pith. machine review for the scientific record. sign in

arxiv: 2604.27364 · v1 · submitted 2026-04-30 · 💻 cs.CV

Recognition: unknown

Hyperspectral Image Classification via Efficient Global Spectral Supertoken Clustering

Peifu Liu , Tingfa Xu , Jie Wang , Huan Chen , Huiyan Bai , Jianan Li

Authors on Pith no claims yet

Pith reviewed 2026-05-07 10:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords hyperspectral image classificationspectral supertokensdual-stage clusteringboundary preservationsoft labelsremote sensingtoken-level prediction
0
0 comments X

The pith

DSCC decouples clustering from classification to produce boundary-aligned predictions from spectral supertokens in hyperspectral images.

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

Prevailing superpixel methods aggregate pixels into regions yet classify each pixel independently, which undermines the regional consistency the clustering intends to deliver. The paper proposes the Dual-stage Spectrum-Constrained Clustering-based Classifier (DSCC) that first forms spectral supertokens via multi-criteria feature distances and locality-aware regularization, then performs classification at the token level. A soft-label scheme encodes class proportions inside each token to accommodate mixed land-cover content, while density-isolation center selection reduces redundancy. Sympathetic readers would care because hyperspectral remote-sensing data routinely contain sharp boundaries and mixed pixels that require both spatial coherence and precise delineation.

Core claim

DSCC is an end-to-end framework that explicitly decouples clustering from classification. It groups spectrally similar and spatially proximate pixels into boundary-preserving spectral supertokens by computing an image-level multi-criteria feature distance, applying locality-aware assignment regularization, and selecting centers via density-isolation. Token-level prediction then uses a soft-label scheme that records class proportions within each supertoken. This design guarantees region-level, boundary-aligned classification outputs while handling mixed compositions and delivering a favorable accuracy-efficiency trade-off.

What carries the argument

The spectral supertoken, a cluster of spectrally similar and spatially proximate pixels carrying soft class-proportion labels, which shifts classification from pixel-wise to token-level and thereby enforces regional consistency.

Where Pith is reading between the lines

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

  • The token-level soft-label scheme could be adapted to standard semantic segmentation tasks where pixels often straddle class boundaries.
  • Density-isolation center selection might generalize to other clustering-based image partitioning problems that suffer from scale variation.
  • The explicit decoupling of stages offers a template for improving spatial coherence in video or multi-temporal remote-sensing classification pipelines.
  • Real-time remote-sensing systems could incorporate the dual-stage design to reduce per-frame compute while preserving edge accuracy.

Load-bearing premise

The multi-criteria feature distance plus locality-aware assignment regularization will reliably produce boundary-preserving supertokens whose soft-label proportions accurately reflect mixed land-cover content without introducing new classification errors.

What would settle it

Quantitative boundary-alignment evaluation or visual inspection on the WHU-OHS dataset showing that a substantial fraction of supertoken edges cross verified land-cover transitions, which would increase per-pixel classification errors relative to pixel-wise baselines.

Figures

Figures reproduced from arXiv: 2604.27364 by Huan Chen, Huiyan Bai, Jianan Li, Jie Wang, Peifu Liu, Tingfa Xu.

