Pith. sign in

REVIEW 3 major objections 6 minor 60 references

Multisource Collaborative Domain Generalization for Cross-Scene Remote Sensing Image Classification

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read This paper proposes MS-CDG, a multi-source domain generalization framework that combines data-aware adversarial augmentation with class-wise prototype and kernel-mixture diversification, and reports the best cross-scene remote sensing…

desk verdict Reported DG gains are likely inflated by tuning hyperparameters on the target test sets; the framework itself is coherent and worth engaging after fixing the evaluation protocol. read the letter →

arxiv 2412.03897 v1 pith:TJ6XKV3Y submitted 2024-12-05 cs.CV cs.LG

classification cs.CVcs.LG
keywords ImageClassificationdomaingeneralizationcross-sceneremotesensingmulti-sourcedataadversarialaugmentationclass-wiseprototypekernelmixture
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

Cross-scene remote sensing classification aims to label imagery of an unseen area using knowledge from labeled source scenes, without any labels from the target. The paper proposes MS-CDG, a multi-source domain generalization framework built from two cooperative ideas: a data-aware adversary network that generates augmented multi-source images by learning realistic channel and distribution shifts while preserving class content, and a model-aware diversification stage that models classes both across domains through class-wise prototypes and within domains through a kernel mixture module. The two streams are trained jointly on original and augmented images with a distribution-consistency loss. The paper claims that MS-CDG outperforms eight domain adaptation and domain generalization baselines on three public multi-source datasets, reaching overall accuracies of 81.87%, 56.56%, and 61.77% on Houston, Germany, and LCZ benchmarks, and that each component contributes to the gain in ablations. If correct, this offers a practical way to exploit multiple existing labeled remote sensing sources to label new scenes without collecting target data.

What carries the argument

The load-bearing mechanism is a two-part learning loop. A partly weight-sharing adversary network with five convolutional layers, independent layers for each source plus shared layers, takes an original multi-source patch and outputs an augmented patch, trained by an adversarial loss that maximizes the cross-entropy of the predicted label as the semantic guide and minimizes total variation to suppress noise. The second part is the domain encoder and diversification module: spatial and channel features are first randomized with AdaIN-style normalization called SpaR and ChaR, then fused across domains through a multi-head cross-attention with a coupled enhancement module, and finally modeled by class-wise prototypes for cross-domain clustering plus a kernel mixture module (KMM) with mixture coefficients, means, and covariances for high-order intra-domain class compactness. A KL-divergence consistency loss aligns predictions on original and augmented images and balances the joint classification objective.

What would settle it

Train MS-CDG, then freeze the classifier and run the trained adversary on held-out source patches; if the generated augmented images are classified by a reference model or by the learned prototypes with accuracy close to chance on classes that the originals classify well, the semantic-preservation assumption is violated and the reported gains would be expected to shrink or reverse when the augmentation is stressed.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that a domain generalization model for multi-source remote sensing can be improved by replacing fixed style-based augmentation with an adversary that learns channel- and distribution-level changes across sources while being constrained to keep class semantics, and by diversifying the classifier with two complementary class models: cross-domain class-wise prototypes computed from multi-head cross-attention features, and an intra-domain kernel mixture that captures high-order class statistics. Trained only on labeled source domains, MS-CDG reports overall accuracies of 81.87% on Houston (HSI plus LiDAR), 56.56% on Germany (EnMAP HSI plus Sentinel-1 SAR), and 61.77% on LCZ (Sentinel-1 plus Sentinel-2), exceeding the best compared baselines by 4.34, 3.01, and 2.71 percentage points respectively. Ablations show that removing the kernel mixture, the adversarial augmentation, or the consistency loss each lowers accuracy, and that shared layers in the adversary are needed for the augmentation to help.

Load-bearing premise

The framework assumes that the augmented samples generated by the adversary preserve the class semantics of the original samples, so training on them adds useful diversity rather than label noise.

Editorial extensions

If this is right

  • If MS-CDG is correct, multi-source remote sensing classification can be pushed past current domain adaptation and domain generalization baselines without any target-domain labels, with reported margins of 2.7 to 4.3 points in overall accuracy across three benchmarks.
  • The paper reports lower per-epoch training and inference times than all compared methods on the same GPU, which suggests that the added modules do not trade away efficiency for accuracy.
  • Ablations indicate that the kernel mixture intra-class constraint gives the largest single-model gain, so high-order class modeling is doing essential work beyond the cross-domain prototype clustering.
  • The shared-layer design of the adversary matters: with no shared layers the augmentation quality and accuracy drop, so cross-source feature interaction is part of what makes the generated samples useful.
  • Because only source data is used at training time, a correct MS-CDG could be applied directly to newly acquired scenes without waiting for target labels, which is the practical goal of cross-scene classification.

