implies the model treats time as reversible; testing whether reversing the image sequence inverts the abundance dynamics would reveal whether the two directions are genuinely complementary.
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

4 major / 5 minor

Summary. The paper proposes MiLD, a multitemporal hyperspectral unmixing framework that models abundance evolution as a neural ODE in latent space, discretizes the ODE via a trapezoidal rule with multistep and bidirectional extensions, and estimates endmembers under the perturbed linear mixing model. The authors claim four contributions: a problem definition via ODEs, a dynamical discretization model, a solution algorithm, and theoretical validation through consistency, convergence, and stability theorems. Experiments are reported on one real dataset (Lake Tahoe) and two synthetic datasets, with comparisons against six baselines. The central advertised strength is the theoretical support for consistency, convergence, and stability of the proposed discretization and algorithm.

Significance. If the theoretical claims were sound, the paper would offer a novel and rare contribution: a neural-ODE-based multitemporal unmixing framework with formal guarantees on its discretization and dynamics. The attempt to provide consistency, convergence, and stability theorems for a hyperspectral unmixing method is commendable and could be of interest to the community. However, the significance is undermined by a circular convergence proof, a stability result that does not address the original ODE, and empirical results that are mixed (the method loses to MUFormer on the second synthetic dataset). The paper also ships no code or machine-checked proofs, so the advertised theoretical validation cannot be independently verified. As it stands, the central claim of theoretical support is not established.

major comments (4)
  1. [Section III-B and Section III-E (Theorem 2)] The derivation of Eq. (20) from Eq. (19) explicitly invokes condition (ii), which states that the sequence {z_t} converges to {A_t} at small Δ, with the text saying "This convergence will be proved in Section III-D." The proof of Theorem 2 then uses Eq. (20) (or Eq. (21)), which already contains the replacement of A_{t'} by z_{t'} for t' ≠ t. This is circular: the convergence result is assumed in the recurrence used to prove it. Moreover, the unsimplified Eq. (19) implies A_t - z_t = A_{t-Δ} - z_{t-Δ} + O(Δ), so by induction |A_t - z_t| ≤ |A_0 - z_0| + O(Δ); the Δ→0 limit vanishes only if the initial encoder error |A_0 - z_0| is zero, which is not shown. Thus Theorem 2 is not established.
  2. [Section III-E (Theorem 3, Stability)] The stability bound in Eq. (44) is derived for the discretized recurrence obtained under the same approximation z ≈ A used in Eq. (20), so it does not establish stability of the original ODE system in Eq. (11). The definition of stability requires continuous dependence on initial conditions over the time horizon, but the proof provides a bound (L2Δ + 1) that grows with Δ and is not uniformly bounded as T increases. For the practical choice Δ = 1, the factor L2 + 1 is not a contraction and gives no long-time stability guarantee. The theorem, as stated, does not support the claim that MiLD is a stable dynamics system.
  3. [Section IV-C (Table I)] The quantitative results on Synthetic data2 show that the proposed method has NRMSEA = 0.307, which is substantially worse than MUFormer (0.185) and ReSUDNN (0.294). The paper acknowledges this but still concludes that the experiments "validated the utility" of the work. No confidence intervals, standard deviations, or repeated-run statistics are reported for any metric, so the reported improvements on Synthetic data1 could be within run-to-run variability. The empirical evidence is therefore insufficient to support the claim of consistent advantage over existing methods.
  4. [Section IV-A and Section III-E] The theoretical results (consistency, convergence, stability) are all asymptotic in the limit Δ → 0, with Assumption 1 requiring lim_{Δ→0} G2(A_t) = 0. However, the experiments set Δ = 1 and K = 2 (Section IV-A). The paper provides no argument that the practical configuration lies within the regime where the asymptotic guarantees apply, nor does it characterize the error at Δ = 1. Consequently, the theoretical support does not cover the algorithm actually evaluated.
minor comments (5)
  1. [Throughout the text] Many equations are missing or garbled in the manuscript (e.g., Eq. (8), Eq. (11), Eq. (12), Eq. (14), and several intermediate expressions are blank or corrupted), making the derivation difficult or impossible to follow.
  2. [Section II-A] The text says "H, W and L represents height, weight and number of channels"; "weight" should be "width."
  3. [Section III-E, Proof of Theorem 2] The proof of Theorem 2 contains a corrupted passage ("e ∆ t e - a e 0 0．d．s．∆．e ．b | t 1 T i h n i i hypothesis") that prevents any rigorous reading; this must be rewritten.
  4. [Table I] The table caption reads "QUALIATIVE RESULTS" instead of "QUALITATIVE RESULTS," and the table reports only point estimates with no indication of variability.
  5. [Section V-B] The real-data experiment on Lake Tahoe is evaluated only qualitatively with a textual claim about the upper-left corner; no quantitative metric or statistical test supports this claim.