Figure 2
Figure 2. Figure 2: Comparison of classification performance and inference speed. Circle radius represents Floating Point Operations (FLOPs). Our DSCC achieves the best accuracy-efficiency trade-off. cover pixels from multiple classes, especially near bound￾aries or in mixed land-cover areas. Training with one-hot labels at the token level would introduce severe label noise. To address this, we propose a Category Proportion-a… view at source ↗
Figure 3
Figure 3. Figure 3: Overall architecture of the proposed (a) Dual-stage Spectrum-Constrained Clustering-based Classifier. In Stage 1, multiple groups of (b) Spectral-consistent Pixel Aggregation is applied to cluster similar pixels into spectral supertokens, followed by (c) Density-Isolation Center Filtering to optimize the distribution of clustering centers. In Stage 2, spectral supertokens are classified using a Transformer… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results on WHU-OHS. DSCC achieves the best classification accuracy with regional consistency (black-box regions) and precise boundaries. accuracy, its inference speed is relatively low at about 10 FPS, limiting its suitability for real-time applications. In contrast, DSCC abandons DSTC’s fixed-grid, local aggre￾gation strategy and adopts a one-shot, global paradigm for modeling pixel-center ass… view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of clustering centers in feature and image space before and after filtering. Density-Isolation Center Filtering removes less representative centers, resulting in a more uniform and dispersed distribution in the feature space. In image space, centers are reduced in large-scale regions. Patch-based Supertoken Global Supertoken Global Classification Patch-based Classification Ground Truth view at source ↗
Figure 6
Figure 6. Figure 6: Compared with the patch-based supertoken aggregation method. Our global supertoken aggregation method avoids the truncation caused by patch division at boundaries, thereby pre￾serving structure integrity and delineates accurate boundaries. This design helps DSCC better preserve fine-grained bound￾aries while maintaining the integrity of large regions. Visualization of Filtered Centers view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results on IP, KSC and HyRANK-Dioni datasets. DSCC achieves the best classification performance with accurate boundary delineation. False-color SSTN CVSSN MambaHSI S2Mamba DSTC DSCC (Ours) Ground Truth view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results on the WHU-Hi-HanChuan dataset. Clustering similar pixels improves regional consistency and boundary accuracy in DSCC’s output. spatial resolution of approximately 0.109 m. It includes 274 spectral bands in the 400-1000 nm range, distributed across 16 land cover classes. The HyRANK-Dioni dataset has a spatial dimension of 250×1376 pixels, with 176 spectral bands and a spatial resolution… view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study of Spectral Derivative (Spec. Deriv.) and Semantic Feature (Sem. Feat.) on pixel-center similarity. White stars indicate selected centers. Red denotes higher similarity, and blue denotes lower similarity. strong and often leading performance across multiple evalu￾ation metrics. It achieves the best results on all three metrics for the IP, KSC, and WHU-Hi-HanChuan datasets. For ex￾ample, on … view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of classification results under different supervision strategies. Soft label used in our DSCC achieves the best performance, underscoring its effectiveness. Compared with the baseline, adding semantic features alone improves OA to 0.801 and CF1 to 0.725, while adding only the spectral derivative increases OA to 0.794 and CF1 to 0.720. The larger gain from semantic features indicates their stron… view at source ↗
read the original abstract

Hyperspectral image classification demands spatially coherent predictions and precise boundary delineation. Yet prevailing superpixel-based methods face an inherent contradiction: clustering aggregates similar pixels into regions, but the subsequent classifier operates pixel-wise, undermining regional consistency. Consequently, existing approaches do not guarantee region-level, boundary-aligned classification. To address this limitation, we propose the Dual-stage Spectrum-Constrained Clustering-based Classifier (DSCC), an end-to-end framework that explicitly decouples clustering from classification by first grouping spectral similar and spatially proximate pixels into spectral supertokens and then performing token-level prediction. At its core, DSCC computes an image-level multi-criteria feature distance between pixels and centers, followed by a locality-aware assignment regularization, enabling the generation of boundary-preserving spectral supertokens. A density-isolation based center selection further yields representative, well-separated centers, reducing redundancy and improving robustness to scale variation. To accommodate mixed land-cover compositions within each token, we introduce a soft-label scheme that encodes class proportions and improves robustness for mixed-class tokens. DSCC attains a CF1 of 0.728 at 197.75 FPS on the WHU-OHS dataset, offering a superior accuracy-efficiency trade-off compared with state-of-the-art methods. Extensive experiments further validate the effectiveness and generality of the proposed dual-stage paradigm for hyperspectral image classification. The source code is available at https://github.com/laprf/DSCC.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The manuscript proposes the Dual-stage Spectrum-Constrained Clustering-based Classifier (DSCC) for hyperspectral image classification. It decouples clustering from classification by first forming boundary-preserving spectral supertokens via an image-level multi-criteria feature distance, locality-aware assignment regularization, and density-isolation center selection, then performing token-level prediction with a soft-label scheme that encodes class proportions for mixed land-cover pixels. The central empirical claim is a CF1 of 0.728 at 197.75 FPS on the WHU-OHS dataset, with a superior accuracy-efficiency trade-off versus state-of-the-art methods, supported by ablation studies on each component and direct comparisons under consistent protocols.