Reading between the lines

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

  • Beyond the paper: the semantic-guide loss that maximizes cross-entropy is a delicate choice; if the adversary learns class-discriminative perturbations, one testable prediction is that its augmented samples should be classified with high accuracy by a fixed reference model, and that accuracy should track the quality of the final classifier.
  • Beyond the paper: the framework is described for two source domains, but the partly weight-sharing architecture and the prototype and kernel modules should extend to three or more sources, with the relative gain of the adversarial augmentation expected to grow as source diversity increases.
  • Beyond the paper: because the benchmarks differ by sensor, city, and season, the same design is a candidate for fusing optical, SAR, and LiDAR sources over time; a concrete check would be replacing one source with a temporally separated revisit to see whether the consistency loss still stabilizes training.
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 / 6 minor

Summary. The manuscript proposes MS-CDG, a multi-source domain generalization framework for cross-scene remote sensing image classification. The method combines a data-aware adversarial augmentation network that generates channel- and distribution-altered multi-source samples with a model-aware multi-level diversification module based on class-wise prototypes and a kernel mixture model, plus a distribution consistency loss that jointly trains on original and augmented samples. Experiments on three multi-source datasets (Houston 2013→2018, Augsburg→Berlin, LCZ Berlin→Hong Kong) report overall accuracies of 81.87%, 56.56%, and 61.77%, surpassing eight DA/DG baselines, and an ablation study shows each component contributes to the reported performance.

Significance. If the claims are verified, the contribution is meaningful: MS-CDG addresses a practical gap by exploiting multiple source modalities for cross-scene domain generalization, and the framework is described with enough detail to be reimplemented. The paper has clear strengths: ten-run mean±std results, an ablation study, computational-cost comparisons, and the proposed module ablations consistently improve OA on all three benchmarks. However, the central empirical claim is currently unverified because key hyperparameters (α1, α2, and the number of adversarial layers) are selected using target test accuracy, which violates the domain-generalization protocol that motivates the paper. The reported state-of-the-art margins may therefore be an artifact of target leakage rather than of the proposed components.

major comments (3)
  1. [Section IV-C, Fig. 11, Table VIII] The regularization parameters α1 and α2 and the number of adversarial-network layers are chosen per dataset by maximizing OA on the target test sets (Houston 2018, Berlin, Hong Kong). Domain generalization assumes the target is unseen and unlabeled, so tuning on target labels makes Tables IV–VI and the ablation in Table IX a model-selection exercise rather than a fair DG evaluation. The baselines are not reported to receive equivalent target-based tuning, so the headline margins (e.g., 81.87 vs. 77.53 on Houston) may be inflated. Please re-run the evaluation with a source-only validation split or a fixed configuration across datasets, and report both validation-selected hyperparameters and test results.
  2. [Section III-B, Eq. (2)] The adversarial augmentation is claimed to preserve class semantics, but the paper does not verify this. The loss L_ADV only couples the augmented image to the label through a classifier's prediction and a total-variation regularizer; there is no analysis of whether class-discriminative spectral or spatial content survives the augmentation. Given that the central claim rests in part on this semantic guide, please add quantitative evidence (e.g., classification accuracy on augmented source samples using the trained classifier, or prototype-distance preservation) or an ablation with a semantically unguided augmentation variant.
  3. [Section IV-B, Tables IV–VI] The comparison protocol mixes DA baselines that use unlabeled target data during training with DG baselines that use only labeled source data, and the conclusion 'MS-CDG can surpass all comparison methods' is drawn across both settings. While this is common practice, the claim conflates the stricter source-only setting with the weaker DA setting; the discussion should explicitly separate the two groups and state that the main DG comparison is against PDEN, SDENet, and LLURNet under source-only training.
minor comments (6)
  1. [Section III-B, Eq. (2)] The phrase 'maximizing the cross-entropy (CE) loss' appears inconsistent with the equation, where L_CE = (1/N)Σ y log(p) is a negative log-likelihood. The adversary is optimized by minimizing L_ADV = -L_CE + L_TV, which actually minimizes the standard cross-entropy (i.e., maximizes the log-likelihood) rather than maximizing a cross-entropy loss. Please correct the terminology and clarify the intended sign.
  2. [Section III-B / Algorithm 1] The provenance of pg_n in Eq. (2) is unclear: it should state explicitly that pg_n is the output of the task model M applied to the augmented image, and which networks are fixed during the adversarial update. This affects the reproducibility of the adversarial loss computation.
  3. [Section IV-A, paragraph 4] The encoder output dimensions dspa (32 or 64) and dcha (3) are described as 'empirically set'; no sensitivity analysis or justification is provided for these architectural free parameters.
  4. [Table V, class-specific rows] The claim that MS-CDG improves by 2% to 5% over MDA-Net and LLURNet on all TDs should be confined to overall accuracy; on several Germany classes (e.g., class 1, class 2, class 4, class 6) MS-CDG is lower than the best baseline.
  5. [Figure 9] The statement about 'higher separability' in the t-SNE visualization is qualitative; consider reporting a quantitative clustering metric (e.g., adjusted Rand index or silhouette score) on the learned features to support the claim.
  6. [Nomenclature / Abstract] The abbreviation 'MS' is used for both 'multi-source' and 'multispectral'; please disambiguate the two uses in the nomenclature to avoid confusion, since 'MS remote sensing data' could also be read as 'multispectral data'.