Circularity Check

2 steps flagged · score 6.0 of 10

Theorem 2's convergence proof assumes the conclusion: Section III-B replaces A with z under a convergence condition that the theorem then 'proves' by reusing that simplified recurrence.

  1. self definitional [Section III-B (Dynamical Discretization), simplification between Eqs. (19) and (20)]
    "For simplifying this equation, we consider two conditions. (i) As mentioned in Section III-A, ≈t can be viewed as an approximation of At without feature fusion from other phrases. (ii) Sequence {≈t } is convergent to sequence {At } at a small ∆. This convergence will be proved in Section III-D. Based on these conditions, when calculating At, we approximate other phrase A(1:T)\t by ≈ (1:T)\t. Under the approximation, eq.(19) equals to:"

    Eq. (19) is A_t − z_t = A_{t−Δ} − z_{t−Δ} plus higher-order terms. To obtain Eq. (20), and hence the algorithm's core recurrence A_t = F_{t−Δ}(z_{t−Δ}) + z_t (Eq. (21)), the paper replaces A_{t−Δ} with z_{t−Δ} under condition (ii) — the statement that {z_t} converges to {A_t}, which is exactly what Theorem 2 claims to prove. The convergence result is thus assumed inside the model definition before it is demonstrated. If one instead propagates the unsimplified Eq. (19), the bound is |A_t − z_t| ≤ |A_0 − z_0| + O(Δ), which vanishes only when the initial encoder error is zero, a fact never established.

  2. self definitional [Section III-E, Theorem 2 (Convergence) proof]
    "Finally, we demonstrate that lim∆→0 |At − ≈t | = 0 by using the inductive hypothesis and deriving a relationship between ≈ t and At as mentioned in eq.(17)."

    The 'relationship between z_t and A_t' invoked here is the recurrence built in Section III-B, which was obtained under condition (ii) by replacing other-time-step abundances A with z. The inductive step therefore carries the theorem's conclusion into its own hypothesis, and the base case (|A_0 − z_0| → 0) is never shown. Moreover, the target A_t is not external ground truth but the solution of the model's own ODE (Eq. (11); 'Solving Eq.(11) is equivalent to addressing MTHU problem'), so Theorem 2 certifies internal self-consistency rather than agreement with data. Theorem 3 likewise bounds deviations of the same approximating recurrence (Eq. (44)), so stability does not restore the link to the original ODE.

full rationale

The paper's empirical core is not circular: MiLD is trained and evaluated against genuine ground-truth abundances on two synthetic datasets (Table I), with NRMSE_A 0.243 on Synthetic data 1 and 0.307 on Synthetic data 2, and the consistency theorem (Theorem 1) is a standard local-truncation-error calculation. There is no load-bearing self-citation chain: Lemma 1 restates the standard universal approximation property without attribution and Definition 3 cites the external reference [53]. However, the advertised theoretical support is undermined by a genuinely circular convergence argument. Section III-B derives the core recurrence A_t = F_{t−Δ}(z_{t−Δ}) + z_t (Eq. (21)) from Eq. (19) only after substituting z for A at other time steps, justified by condition (ii): 'Sequence {z_t} is convergent to sequence {A_t} at a small Δ. This convergence will be proved in Section III-D.' Theorem 2 then proves lim|A_t − z_t| = 0 by using the relation derived from that same already-simplified recurrence together with an inductive hypothesis, so the proof presupposes the convergence it announces. Compounding the issue, the 'true solution' A_t is defined as the solution of the model's own ODE (Eq. (11); 'Solving Eq.(11) is equivalent to addressing MTHU problem'), so convergence is to an internal construct, and the stability theorem (Theorem 3, Eq. (44)) bounds deviations of that same approximating recurrence, not the original ODE. Finally, the theory is asymptotic (Δ → 0) while the deployed configuration uses unit-spaced time indices with K = 2 (Section IV-A), so the guarantees do not cover the experimental regime. Net: one central theorem reduces by construction, giving partial circularity, while the framework retains independent empirical content that justifies a score of 6 rather than higher.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