Significance. If the reported performance holds, the work is significant because it resolves the regional-consistency contradiction inherent in prior superpixel pipelines by enforcing an explicit dual-stage separation and soft-label encoding. The efficiency (nearly 200 FPS) and open-source code are practical strengths that could influence real-time remote-sensing applications. The ablations and reproducible implementation provide a solid basis for follow-on research.

minor comments (3)
  1. [Abstract] Abstract: The performance figures (CF1 and FPS) are stated without reference to hardware platform or batch size; while the full experimental section supplies these details, a brief qualifier in the abstract would improve immediate readability.
  2. [§2] §2: The contrast between spectral supertokens and conventional superpixels is conceptually clear, yet a short quantitative comparison (e.g., average token size or boundary F-score) in the related-work discussion would sharpen the novelty statement.
  3. [§4.2] §4.2: The density-isolation center selection is described algorithmically, but the sensitivity of the isolation threshold to image resolution is not tabulated; adding a one-line sensitivity plot would strengthen the robustness claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive review, the recognition of the dual-stage paradigm's resolution of regional-consistency issues, and the recommendation to accept. We appreciate the note on practical strengths and reproducibility.

Circularity Check

0 steps flagged

No significant circularity; empirical performance claim

full rationale

The paper proposes an algorithmic framework (DSCC) with multi-criteria distance, locality-aware regularization, density-isolation center selection, and soft-label encoding, then reports empirical results (CF1 0.728 at 197.75 FPS on WHU-OHS) plus ablations and comparisons. No equations, fitted parameters, or predictions are defined in terms of themselves; the central claim is a measured trade-off under consistent protocols rather than a self-referential derivation. Self-citations, if present, are not load-bearing for any uniqueness theorem or ansatz that would force the result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Based solely on the abstract, the method introduces several design choices (multi-criteria distance, locality regularization, density-isolation center selection, soft-label scheme) whose effectiveness is asserted but not derived from first principles or external benchmarks.

invented entities (1)
  • spectral supertokens no independent evidence
    purpose: Group spectrally similar and spatially proximate pixels for token-level classification instead of pixel-wise prediction
    New term and concept introduced to resolve the clustering-classification mismatch described in the abstract

pith-pipeline@v0.9.0 · 5563 in / 1211 out tokens · 66896 ms · 2026-05-07T10:00:41.105468+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

54 extracted references · 36 canonical work pages

  1. [1]

    A machine learning framework for drill-core mineral mapping using hyperspectral and high-resolution mineralogical data fusion

    Acosta,I.C.C.,Khodadadzadeh,M.,Tusa,L.,Ghamisi,P.,Gloaguen, R., 2019. A machine learning framework for drill-core mineral mapping using hyperspectral and high-resolution mineralogical data fusion. IEEE Journal of Selected Topics in Applied Earth Observa- tions and Remote Sensing 12, 4829–4842. doi:10.1109/JSTARS.2019. 2924292

    work page doi:10.1109/jstars.2019 2019

  2. [2]

    Content-driven magnitude-derivative spectrum complementary learning for hyper- spectral image classification

    Bai, H., Xu, T., Chen, H., Liu, P., Li, J., 2024. Content-driven magnitude-derivative spectrum complementary learning for hyper- spectral image classification. IEEE Transactions on Geoscience and Remote Sensing 62, 1–14. doi:10.1109/TGRS.2024.3435079

    work page doi:10.1109/tgrs.2024.3435079 2024

  3. [4]

    Prosail-net: A transfer learning-based dual stream neural network to estimate leaf chlorophyll and leaf angle of crops from uav hyperspectralimages

    Bhadra, S., Sagan, V., Sarkar, S., Braud, M., Mockler, T.C., Eveland, A.L., 2024. Prosail-net: A transfer learning-based dual stream neural network to estimate leaf chlorophyll and leaf angle of crops from uav hyperspectralimages. ISPRSJournalofPhotogrammetryandRemote Sensing 210, 1–24. doi:https://doi.org/10.1016/j.isprsjprs.2024. 02.020

    work page doi:10.1016/j.isprsjprs.2024 2024

  4. [6]