Circularity Check

2 steps flagged · score 6.0 of 10

Target test OAs are used to select α1, α2, and adversarial-network depth (Sec. IV-C, Fig. 11, Table VIII), so the reported SOTA margins may reflect target leakage rather than unbiased domain generalization.

  1. fitted input called prediction [Section IV-C, Fig. 11; Table VIII]
    "Fig. 11 provides OA results of MS-CDG corresponding to various combinations of α1 and α2 on three types of MS remote sensing datasets. ... Consequently, the optimal parameters of α1 and α2 are listed as follows: α1 = 0.01 and α2 = 0.1 for Houston datasets, α1 = 0.01 and α2 = 0.05 for Germany datasets, α1 = 0.1 and α2 = 0.1 for LCZ datasets."

    In a domain generalization benchmark the target should remain unseen, but Section IV-C tunes (α1, α2) and the adversarial network depth on the target test OA (Houston 2018, Berlin, Hong Kong) and then reports the resulting accuracies in Tables IV-VI as the method's performance. The target accuracy is therefore not an unbiased prediction for an unseen domain; it is a selection criterion that the reported values were chosen to maximize. The comparison baselines are not reported to receive the same target-based tuning, so the claimed margins over them may be an artifact of this asymmetry rather than evidence for the proposed components.

  2. other [Section IV-B, first paragraph; Section IV-C]
    "All the competing approaches and the proposed method are trained without the utilization of real labels in the TD, which ensures the fairness of the experiment."

    This fairness assertion is undermined by the protocol in Section IV-C, where target OA is used to select α1, α2, and the number of adversarial layers. Even if target labels are not used in the classification loss, using them for model selection is a form of target leakage that breaks the 'unseen target' premise of DG and invalidates the claim that the comparison is a fair test of generalization.

full rationale

The model equations (1)-(17) are derived from labeled source data only: the adversarial augmentation, prototype clustering, KMM, and consistency losses contain no target-domain terms, so there is no self-definitional circularity in the method itself. The circularity is localized to the evaluation protocol. The paper's central claim that MS-CDG 'can surpass all comparison methods' rests on Tables IV-VI, but those exact target OA values were used in Section IV-C to choose the regularization parameters and adversarial-network depth on the target test sets. This is a fitted-input-called-prediction pattern: the reported accuracy is partly constructed by selecting the configuration that maximizes it, rather than measured on a genuinely unseen target. The comparison baselines are not described as receiving equivalent target-based tuning, which exaggerates the fairness asymmetry. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling-via-citation was found. Because the central empirical claim is contaminated but the model derivation itself is not circular, a score of 6 is appropriate.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on tuned hyperparameters rather than a derived theory. The most consequential free parameters are alpha1, alpha2, and the adversarial depth, all selected using target test accuracy. The main domain assumptions are spatial alignment of modalities, label preservation under augmentation, and sufficiency of AdaIN-style randomization. No new physical or external entities are postulated.