MiLD introduces no new physical entities; the latent dynamics layer is a model component, not an invented physical entity. The assumptions and free parameters listed above are what the central claim rests on.

free parameters (3)
  • time step Δ = 1 (image index steps)
    The discretization theorems require Δ→0, but the algorithm uses adjacent time indices, so Δ=1 or K-step averages; the small-Δ theory does not apply.
  • number of steps K = 2
    Hyperparameter in Eq. (26) set to 2 for experiments; affects how many temporal neighbors are used.
  • loss weights α and β = not reported in paper
    Reconstruction and endmember variability trade-off weights in Eq. (29); values chosen by hand and not reported.
assumptions (4)
  • domain assumption Linear Mixing Model Y_t = A_t E_t + N_t
    Standard model in spectroscopy; the paper uses it to define the problem and the autoencoder.
  • ad hoc to paper Assumption 1: Continuity of material change, lim_{Δ→0} G2(A_t) = 0
    The limit notation is ill-posed since G2 does not depend on Δ, and the assumption is used to conclude convergence in the limit but is not verified empirically.
  • standard math Assumption 2: Lipschitz conditions on G1 and G2
    Standard for ODE numerical methods; however, no evidence is given that trained neural networks satisfy the condition with the required constants.
  • standard math Universal approximation theorem (Lemma 1)
    Invoked to justify neural approximations of the feature maps; standard but does not provide error bounds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multitemporal Latent Dynamical Framework for Hyperspectral Images Unmixing." pith.science (2026). https://pith.science/paper/5AR3HC4N

@misc{pith2026250520902,
  author       = {Pith},
  title        = {Pith review of: Multitemporal Latent Dynamical Framework for Hyperspectral Images Unmixing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5AR3HC4N}},
  note         = {Machine review of arXiv:2505.20902}
}
read the original abstract

Multitemporal hyperspectral unmixing can capture dynamical evolution of materials. Despite its capability, current methods emphasize variability of endmembers while neglecting dynamics of abundances, which motivates our adoption of neural ordinary differential equations to model abundances temporally. However, this motivation is hindered by two challenges: the inherent complexity in defining, modeling and solving problem, and the absence of theoretical support. To address above challenges, in this paper, we propose a multitemporal latent dynamical (MiLD) unmixing framework by capturing dynamical evolution of materials with theoretical validation. For addressing multitemporal hyperspectral unmixing, MiLD consists of problem definition, mathematical modeling, solution algorithm and theoretical support. We formulate multitemporal unmixing problem definition by conducting ordinary differential equations and developing latent variables. We transfer multitemporal unmixing to mathematical model by dynamical discretization approaches, which describe the discreteness of observed sequence images with mathematical expansions. We propose algorithm to solve problem and capture dynamics of materials, which approximates abundance evolution by neural networks. Furthermore, we provide theoretical support by validating the crucial properties, which verifies consistency, convergence and stability theorems. The major contributions of MiLD include defining problem by ordinary differential equations, modeling problem by dynamical discretization approach, solving problem by multitemporal unmixing algorithm, and presenting theoretical support. Our experiments on both synthetic and real datasets have validated the utility of our work

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 52 canonical work pages

  1. [1]

    Spectral-spatial latent reconstruction for open-set hyperspectral image classification,

    J. Yue, L. Fang, and M. He, “Spectral-spatial latent reconstruction for open-set hyperspectral image classification,” IEEE Transactions on Image Processing, vol. 31, pp. 5227–5241, 2022

    2022

  2. [2]

    Dual-stage approach toward hyperspectral image super-resolution,

    Q. Li, Y. Yuan, X. Jia, and Q. Wang, “Dual-stage approach toward hyperspectral image super-resolution,” IEEE Transactions on Image Processing, vol. 31, pp. 7252–7263, 2022

    work page 2022

  3. [3]

    Binary change guided hy- perspectral multiclass change detection,

    M. Hu, C. Wu, B. Du, and L. Zhang, “Binary change guided hy- perspectral multiclass change detection,” IEEE Transactions on Image Processing, vol. 32, pp. 791–806, 2023

    work page 2023

  4. [4]

    Few-shot learning with mutual information enhancement for hyperspectral image classification,