    Neural clustering based visual representation learning, in: CVPR, pp

    Chen, G., Li, X., Yang, Y., Wang, W., 2024. Neural clustering based visual representation learning, in: CVPR, pp. 5714–5725

    2024

  5. [7]

    Bridging3-d and2-dconvolutionforhyperspectralimageswithcross-dimensional spectral attention

    Chen,H.,Xu,T.,Liu,P.,Bai,H.,Bian,Z.,Li,J.,2025a. Bridging3-d and2-dconvolutionforhyperspectralimageswithcross-dimensional spectral attention. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 19, 2497–2510. doi:10.1109/ JSTARS.2025.3608249

    work page arXiv 2025

  6. [8]

    Deep feature extraction and classification of hyperspectral images based on con- volutional neural networks

    Chen, Y., Jiang, H., Li, C., Jia, X., Ghamisi, P., 2016. Deep feature extraction and classification of hyperspectral images based on con- volutional neural networks. IEEE Transactions on Geoscience and Remote Sensing 54, 6232–6251. doi:10.1109/TGRS.2016.2584107

    work page doi:10.1109/tgrs.2016.2584107 2016

  7. [9]

    Sit: Scale-interaction transformer for hyperspectral image classification

    Chen, Z., Kumar Roy, S., Gao, H., Ding, Y., Zhang, B., 2025b. Sit: Scale-interaction transformer for hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing 63, 1–17. doi:10.1109/TGRS.2025.3598290

    work page doi:10.1109/tgrs.2025.3598290 2025

  8. [10]

    R-pred:Two-stagemotion prediction via tube-query attention-based trajectory refinement, in: ProceedingsoftheIEEE/CVFInternationalConferenceonComputer Vision (ICCV), pp

    Choi,S.,Kim,J.,Yun,J.,Choi,J.W.,2023. R-pred:Two-stagemotion prediction via tube-query attention-based trajectory refinement, in: ProceedingsoftheIEEE/CVFInternationalConferenceonComputer Vision (ICCV), pp. 8525–8535

    2023

  9. [11]

    An image is worth 16x16 words: Transformers for image recognition at scale, in: ICLR

    Dosovitskiy,A.,Beyer,L.,Kolesnikov,A.,Weissenborn,D.,Zhai,X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N., 2021. An image is worth 16x16 words: Transformers for image recognition at scale, in: ICLR

    2021

  10. [12]

    Simple hardware-efficient long convolutions for sequencemodeling, in: Proceedings of International Conference on Machine Learning, pp

    Fu, D.Y., Epstein, E.L., Nguyen, E., Thomas, A.W., Zhang, M., Dao, T., Rudra, A., Re, C., 2023. Simple hardware-efficient long convolutions for sequencemodeling, in: Proceedings of International Conference on Machine Learning, pp. 10373–10391

    2023

  11. [13]

    Mamba: Linear-time sequence modeling with selective state spaces

    Gu, A., Dao, T., 2024. Mamba: Linear-time sequence modeling with selective state spaces

    2024

  12. [14]

    Han, Z., Yang, J., Gao, L., Zeng, Z., Zhang, B., Chanussot, J.,

  13. [15]

    IEEETransactionsonGeoscienceandRemoteSensing 63, 1–14

    Subpixel spectral variability network for hyperspectral image classification. IEEETransactionsonGeoscienceandRemoteSensing 63, 1–14. doi:10.1109/TGRS.2025.3535749

    work page doi:10.1109/tgrs.2025.3535749 2025

  14. [16]

    Spectralformer: Rethinking hyperspectral image classifica- tionwithtransformers.IEEETransactionsonGeoscienceandRemote Sensing 60, 1–15