free parameters (5)
  • alpha1 = 0.01 (Houston), 0.01 (Germany), 0.1 (LCZ)
    Balance weight between prototype clustering and kernel mixture losses. Selected by target OA in Fig. 11, Section IV-C.
  • alpha2 = 0.1 (Houston), 0.05 (Germany), 0.1 (LCZ)
    Regularization parameter balancing pre-training and consistency loss. Selected by target OA in Fig. 11, Section IV-C.
  • Adversarial network depth = 5 layers (Houston, Germany), 4 layers (LCZ)
    Chosen based on target OA in Table VIII, Section IV-C. Controls degree of weight sharing in the augmentation network.
  • gamma_plus and gamma_minus = 2 and 4
    Hyperparameters in the asymmetric loss for the kernel mixture module. Fixed across all datasets without sensitivity analysis.
  • Encoder output dimensions dspa and dcha = dspa 32 or 64, dcha 3
    Empirically set per dataset in Section IV-A. These affect the capacity of the spatial and channel encoders.
assumptions (5)
  • domain assumption The two input source modalities are spatially aligned and pixel-corresponding patches are meaningful.
    Invoked throughout Section III-A and in the experimental setup (Section IV-A), where HSI, LiDAR, SAR, and multispectral images are co-registered. If the sources are not aligned, the cross-attention and fusion operations lose meaning.
  • domain assumption The augmented samples preserve the class labels of the originals.
    Section III-B relies on the adversarial loss and total variation regularizer to keep semantic content. This is asserted, not proven or measured on the target.
  • domain assumption AdaIN-based randomization of spatial and channel statistics is sufficient to model cross-domain style variation.
    Section III-C1 introduces SpaR and ChaR to reduce imaging attribute differences. The effectiveness of this modeling choice is assumed and only indirectly validated by final accuracies.
  • domain assumption The target domain is unseen and unavailable during training.
    This is the domain generalization setting used for MS-CDG and the DG baselines. It is stated in Section IV-B, while DA baselines additionally use unlabeled target data.
  • standard math Standard loss functions such as cross-entropy, KL divergence, cosine similarity, and Gaussian kernels behave as expected.
    These mathematical tools are used in Eqs. (2), (10), (13), and (16) without proof, as is normal in this literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multisource Collaborative Domain Generalization for Cross-Scene Remote Sensing Image Classification." pith.science (2026). https://pith.science/paper/TJ6XKV3Y

@misc{pith2026241203897,
  author       = {Pith},
  title        = {Pith review of: Multisource Collaborative Domain Generalization for Cross-Scene Remote Sensing Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TJ6XKV3Y}},
  note         = {Machine review of arXiv:2412.03897}
}
read the original abstract

Cross-scene image classification aims to transfer prior knowledge of ground materials to annotate regions with different distributions and reduce hand-crafted cost in the field of remote sensing. However, existing approaches focus on single-source domain generalization to unseen target domains, and are easily confused by large real-world domain shifts due to the limited training information and insufficient diversity modeling capacity. To address this gap, we propose a novel multi-source collaborative domain generalization framework (MS-CDG) based on homogeneity and heterogeneity characteristics of multi-source remote sensing data, which considers data-aware adversarial augmentation and model-aware multi-level diversification simultaneously to enhance cross-scene generalization performance. The data-aware adversarial augmentation adopts an adversary neural network with semantic guide to generate MS samples by adaptively learning realistic channel and distribution changes across domains. In views of cross-domain and intra-domain modeling, the model-aware diversification transforms the shared spatial-channel features of MS data into the class-wise prototype and kernel mixture module, to address domain discrepancies and cluster different classes effectively. Finally, the joint classification of original and augmented MS samples is employed by introducing a distribution consistency alignment to increase model diversity and ensure better domain-invariant representation learning. Extensive experiments on three public MS remote sensing datasets demonstrate the superior performance of the proposed method when benchmarked with the state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2412.03897 by the authors.

Figure 1
Figure 1. Illustrative comparison for MS data augmentation and diversity mod [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of the proposed MS-CDG, including data-aware adversarial augmentation and model-aware multi-level diversification. The multi-domain [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the domain encoder consisting of spatial randomiza [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The conceptual illustration of different feature fusion strategies. (a) [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: MS remote sensing datasets, including two remote sensing data sources and the corresponding ground-truth (GT) map. (a) Houston 2013 dataset. (b) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Visualization results of different compared methods on the Houston datasets. (a) pseudocolor HSI. (b) LiDAR image. (c) Ground-truth map. (d) GAN [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Visualization results of different compared methods on the Germany datasets. (a) pseudocolor HSI. (b) SAR image. (c) Ground-truth map. (d) GAN [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 9
Figure 9. Figure 9: t-SNE embedding visualization of different cross-scene classification [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 8
Figure 8. Figure 8: Visualization results of different compared methods on the LCZ [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 10
Figure 10. Figure 10: Visualization results of the augmented and original MS image on [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Parameter tuning of α1 and α2 for the proposed MS-CDG using all the three experimental datasets. (a) Houston. (b) Germany. (c) LCZ. than other comparison methods on all datasets. The double￾branch network design for cross-domain and intra-domain class modeling ensures…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 57 canonical work pages

  1. [1]