    Q. Zhang, J. Peng, W. Sun, and Q. Liu, “Few-shot learning with mutual information enhancement for hyperspectral image classification,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14,2024

    work page 2024

  5. [5]

    Rotation-invariant attention network for hyperspectral image classification,

    X. Zheng, H. Sun, X. Lu, and W. Xie, “Rotation-invariant attention network for hyperspectral image classification,” IEEE Transactions on Image Processing, vol. 31, pp. 4251–4265, 2022

    work page 2022

  6. [6]

    Target-constrained interference-minimized band selection for hyper- spectraltarget detection,

    X. Shang, M. Song, Y. Wang, C. Yu, H. Yu, F. Li, and C.-I. Chang, “Target-constrained interference-minimized band selection for hyper- spectraltarget detection,” IEEE Transactions on Geoscienceand Remote Sensing, vol. 59, no. 7, pp. 6044–6064, 2021

    work page 2021

  7. [7]

    Cascade superpixel regularized gabor feature fusion for hyperspectral image classification,

    S. Jia, Z. Lin, B. Deng, J. Zhu, and Q. Li, “Cascade superpixel regularized gabor feature fusion for hyperspectral image classification,” IEEETransactions on Neural Networks andLearning Systems, vol. 31, no. 5, pp. 1638–1652, 2020

    work page 2020

  8. [8]

    Fusing hyperspectral and multispectral images via coupled sparse tensor factorization,

    S. Li, R. Dian, L. Fang, and J. M. Bioucas-Dias, “Fusing hyperspectral and multispectral images via coupled sparse tensor factorization,” IEEE Transactions on Image Processing, vol. 27, no. 8, pp. 4118– 4130, 2018

    work page 2018

Show all 53 references
  1. [9]

    Coded hyperspectral image reconstruction using deep external and internal learning,

    Y. Fu, T. Zhang, L. Wang, and H. Huang, “Coded hyperspectral image reconstruction using deep external and internal learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44,no. 7, pp. 3404–3420, 2022

    2022

  2. [10]

    Fully tensorized lightweight convlstm neural networks for hyperspectral image classifi- cation,

    T.-Y. Ma, H.-C. Li, Y.-B. Zheng, Q. Du, and A. Plaza, “Fully tensorized lightweight convlstm neural networks for hyperspectral image classifi- cation,” IEEE Transactions onNeuralNetworks andLearning Systems,pp. 1–15, 2024

    2024

  3. [11]

    Vertex component analysis: a fast algorithm to unmix hyperspectral data,

    J. Nascimento and J. Dias, “Vertex component analysis: a fast algorithm to unmix hyperspectral data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 43, no. 4, pp. 898–910, 2005

    2005

  4. [12]

    Hyperspectral blind unmixing using a double deep image prior,

    C. Zhou and M. R. D. Rodrigues, “Hyperspectral blind unmixing using a double deep image prior,” IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 11, pp. 16478–16492, 2024

    2024

  5. [13]

    Hyperspectral unmixing overview: Geometrical, statistical, and sparse regression-based approaches,

    J. M. Bioucas-Dias, A. Plaza, N. Dobigeon, M. Parente, Q. Du, P. Gader, and J. Chanussot, “Hyperspectral unmixing overview: Geometrical, statistical, and sparse regression-based approaches,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,vol. ...

    2012

  6. [14]

    An abundance-guided attention network for hyperspectral unmixing,

    X. Tao, M. E. Paoletti, Z. Wu, J. M. Haut, P. Ren, and A. Plaza, “An abundance-guided attention network for hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14,2024

    2024

  7. [15]

    Nonlinear unmixing of hyperspectral images using a generalized bilinear model,

    A. Halimi, Y. Altmann, N. Dobigeon, and J.-Y. Tourneret, “Nonlinear unmixing of hyperspectral images using a generalized bilinear model,” IEEETransactionson GeoscienceandRemoteSensing, vol. 49, no. 11, pp. 4153–4162, 2011

    2011

  8. [16]

    Nonlinear spectral unmixing using b zier surfaces,

    B. Koirala, B. Rasti, Z. Bnoulkacem, and P. Scheunders, “Nonlinear spectral unmixing using b zier surfaces,” IEEE Transactions on Geo - science andRemote Sensing, vol. 62, pp. 1–16, 2024

    2024

  9. [17]