    Hong, D., Han, Z., Yao, J., Gao, L., Zhang, B., Plaza, A., Chanussot, J., 2022. Spectralformer: Rethinking hyperspectral image classifica- tionwithtransformers.IEEETransactionsonGeoscienceandRemote Sensing 60, 1–15. doi:10.1109/TGRS.2021.3130716

    work page doi:10.1109/tgrs.2021.3130716 2022

  15. [17]

    Deep diversity-enhanced feature representation of hyperspectral images

    Hou, J., Zhu, Z., Hou, J., Liu, H., Zeng, H., Meng, D., 2024. Deep diversity-enhanced feature representation of hyperspectral images. IEEE Transactions on Pattern Analysis and Machine Intelligence 46, 8123–8138. doi:10.1109/TPAMI.2024.3399753

    work page doi:10.1109/tpami.2024.3399753 2024

  16. [18]

    Super- pixel sampling networks, in: ECCV

    Jampani, V., Sun, D., Liu, M.Y., Yang, M.H., Kautz, J., 2018. Super- pixel sampling networks, in: ECCV

    2018

  17. [19]

    Whu-ohs: A benchmark dataset for large-scale hersepctral image classification

    Li, J., Huang, X., Tu, L., 2022. Whu-ohs: A benchmark dataset for large-scale hersepctral image classification. International Journal of Applied Earth Observation and Geoinformation 113, 103022

    2022

  18. [20]

    Exploring the relationship between center and neighborhoods: Central vector ori- ented self-similarity network for hyperspectral image classification

    Li, M., Liu, Y., Xue, G., Huang, Y., Yang, G., 2023. Exploring the relationship between center and neighborhoods: Central vector ori- ented self-similarity network for hyperspectral image classification. IEEETransactionsonCircuitsandSystemsforVideoTechnology33, 1979–1993. doi:10.1109/TCSVT.2022.3218284

    work page doi:10.1109/tcsvt.2022.3218284 2023

  19. [21]

    Mambahsi: Spatial–spectral mamba for hyperspectral image classification

    Li, Y., Luo, Y., Zhang, L., Wang, Z., Du, B., 2024. Mambahsi: Spatial–spectral mamba for hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing 62, 1–16. doi:10. Peifu Liu, et al:Preprint submitted to ElsevierPage 13 of 14 Hyperspectral Image Classification via Efficient Global Spectral Supertoken Clustering 1109/TGRS.2024.3430985

    work page arXiv 2024

  20. [22]

    Dual-stage hyperspectralimageclassificationmodelwithspectralsupertoken,in: ECCV, pp

    Liu, P., Xu, T., Wang, J., Chen, H., Bai, H., Li, J., 2025a. Dual-stage hyperspectralimageclassificationmodelwithspectralsupertoken,in: ECCV, pp. 368–386

  21. [23]

    Hypermamba: A spectral-spatial adaptive mamba for hyperspectral image classifi- cation

    Liu, Q., Yue, J., Fang, Y., Xia, S., Fang, L., 2024. Hypermamba: A spectral-spatial adaptive mamba for hyperspectral image classifi- cation. IEEE Transactions on Geoscience and Remote Sensing 62, 1–14. doi:10.1109/TGRS.2024.3482473

    work page doi:10.1109/tgrs.2024.3482473 2024

  22. [24]

    Spatial–spectral enhancement and fusion network for hyperspectral image classifica- tionwithfewlabeledsamples

    Liu, S., Fu, C., Duan, Y., Wang, X., Luo, F., 2025b. Spatial–spectral enhancement and fusion network for hyperspectral image classifica- tionwithfewlabeledsamples. IEEETransactionsonGeoscienceand Remote Sensing 63, 1–14. doi:10.1109/TGRS.2024.3523578

    work page doi:10.1109/tgrs.2024.3523578 2024

  23. [25]

    Grouped multi-attention network for hyperspectral image spectral-spatial classification

    Lu, T., Liu, M., Fu, W., Kang, X., 2023. Grouped multi-attention network for hyperspectral image spectral-spatial classification. IEEE Transactions on Geoscience and Remote Sensing 61, 1–12. doi:10. 1109/TGRS.2023.3263851

    work page arXiv 2023

  24. [26]