    A review of supervised object-based land-cover image classification,

    L. Ma, M. Li, X. Ma, L. Cheng, P. Du, and Y . Liu, “A review of supervised object-based land-cover image classification,” ISPRS J. Photogramm. Remote Sens. , vol. 130, pp. 277–293, 2017

  2. [2]

    Joint deep learning for land cover and land use classification,

    C. Zhang, I. Sargent, X. Pan, H. Li, A. Gardiner, J. Hare, and P. M. Atkinson, “Joint deep learning for land cover and land use classification,” Remote Sens. Environ., vol. 221, pp. 173–187, 2019

  3. [3]

    Towards computer vision and deep learning facilitated pollination monitoring for agriculture,

    M. N. Ratnayake, A. G. Dyer, and A. Dorin, “Towards computer vision and deep learning facilitated pollination monitoring for agriculture,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2021, pp. 2921–2930

  4. [4]

    Spatio-temporal multi-level attention crop mapping method using time- series sar imagery,

    Z. Han, C. Zhang, L. Gao, Z. Zeng, B. Zhang, and P. M. Atkinson, “Spatio-temporal multi-level attention crop mapping method using time- series sar imagery,” ISPRS J. Photogramm. Remote Sens. , vol. 206, pp. 293–310, 2023

  5. [5]

    Understanding urban landuse from the above and ground perspectives: A deep learning, multimodal solution,

    S. Srivastava, J. E. Vargas-Munoz, and D. Tuia, “Understanding urban landuse from the above and ground perspectives: A deep learning, multimodal solution,” Remote Sens. Environ. , vol. 228, pp. 129–143, 2019

  6. [6]

    Local climate zone-based urban land cover classification from multi-seasonal sentinel-2 images with a recurrent residual network,

    C. Qiu, L. Mou, M. Schmitt, and X. X. Zhu, “Local climate zone-based urban land cover classification from multi-seasonal sentinel-2 images with a recurrent residual network,” ISPRS J. Photogramm. Remote Sens., vol. 154, pp. 151–162, 2019

  7. [7]

    Domain adaptation network for cross-scene classification,

    E. Othman, Y . Bazi, F. Melgani, H. Alhichri, N. Alajlan, and M. Zuair, “Domain adaptation network for cross-scene classification,” IEEE Trans. Geosci. Remote Sens. , vol. 55, no. 8, pp. 4441–4456, 2017

  8. [8]

    Domain adaptation by joint distribution invariant projections,

    S. Chen, M. Harandi, X. Jin, and X. Yang, “Domain adaptation by joint distribution invariant projections,” IEEE Trans. Image Process., vol. 29, pp. 8264–8277, 2020

Show all 60 references
  1. [9]

    Heterogeneous domain adaptation by information capturing and distribution matching,

    H. Wu, H. Zhu, Y . Yan, J. Wu, Y . Zhang, and M. K. Ng, “Heterogeneous domain adaptation by information capturing and distribution matching,” IEEE Trans. Image Process. , vol. 30, pp. 6364–6376, 2021

  2. [10]

    Domain adaptation in remote sensing image classification: A survey,

    J. Peng, Y . Huang, W. Sun, N. Chen, Y . Ning, and Q. Du, “Domain adaptation in remote sensing image classification: A survey,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. , vol. 15, pp. 9842–9859, 2022

  3. [11]

    Domain adap- tation for structured output via discriminative patch representations,

    Y .-H. Tsai, K. Sohn, S. Schulter, and M. Chandraker, “Domain adap- tation for structured output via discriminative patch representations,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , Oct. 2019

  4. [12]

    Instance level affinity-based transfer for unsupervised domain adaptation,

    A. Sharma, T. Kalluri, and M. Chandraker, “Instance level affinity-based transfer for unsupervised domain adaptation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2021, pp. 5361–5371

  5. [13]

    Transferable representation learning with deep adaptation networks,

    M. Long, Y . Cao, Z. Cao, J. Wang, and M. I. Jordan, “Transferable representation learning with deep adaptation networks,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 41, no. 12, pp. 3071–3085, 2019

  6. [14]

    Deep subdomain adaptation network for image classification,

    Y . Zhu, F. Zhuang, J. Wang, G. Ke, J. Chen, J. Bian, H. Xiong, and Q. He, “Deep subdomain adaptation network for image classification,” IEEE Trans. Neural Netw. Learn. Syst. , vol. 32, no. 4, pp. 1713–1722, 2020. ACCEPTED BY IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING,...