    Blind spectral unmixing based on sparse nonnegative matrix factorization,

    Z. Yang, G. Zhou, S. Xie, S. Ding, J.-M. Yang, and J. Zhang, “Blind spectral unmixing based on sparse nonnegative matrix factorization,” IEEE Transactions on Image Processing, vol. 20, no. 4, pp. 1112–1125, 2011

    2011

  10. [18]

    A plug-and-play priors framework for hyperspectral unmixing,

    M. Zhao, X. Wang, J. Chen, and W. Chen, “A plug-and-play priors framework for hyperspectral unmixing,” IEEE Transactions on Geo - scienceand RemoteSensing, vol. 60, pp. 1–13, 2022

    2022

  11. [19]

    Simultaneously multiob- jective sparse unmixing and library pruning for hyperspectral imagery,

    X. Xu, B. Pan, Z. Chen, Z. Shi, and T. Li, “Simultaneously multiob- jective sparse unmixing and library pruning for hyperspectral imagery,” IEEE Transactions on Geoscience andRemote Sensing, vol. 59, no. 4, pp. 3383–3395, 2021

    2021

  12. [20]

    A novel NMF guided for hyper- U A IEEE TRANSACTIONS ON IMAGE PROCESSING 11 spectral unmixing from incomplete and noisy data,

    L. Dong, X. Lu, G. Liu, and Y. Yuan, “A novel NMF guided for hyper- U A IEEE TRANSACTIONS ON IMAGE PROCESSING 11 spectral unmixing from incomplete and noisy data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–15, 2022

    2022

  13. [21]

    Superpixel-based reweighted low-rank and total variation sparse unmixing for hyperspec- tral remote sensing imagery,

    H. Li, R. Feng, L. Wang, Y. Zhong, and L. Zhang, “Superpixel-based reweighted low-rank and total variation sparse unmixing for hyperspec- tral remote sensing imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 1, pp. 629–647, 2021

    2021

  14. [22]

    Spectral variability aware blind hyperspectral image unmixing based on convex geometry,

    L. Drumetz, J. Chanussot, C. Jutten, W.-K. Ma, and A. Iwasaki, “Spectral variability aware blind hyperspectral image unmixing based on convex geometry,” IEEE Transactions on Image Processing, vol. 29,pp. 4568–4582, 2020

    2020

  15. [23]

    MAHUM: A multitasks autoencoder hyperspectral unmixing model,

    J. Chen, P. Gamba, and J. Li, “MAHUM: A multitasks autoencoder hyperspectral unmixing model,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–16, 2023

    2023

  16. [24]

    Dynamical spectral unmixing of multitemporal hyperspectral images,

    S. Henrot, J. Chanussot, and C. Jutten, “Dynamical spectral unmixing of multitemporal hyperspectral images,” IEEE Transactions on Image Processing, vol. 25, no. 7, pp. 3219–3232, 2016

    2016

  17. [25]

    Sparse distributed multitemporal hyperspectral unmixing,

    J. Sigurdsson, M. O. Ulfarsson, J. R. Sveinsson, and J. M. Bioucas- Dias, “Sparse distributed multitemporal hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 11, pp.6069–6084, 2017

    2017

  18. [26]

    A dynamic unmixing framework for plant production system monitoring,

    M.-D. Iordache, L. Tits, J. M. Bioucas-Dias, A. Plaza, and B. Somers, “A dynamic unmixing framework for plant production system monitoring,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 7, no. 6, pp. 2016–2034, 2014

    2016

  19. [27]

    Multisensor coupled spectral unmixing for time-series analysis,

    N. Yokoya, X. X. Zhu, and A. Plaza, “Multisensor coupled spectral unmixing for time-series analysis,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 5, pp. 2842–2857, 2017

    2017

  20. [28]

    Monitoring tropical forest degradation using spectral unmixing and landsat time series analysis,

    E. L. Bullock, C. E. Woodcock, and P. Olofsson, “Monitoring tropical forest degradation using spectral unmixing and landsat time series analysis,” Remote Sensing of Environment, vol. 238, p. 110968, 2020

    2020

  21. [29]

    Online unmixing of multitemporal hyperspectral images accounting for spectral variability,

    P.-A. Thouvenin, N. Dobigeon, and J.-Y. Tourneret, “Online unmixing of multitemporal hyperspectral images accounting for spectral variability,” IEEE Transactions on Image Processing, vol. 25, no. 9, pp. 3979–3990, 2016