    Image as set of points, in: ICLR

    Ma, X., Zhou, Y., Wang, H., Qin, C., Sun, B., Liu, C., Fu, Y., 2023. Image as set of points, in: ICLR

    2023

  25. [27]

    Hyperspectralimageclassifi- cationusinggroup-awarehierarchicaltransformer.IEEETransactions onGeoscienceandRemoteSensing60,1–14

    Mei,S.,Song,C.,Ma,M.,Xu,F.,2022. Hyperspectralimageclassifi- cationusinggroup-awarehierarchicaltransformer.IEEETransactions onGeoscienceandRemoteSensing60,1–14. doi:10.1109/TGRS.2022. 3207933

    work page doi:10.1109/tgrs.2022 2022

  26. [28]

    Picovs: Pixel-level with covariance pooling feature and superpixel-level fea- ture fusion for hyperspectral image classification

    Nartey, O.T., Sarpong, K., Addo, D., Rao, Y., Qin, Z., 2023. Picovs: Pixel-level with covariance pooling feature and superpixel-level fea- ture fusion for hyperspectral image classification. IEEE Transactions onGeoscienceandRemoteSensing61,1–20. doi:10.1109/TGRS.2023. 3322641

    work page doi:10.1109/tgrs.2023 2023

  27. [29]

    Hyperloopnet: Hyperspectral image classification using multiscale self-looping convolutional networks

    Pande, S., Banerjee, B., 2022. Hyperloopnet: Hyperspectral image classification using multiscale self-looping convolutional networks. ISPRS Journal of Photogrammetry and Remote Sensing 183, 422–

    2022

  28. [30]

    doi:https://doi.org/10.1016/j.isprsjprs.2021.11.021

    work page doi:10.1016/j.isprsjprs.2021.11.021 2021

  29. [31]

    Hsod-bit-v2: A challenging benchmark for hyperspectral salient object detection

    Qiu, Y., Bai, S., Xu, T., Liu, P., Qin, H., Li, J., 2025. Hsod-bit-v2: A challenging benchmark for hyperspectral salient object detection. Proceedings of the AAAI Conference on Artificial Intelligence 39, 6630–6638

    2025

  30. [32]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Ren, S., He, K., Girshick, R., Sun, J., 2017. Faster r-cnn: Towards real-time object detection with region proposal networks. IEEE TransactionsonPatternAnalysisandMachineIntelligence39,1137–

    2017

  31. [33]

    doi:10.1109/TPAMI.2016.2577031

    work page doi:10.1109/tpami.2016.2577031 2016

  32. [34]

    Spectral–spatial morphological attention transformer for hyperspec- tral image classification

    Roy, S.K., Deria, A., Shah, C., Haut, J.M., Du, Q., Plaza, A., 2023. Spectral–spatial morphological attention transformer for hyperspec- tral image classification. IEEE Transactions on Geoscience and Remote Sensing 61, 1–15. doi:10.1109/TGRS.2023.3242346

    work page doi:10.1109/tgrs.2023.3242346 2023

  33. [35]

    Remote sensing monitoring of multi-scale watersheds impermeability for urban hy- drological evaluation

    Shao, Z., Fu, H., Li, D., Altan, O., Cheng, T., 2019. Remote sensing monitoring of multi-scale watersheds impermeability for urban hy- drological evaluation. Remote Sensing of Environment 232, 111338

    2019

  34. [36]

    Dualmamba: A lightweight spectral–spatial mamba-convolution network for hyper- spectral image classification

    Sheng, J., Zhou, J., Wang, J., Ye, P., Fan, J., 2025. Dualmamba: A lightweight spectral–spatial mamba-convolution network for hyper- spectral image classification. IEEE Transactions on Geoscience and Remote Sensing 63, 1–15. doi:10.1109/TGRS.2024.3516817

    work page doi:10.1109/tgrs.2024.3516817 2025

  35. [37]

    Hyperspectral image classification using a superpixel–pixel–subpixel multilevel network

    Tu, B., Ren, Q., Li, Q., He, W., He, W., 2023. Hyperspectral image classification using a superpixel–pixel–subpixel multilevel network. IEEE Transactions on Instrumentation and Measurement 72, 1–16. doi:10.1109/TIM.2023.3271713

    work page doi:10.1109/tim.2023.3271713 2023

  36. [38]