  7. [15]

    Optimal transport for domain adaptation,

    R. Flamary, N. Courty, D. Tuia, and A. Rakotomamonjy, “Optimal transport for domain adaptation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 1, no. 1-40, p. 2, 2016

  8. [16]

    Topological structure and semantic information transfer network for cross-scene hyperspectral image classification,

    Y . Zhang, W. Li, M. Zhang, Y . Qu, R. Tao, and H. Qi, “Topological structure and semantic information transfer network for cross-scene hyperspectral image classification,” IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 6, pp. 2817–2830, 2021

  9. [17]

    Adversarial dis- criminative domain adaptation,

    E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell, “Adversarial dis- criminative domain adaptation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 2962–2971

  10. [18]

    Transfer learning with dynamic adversarial adaptation network,

    C. Yu, J. Wang, Y . Chen, and M. Huang, “Transfer learning with dynamic adversarial adaptation network,” in Proc. IEEE Int. Conf. Data Min. (ICDM), 2019, pp. 778–786

  11. [19]

    Class-wise distribution adaptation for unsupervised classification of hyperspectral remote sensing images,

    Z. Liu, L. Ma, and Q. Du, “Class-wise distribution adaptation for unsupervised classification of hyperspectral remote sensing images,” IEEE Trans. Geosci. Remote Sens. , vol. 59, no. 1, pp. 508–521, 2020

  12. [20]

    Domain gener- alization: A survey,

    K. Zhou, Z. Liu, Y . Qiao, T. Xiang, and C. C. Loy, “Domain gener- alization: A survey,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 45, no. 4, pp. 4396–4415, 2023

  13. [21]

    Deep domain generalization via conditional invariant adversarial networks,

    Y . Li, X. Tian, M. Gong, Y . Liu, T. Liu, K. Zhang, and D. Tao, “Deep domain generalization via conditional invariant adversarial networks,” in Proc. Eur. Conf. Comput. Vis. (ECCV) , 2018, pp. 624–639

  14. [22]

    Domain generalization with adversarial feature learning,

    H. Li, S. J. Pan, S. Wang, and A. C. Kot, “Domain generalization with adversarial feature learning,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 5400–5409

  15. [23]

    Domain generalization by solving jigsaw puzzles,

    F. M. Carlucci, A. D’Innocente, S. Bucci, B. Caputo, and T. Tommasi, “Domain generalization by solving jigsaw puzzles,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2019, pp. 2229–2238

  16. [24]

    Selfreg: Self- supervised contrastive regularization for domain generalization,

    D. Kim, Y . Yoo, S. Park, J. Kim, and J. Lee, “Selfreg: Self- supervised contrastive regularization for domain generalization,” inProc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , 2021, pp. 9619–9628

  17. [25]

    Metareg: Towards domain generalization using meta-regularization,

    Y . Balaji, S. Sankaranarayanan, and R. Chellappa, “Metareg: Towards domain generalization using meta-regularization,” in Proc. Adv. Neural Inf. Process. Syst. , vol. 31, 2018

  18. [26]

    Learning to generalize: Meta-learning for domain generalization,

    D. Li, Y . Yang, Y .-Z. Song, and T. Hospedales, “Learning to generalize: Meta-learning for domain generalization,” in Proc. AAAI Conf. Artif. Intell., vol. 32, no. 1, 2018

  19. [27]

    Few-shot sar target recognition through meta-adaptive hyperparameters learning for fast adaptation,

    Z. Zeng, J. Sun, Y . Wang, D. Gu, Z. Han, and W. Hong, “Few-shot sar target recognition through meta-adaptive hyperparameters learning for fast adaptation,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1–17, 2023

  20. [28]

    Generalizing across domains via cross-gradient training,

    S. Shankar, V . Piratla, S. Chakrabarti, S. Chaudhuri, P. Jyothi, and S. Sarawagi, “Generalizing across domains via cross-gradient training,” Proc. Int. Conf. Learn. Repre. , 2018

  21. [29]

    Learning to generate novel domains for domain generalization,

    K. Zhou, Y . Yang, T. Hospedales, and T. Xiang, “Learning to generate novel domains for domain generalization,” in Proc. Eur. Conf. Comput. Vis. (ECCV), 2020, pp. 561–578

  22. [30]

    Progressive domain expansion network for single domain generalization,

    L. Li, K. Gao, J. Cao, Z. Huang, Y . Weng, X. Mi, Z. Yu, X. Li, and B. Xia, “Progressive domain expansion network for single domain generalization,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2021, pp. 224–233

  23. [31]