    2016

  22. [30]

    Learning end- member dynamics in multitemporal hyperspectral data using a state- space model formulation,

    L. Drumetz, M. D. Mura, G. Tochon, and R. Fablet, “Learning end- member dynamics in multitemporal hyperspectral data using a state- space model formulation,” in ICASSP 2020 - 2020 IEEE International ConferenceonAcoustics,SpeechandSignal Processing(ICASSP),2020,pp. 2483–2487

    2020

  23. [31]

    Kalman filtering and expectation maximization for multitemporal spec- tral unmixing,

    R. A. Borsoi, T. Imbiriba, P. Closas, J. C. M. Bermudez, and C. Richard, “Kalman filtering and expectation maximization for multitemporal spec- tral unmixing,” IEEE Geoscience and Remote SensingLetters, vol. 19,pp. 1–5, 2022

    2022

  24. [32]

    Bayesian unmixing of hyperspectral image sequence with composite priors for abundance and endmember variability,

    H. Liu, Y. Lu, Z. Wu, Q. Du, J. Chanussot, and Z. Wei, “Bayesian unmixing of hyperspectral image sequence with composite priors for abundance and endmember variability,” IEEE Transactions on Geo - science and RemoteSensing, vol. 60, pp. 1–15, 2022

    2022

  25. [33]

    Dynamical hyperspectral unmixing with variational recurrent neural networks,

    R. A. Borsoi, T. Imbiriba, and P. Closas, “Dynamical hyperspectral unmixing with variational recurrent neural networks,” IEEE Transactions on Image Processing, vol. 32, pp. 2279–2294, 2023

    2023

  26. [34]

    Spatio-temporal spec- tral unmixing of time-series images,

    Q. Wang, X. Ding, X. Tong, and P. M. Atkinson, “Spatio-temporal spec- tral unmixing of time-series images,” Remote Sensing of Environment,vol. 259, p. 112407, 2021

    2021

  27. [35]

    Transformer for multitemporal hyperspectral image unmixing,

    H. Li, Q. Dong, X. Xie, X. Xu, T. Li, and Z. Shi, “Transformer for multitemporal hyperspectral image unmixing,” 2024. [Online]. Available: https://arxiv.org/abs/2407.10427

    2024 arXiv

  28. [36]

    DAEN: Deep autoencoder networks for hyperspectral unmixing,

    Y. Su, J. Li, A. Plaza, A. Marinoni, P. Gamba, and S. Chakravortty, “DAEN: Deep autoencoder networks for hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 7, pp.4309–4321, 2019

    2019

  29. [37]

    Convolutional autoen- coder for spectral–spatial hyperspectral unmixing,

    B. Palsson, M. O. Ulfarsson, and J. R. Sveinsson, “Convolutional autoen- coder for spectral–spatial hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 1, pp. 535–549, 2021

    2021

  30. [38]

    Endmember-guided unmixing network (EGU-Net): A gen- eral deep learning framework for self-supervised hyperspectral unmix- ing,

    D. Hong, L. Gao, J. Yao, N. Yokoya, J. Chanussot, U. Heiden, and B. Zhang, “Endmember-guided unmixing network (EGU-Net): A gen- eral deep learning framework for self-supervised hyperspectral unmix- ing,” IEEE Transactions on Neural Networks and Learning Systems,vol. 33, no. 11...

    2022

  31. [39]

    Adversarial autoencoder network for hyperspectral unmixing,

    Q. Jin, Y. Ma, F. Fan, J. Huang, X. Mei, and J. Ma, “Adversarial autoencoder network for hyperspectral unmixing,” IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 8, pp. 4555–4569, 2023

    2023

  32. [40]

    uDAS: An untied denoising autoencoder with sparsity for spectral unmixing,

    Y. Qu and H. Qi, “uDAS: An untied denoising autoencoder with sparsity for spectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 3, pp. 1698–1712, 2019

    2019

  33. [41]

    EndNet: Sparse autoencoder network for endmember extraction and hyperspectral unmixing,

    S. Ozkan, B. Kaya, and G. B. Akar, “EndNet: Sparse autoencoder network for endmember extraction and hyperspectral unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 1, pp.482–496, 2019

    2019

  34. [42]

    CyCU- Net: Cycle-consistency unmixing network by learning cascaded autoen- coders,

    L. Gao, Z. Han, D. Hong, B. Zhang, and J. Chanussot, “CyCU- Net: Cycle-consistency unmixing network by learning cascaded autoen- coders,” IEEETransactions onGeoscienceandRemoteSensing,vol. 60,pp. 1–14, 2022

    2022

  35. [43]

    Neu- ral ordinary differential equations,

    R. T. Q. Chen, Y. Rubanova, J. Bettencourt, and D. K. Duvenaud, “Neu- ral ordinary differential equations,” in Advances in Neural Information Processing Systems, S. Bengio, H.Wallach, H. Larochelle, K. Grauman,N. Cesa-Bianchi, and R. Garnett, Eds., vol. 31. Curran Associates, ...