    Dual-stage construction of probability for hyperspectral image classification

    Tu, B., Wang, J., Zhao, G., Zhang, X., Zhang, G., 2019. Dual-stage construction of probability for hyperspectral image classification. IEEE Geoscience and Remote Sensing Letters 17, 889–893

    2019

  37. [39]

    Attentionisallyouneed,in: NIPS

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez, A.N.,Kaiser,L.u.,Polosukhin,I.,2017. Attentionisallyouneed,in: NIPS

    2017

  38. [40]

    Prosdm:Applicabilityofprospectmodelcoupledwith spectralderivativesandsimilaritymetricstoretrieveleafbiochemical traitsfrombidirectionalreflectance

    Wan, L., Zhang, J., Xu, Y., Huang, Y., Zhou, W., Jiang, L., He, Y., Cen,H.,2021. Prosdm:Applicabilityofprospectmodelcoupledwith spectralderivativesandsimilaritymetricstoretrieveleafbiochemical traitsfrombidirectionalreflectance. RemoteSensingofEnvironment 267, 112761. doi:https://doi.org/10.1016/j.rse.2021.112761

    work page doi:10.1016/j.rse.2021.112761 2021

  39. [41]

    Retrieval-Augmented Embodied Agents

    Wang, G., Guo, Y., Xu, Z., Kankanhalli, M., 2024a. Bilateral adaptation for human-object interaction detection with occlusion- robustness, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 27970–27980. doi:10.1109/ CVPR52733.2024.02642

    work page arXiv 2024

  40. [42]

    S2mamba: A spatial–spectral state space model for hyperspectral image classi- fication

    Wang, G., Zhang, X., Peng, Z., Zhang, T., Jiao, L., 2025. S2mamba: A spatial–spectral state space model for hyperspectral image classi- fication. IEEE Transactions on Geoscience and Remote Sensing 63, 1–13. doi:10.1109/TGRS.2025.3530993

    work page doi:10.1109/tgrs.2025.3530993 2025

  41. [43]

    Neurodin: A two-stage framework for high- fidelity neural surface reconstruction, in: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Wang, Y., Huang, D., Ye, W., Zhang, G., Ouyang, W., He, T., 2024b. Neurodin: A two-stage framework for high- fidelity neural surface reconstruction, in: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc.. pp. 103168–103197. URL: https://pro...

    work page doi:10.52202/079017-3278 2024

  42. [44]

    1955–1964

    Wang,Y.C.,Wang,C.Y.,Lai,S.H.,2022.Disentangledrepresentation withdual-stagefeaturelearningforfaceanti-spoofing,in:WACV,pp. 1955–1964

    2022

  43. [45]

    From center to surrounding: An interactivelearningframeworkforhyperspectralimageclassification

    Yang, J., Du, B., Zhang, L., 2023. From center to surrounding: An interactivelearningframeworkforhyperspectralimageclassification. ISPRS Journal of Photogrammetry and Remote Sensing 197, 145–

    2023

  44. [46]

    doi:https://doi.org/10.1016/j.isprsjprs.2023.01.024

    work page doi:10.1016/j.isprsjprs.2023.01.024 2023

  45. [47]

    A cross-level spectral–spatial joint encode learning framework for imbalanced hy- perspectral image classification

    Yu, D., Li, Q., Wang, X., Xu, C., Zhou, Y., 2022. A cross-level spectral–spatial joint encode learning framework for imbalanced hy- perspectral image classification. IEEE Transactions on Geoscience and Remote Sensing 60, 1–17. doi:10.1109/TGRS.2022.3203980

    work page doi:10.1109/tgrs.2022.3203980 2022

  46. [48]

    Tcformer: Visual recognition via token clustering transformer

    Zeng, W., Jin, S., Xu, L., Liu, W., Qian, C., Ouyang, W., Luo, P., Wang, X., 2024. Tcformer: Visual recognition via token clustering transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence 46, 9521–9535. doi:10.1109/TPAMI.2024.3425768

    work page doi:10.1109/tpami.2024.3425768 2024

  47. [49]