    Seasonal variation in land-cover classification accuracy in a diverse region,

    P. Sinha, L. Kumar, and N. Reid, “Seasonal variation in land-cover classification accuracy in a diverse region,” Photogramm. Eng. Remote Sens., vol. 78, no. 3, pp. 271–280, 2012

  24. [32]

    Classifier-constrained deep adversarial domain adaptation for cross-domain semisupervised classification in remote sensing images,

    W. Teng, N. Wang, H. Shi, Y . Liu, and J. Wang, “Classifier-constrained deep adversarial domain adaptation for cross-domain semisupervised classification in remote sensing images,” IEEE Geosci. Remote Sens. Lett., vol. 17, no. 5, pp. 789–793, 2019

  25. [33]

    Transfer learning in environmental remote sensing,

    Y . Ma, S. Chen, S. Ermon, and D. B. Lobell, “Transfer learning in environmental remote sensing,” Remote Sens. Environ. , vol. 301, p. 113924, 2024

  26. [34]

    Style and content separation network for remote sensing image cross-scene generalization,

    S. Zhu, C. Wu, B. Du, and L. Zhang, “Style and content separation network for remote sensing image cross-scene generalization,” ISPRS J. Photogramm. Remote Sens. , vol. 201, pp. 1–11, 2023

  27. [35]

    Dual- branch subpixel-guided network for hyperspectral image classification,

    Z. Han, J. Yang, L. Gao, Z. Zeng, B. Zhang, and J. Chanussot, “Dual- branch subpixel-guided network for hyperspectral image classification,” IEEE Trans. Geosci. Remote Sens. , vol. 62, pp. 1–13, 2024

  28. [36]

    Single-source domain expansion network for cross-scene hyperspectral image classification,

    Y . Zhang, W. Li, W. Sun, R. Tao, and Q. Du, “Single-source domain expansion network for cross-scene hyperspectral image classification,” IEEE Trans. Image Process. , vol. 32, pp. 1498–1512, 2023

  29. [37]

    Locally linear unbiased randomization network for cross-scene hyperspectral image classification,

    H. Zhao, J. Zhang, L. Lin, J. Wang, S. Gao, and Z. Zhang, “Locally linear unbiased randomization network for cross-scene hyperspectral image classification,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1–12, 2023

  30. [38]

    More diverse means better: Multimodal deep learning meets remote- sensing imagery classification,

    D. Hong, L. Gao, N. Yokoya, J. Yao, J. Chanussot, Q. Du, and B. Zhang, “More diverse means better: Multimodal deep learning meets remote- sensing imagery classification,” IEEE Trans. Geosci. Remote Sens. , vol. 59, no. 5, pp. 4340–4354, 2020

  31. [39]

    Multimodal hyperspectral unmixing: Insights from attention networks,

    Z. Han, D. Hong, L. Gao, J. Yao, B. Zhang, and J. Chanussot, “Multimodal hyperspectral unmixing: Insights from attention networks,” IEEE Trans. Geosci. Remote Sens. , vol. 60, pp. 1–13, 2022

  32. [40]

    mixup: Beyond empirical risk minimization,

    H. Zhang, M. Cisse, Y . N. Dauphin, and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” in Proc. Int. Conf. Learn. Repre. , 2018

  33. [41]

    Generalisation in humans and deep neural networks,

    R. Geirhos, C. R. Temme, J. Rauber, H. H. Sch ¨utt, M. Bethge, and F. A. Wichmann, “Generalisation in humans and deep neural networks,” in Proc. Adv. Neural Inf. Process. Syst. , vol. 31, 2018, pp. 7549–7561

  34. [42]

    Deep domain generalization with structured low- rank constraint,

    Z. Ding and Y . Fu, “Deep domain generalization with structured low- rank constraint,” IEEE Trans. Image Process. , vol. 27, no. 1, pp. 304– 313, 2018

  35. [43]

    Model-based domain gener- alization,

    A. Robey, G. J. Pappas, and H. Hassani, “Model-based domain gener- alization,” in Proc. Adv. Neural Inf. Process. Syst. , vol. 34, 2021, pp. 20 210–20 229

  36. [44]

    Situational fusion of visual representation for visual navigation,

    W. B. Shen, D. Xu, Y . Zhu, L. J. Guibas, L. Fei-Fei, and S. Savarese, “Situational fusion of visual representation for visual navigation,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , 2019, pp. 2881–2890

  37. [45]

    Learning from extrinsic and intrinsic supervisions for domain generalization,

    S. Wang, L. Yu, C. Li, C.-W. Fu, and P.-A. Heng, “Learning from extrinsic and intrinsic supervisions for domain generalization,” in Proc. Eur. Conf. Comput. Vis. (ECCV) . Springer, 2020, pp. 159–176