    2018

  36. [44]

    Latent ordinary differential equations for irregularly-sampled time series,

    Y. Rubanova, R. T. Q. Chen, and D. K. Duvenaud, “Latent ordinary differential equations for irregularly-sampled time series,” in Advances in Neural Information Processing Systems, H. Wallach, H. Larochelle,A. Beygelzimer, F. d'Alch -Buc, E. Fox, and R. Garnett, Eds., vol. 32. ...

    2019

  37. [45]

    Neural controlled differential equations for irregular time series,

    P. Kidger, J. Morrill, J. Foster, and T. Lyons, “Neural controlled differential equations for irregular time series,” in Advances in Neural Information Processing Systems, H. Larochelle,M. Ranzato, R. Hadsell,M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020,...

    2020

  38. [46]

    Neural Lad: Aneural latent dynamics framework for times series modeling,

    t. li, J. Li, and Z. Zhu, “Neural Lad: Aneural latent dynamics framework for times series modeling,” in Advances in Neural Information Process - ing Systems, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt,and S. Levine, Eds., vol. 36. Curran Associates, Inc., 2023, pp. 1...

    2023

  39. [47]

    Unsupervised unmixing of hyperspectral images accounting for endmember variability,

    A. Halimi, N. Dobigeon, and J.-Y. Tourneret, “Unsupervised unmixing of hyperspectral images accounting for endmember variability,” IEEE Transactions onImageProcessing,vol.24,no. 12,pp.4904– 4917,2015

    2015

  40. [48]

    Hyperspectral unmixing with spectral variability using a perturbed linear mixing model,

    P.-A. Thouvenin, N. Dobigeon, and J.-Y. Tourneret, “Hyperspectral unmixing with spectral variability using a perturbed linear mixing model,” IEEE Transactions on Signal Processing, vol. 64, no. 2, pp.525–538, 2016

    2016

  41. [49]

    Blind hyperspectral unmixing using an extended linear mixing model to address spectral variability,

    L. Drumetz, M.-A. Veganzones, S. Henrot, R. Phlypo, J. Chanussot, and C. Jutten, “Blind hyperspectral unmixing using an extended linear mixing model to address spectral variability,” IEEE Transactions on Image Processing, vol. 25, no. 8, pp. 3890–3905, 2016

    2016

  42. [50]

    An augmented linear mixing model to address spectral variability for hyperspectral unmixing,

    D. Hong, N. Yokoya, J. Chanussot, and X. X. Zhu, “An augmented linear mixing model to address spectral variability for hyperspectral unmixing,” IEEE Transactions on Image Processing, vol. 28, no. 4, pp. 1923–1938, 2019

    1923

  43. [51]

    Hyperspectral un- mixing with endmember variability via alternating angle minimization,

    R. Heylen, A. Zare, P. Gader, and P. Scheunders, “Hyperspectral un- mixing with endmember variability via alternating angle minimization,” IEEE Transactions on Geoscience andRemote Sensing, vol. 54, no. 8, pp. 4983–4993, 2016

    2016

  44. [52]

    A new admm- based hyperspectral unmixing algorithm associated with a linear mixing model addressing spectral variability with a multiplicative structure,

    F. Z. Benhalouche, M. S. Karoui, and Y. Deville, “A new admm- based hyperspectral unmixing algorithm associated with a linear mixing model addressing spectral variability with a multiplicative structure,” in IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing S...

    2024

  45. [53]

    On the convergence, consistency and stability of a one-step method for numerical integration of ordinary differential equation,

    E. Ibijola and P. Kama, “On the convergence, consistency and stability of a one-step method for numerical integration of ordinary differential equation,” International Journal of Computer Mathematics, vol. 73,no. 2, pp. 261–277, 1999

    1999

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.