    Three-dimensionspatial–spectralattentiontransformerfor hyperspectralimagedenoising.IEEETransactionsonGeoscienceand Remote Sensing 62, 1–13

    Zhang, Q., Dong, Y., Zheng, Y., Yu, H., Song, M., Zhang, L., Yuan, Q.,2024a. Three-dimensionspatial–spectralattentiontransformerfor hyperspectralimagedenoising.IEEETransactionsonGeoscienceand Remote Sensing 62, 1–13. doi:10.1109/TGRS.2024.3458174

    work page doi:10.1109/tgrs.2024.3458174 2024

  48. [50]

    Cooperated spectrallow-ranknessprioranddeepspatialpriorforhsiunsupervised denoising

    Zhang,Q.,Yuan,Q.,Song,M.,Yu,H.,Zhang,L.,2022a. Cooperated spectrallow-ranknessprioranddeepspatialpriorforhsiunsupervised denoising. IEEE Transactions on Image Processing 31, 6356–6368. doi:10.1109/TIP.2022.3211471

    work page doi:10.1109/tip.2022.3211471 2022

  49. [51]

    Hyperspectral image denoising: From model-driven, data- driven,tomodel-data-driven

    Zhang, Q., Zheng, Y., Yuan, Q., Song, M., Yu, H., Xiao, Y., 2024b. Hyperspectral image denoising: From model-driven, data- driven,tomodel-data-driven. IEEETransactionsonNeuralNetworks and Learning Systems 35, 13143–13163. doi:10.1109/TNNLS.2023. 3278866

    work page doi:10.1109/tnnls.2023 2023

  50. [52]

    Spectral–spatial self-attention networks for hyperspectral image classification

    Zhang, X., Sun, G., Jia, X., Wu, L., Zhang, A., Ren, J., Fu, H., Yao, Y., 2022b. Spectral–spatial self-attention networks for hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing 60, 1–15. doi:10.1109/TGRS.2021.3102143

    work page doi:10.1109/tgrs.2021.3102143 2021

  51. [53]

    completely blind

    Zhao, C., Zhu, W., Feng, S., 2022. Superpixel guided deformable convolution network for hyperspectral image classification. IEEE Transactions on Image Processing 31, 3838–3851. doi:10.1109/TIP. 2022.3176537

    work page doi:10.1109/tip 2022

  52. [54]

    Fpga:Fastpatch-free global learning framework for fully end-to-end hyperspectral image classification

    Zheng,Z.,Zhong,Y.,Ma,A.,Zhang,L.,2020. Fpga:Fastpatch-free global learning framework for fully end-to-end hyperspectral image classification. IEEETransactionsonGeoscienceandRemoteSensing 58, 5612–5626. doi:10.1109/TGRS.2020.2967821

    work page doi:10.1109/tgrs.2020.2967821 2020

  53. [55]

    Whu-hi: Uav-borne hyperspectral with high spatial resolution (h2) benchmarkdatasetsandclassifierforprecisecropidentificationbased on deep convolutional neural network with crf

    Zhong, Y., Hu, X., Luo, C., Wang, X., Zhao, J., Zhang, L., 2020. Whu-hi: Uav-borne hyperspectral with high spatial resolution (h2) benchmarkdatasetsandclassifierforprecisecropidentificationbased on deep convolutional neural network with crf. Remote Sensing of Environment 250, 112012. doi:https://doi.org/10.1016/j.rse.2020. 112012

    work page doi:10.1016/j.rse.2020 2020

  54. [56]

    Spectral–spatial transformer network for hyperspectral image classification: A fac- torized architecture search framework

    Zhong, Z., Li, Y., Ma, L., Li, J., Zheng, W.S., 2022. Spectral–spatial transformer network for hyperspectral image classification: A fac- torized architecture search framework. IEEE Transactions on Geo- science and Remote Sensing 60, 1–15. doi:10.1109/TGRS.2021. 3115699. Peifu Liu, et al:Preprint submitted to ElsevierPage 14 of 14

    work page doi:10.1109/tgrs.2021 2022