  38. [46]

    Domain generalization via model-agnostic learning of semantic features,

    Q. Dou, D. Coelho de Castro, K. Kamnitsas, and B. Glocker, “Domain generalization via model-agnostic learning of semantic features,” in Proc. Adv. Neural Info. Process. Syst. , vol. 32, 2019, pp. 6447–6458

  39. [47]

    Adversarial domain adaptation with domain mixup,

    M. Xu, J. Zhang, B. Ni, T. Li, C. Wang, Q. Tian, and W. Zhang, “Adversarial domain adaptation with domain mixup,” in Proc. AAAI Conf. Artif. Intell. , vol. 34, no. 04, 2020, pp. 6502–6509

  40. [48]

    Open domain gen- eralization with domain-augmented meta-learning,

    Y . Shu, Z. Cao, C. Wang, J. Wang, and M. Long, “Open domain gen- eralization with domain-augmented meta-learning,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2021, pp. 9624–9633

  41. [49]

    Two- branch attention adversarial domain adaptation network for hyperspectral image classification,

    Y . Huang, J. Peng, W. Sun, N. Chen, Q. Du, Y . Ning, and H. Su, “Two- branch attention adversarial domain adaptation network for hyperspectral image classification,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1– 13, 2022

  42. [50]

    Cross-scene joint classification of multisource data with multilevel domain adaption network,

    M. Zhang, X. Zhao, W. Li, Y . Zhang, R. Tao, and Q. Du, “Cross-scene joint classification of multisource data with multilevel domain adaption network,” IEEE Trans. Neural Netw. Learn. Syst. , pp. 1–13, 2023

  43. [51]

    Language-aware domain generalization network for cross-scene hyperspectral image classification,

    Y . Zhang, M. Zhang, W. Li, S. Wang, and R. Tao, “Language-aware domain generalization network for cross-scene hyperspectral image classification,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1–12, 2023

  44. [52]

    A style-based generator architecture for generative adversarial networks,

    T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2019, pp. 4401–4410

  45. [53]

    Prototype selection for nearest neighbor classification: Taxonomy and empirical study,

    S. Garcia, J. Derrac, J. Cano, and F. Herrera, “Prototype selection for nearest neighbor classification: Taxonomy and empirical study,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 34, no. 3, pp. 417–435, 2012

  46. [54]

    Fast and accurate deep network learning by exponential linear units (elus),

    D.-A. Clevert, T. Unterthiner, and S. Hochreiter, “Fast and accurate deep network learning by exponential linear units (elus),” arXiv preprint arXiv:1511.07289, 2015

  47. [55]

    On information and sufficiency,

    S. Kullback and R. A. Leibler, “On information and sufficiency,” The Ann. Math. Stat. , vol. 22, no. 1, pp. 79–86, 1951

  48. [56]

    Open set domain adaptation for hyperspectral image classification using generative adversarial network,

    S. Nirmal, V . Sowmya, and K. Soman, “Open set domain adaptation for hyperspectral image classification using generative adversarial network,” in Proc. Inventive Commun. Comput. Tech. Springer, 2020, pp. 819– 827

  49. [57]

    Hyperspectral and lidar data fusion: Outcome of the 2013 grss data fusion contest,

    C. Debes, A. Merentitis, R. Heremans, J. Hahn, N. Frangiadakis, T. van Kasteren, W. Liao, R. Bellens, A. Pi ˇzurica, S. Gautama et al. , “Hyperspectral and lidar data fusion: Outcome of the 2013 grss data fusion contest,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. , ...

  50. [58]

    2018 ieee grss data fusion contest: Multimodal land use classification [technical com- mittees],

    B. Le Saux, N. Yokoya, R. H ¨ansch, and S. Prasad, “2018 ieee grss data fusion contest: Multimodal land use classification [technical com- mittees],” IEEE Geosci. Remote Sens. Mag. , vol. 6, no. 1, pp. 52–54, 2018

  51. [59]

    The enmap spaceborne imaging spectroscopy mission for earth observation,

    L. Guanter, H. Kaufmann, K. Segl, S. Foerster, C. Rogass, S. Chabrillat, T. Kuester, A. Hollstein, G. Rossner, C. Chlebek et al. , “The enmap spaceborne imaging spectroscopy mission for earth observation,”Remote Sens., vol. 7, no. 7, pp. 8830–8857, 2015

  52. [60]

    Visualizing data using t-sne

    L. Van der Maaten and G. Hinton, “Visualizing data using t-sne.” J. Mach. Learn. Res. , vol. 9, no. 11, 2008

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.