Pith. sign in

REVIEW 4 major objections 6 minor 73 references

Fine-grained Hierarchical Crop Type Classification from Integrated Hyperspectral EnMAP Data and Multispectral Sentinel-2 Time Series: A Large-scale Dataset and Dual-stream Transformer Method

T0 review · 4 major / 6 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 Adding EnMAP hyperspectral imagery to Sentinel-2 time series lifts fine-grained crop classification F1 by an average of 4.2 percentage points.

desk verdict A valuable new benchmark dataset, but the headline 4.2% hyperspectral gain is not yet separated from added model capacity. read the letter →

arxiv 2506.06155 v2 pith:SQSVQGAK submitted 2025-06-06 cs.CV cs.LG

Wenyuan Li , Shunlin Liang , Yuxiang Zhang , Liqin Liu , Keyan Chen , Yongzhe Chen , Han Ma , Jianglei Xu
show 3 more authors
Yichuan Ma Shikang Guan Zhenwei Shi
This is my paper
classification cs.CVcs.LG
keywords croptypeclassificationhyperspectralimageryEnMAPSentinel-2timeserieshierarchicaltaxonomydual-streamtransformermulti-modalfusionH2Cropdataset
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces H2Crop, a large benchmark that pairs 30-m EnMAP hyperspectral images with monthly 10-m Sentinel-2 time series over France, labeling more than a million field parcels across a four-tier taxonomy (6, 36, 82, and 101 crop classes). On top of this dataset, it proposes a dual-stream Transformer that processes hyperspectral and multispectral-temporal data separately and fuses them via pixel-shuffle upsampling and convolutional layers, with a cascade of hierarchical classification heads. The paper's central claim is that adding hyperspectral EnMAP data to Sentinel-2 time series raises average F1 by 4.2 percentage points, with gains peaking at 6.3% at finer taxonomic levels. The authors argue this shows that nanometer-scale spectral resolution provides information that temporal multispectral data alone cannot capture, and they present the dataset and code as an open benchmark for testing that hypothesis.

What carries the argument

The architecture is a dual-stream Transformer. A spectral-spatial decoupled Vision Transformer treats hyperspectral pixel spectra as tokens in one branch and spatial patches in another; a Video Swin Transformer processes the Sentinel-2 monthly stack to learn spatiotemporal phenological patterns; pixel-shuffle upsampling and convolutional fusion reconcile the 30-m versus 10-m resolution difference; and a cascade of four classification heads combines fused features with the previous level's probability output and prior-year crop-type embeddings. This carries the argument by isolating where hyperspectral information enters the decision.

What would settle it

Retrain the model on a cleaned label subset where LPIS declarations have been verified by field surveys or independent high-resolution imagery. If the 4.2-point F1 gain shrinks or disappears on verified labels, the reported benefit would reflect administrative label noise rather than true spectral separability.

Watch

Extended reading notes

Core claim

The paper claims that hyperspectral imagery is not redundant with phenological time series: when EnMAP data is added to the Sentinel-2 baseline, average F1 rises from 35.6 to 39.8 (a 4.2-point gain) with 6-month inputs, and the gain is concentrated at Levels 3 and 4, where F1 improves by 6.3 and 5.5 points. With prior-year crop maps included, hyperspectral still adds 3.1 points on average. The gain persists across temporal windows up to 12 months and is largest in areas where the crop changed from the previous year, where hyperspectral data counteracts the misleading effect of prior knowledge.

Load-bearing premise

The load-bearing premise is that France's LPIS farmer declarations are accurate enough, at parcel level, to serve as ground truth for 101 fine-grained crop types, including variety distinctions such as winter versus spring common soft wheat.

Editorial extensions

If this is right

  • Adding hyperspectral data delivers a measurable F1 gain even with early-season inputs, supporting in-season monitoring before the full growing cycle is observed.
  • The largest gains at the finest taxonomic levels mean that differentiating similar varieties, such as winter versus spring common soft wheat, benefits most from spectral detail.
  • Prior-year crop maps help classification in stable areas but hurt in rotated areas, and hyperspectral data partially offsets that damage, which gives practical guidance on when to use historical labels.
  • The public H2Crop dataset gives the community a common basis for comparing hyperspectral-multispectral fusion methods and for studying rare-class and hierarchical-label problems.

Reading between the lines

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

  • If LPIS labels contain misreporting at variety level, the absolute F1 numbers may compress toward the noise ceiling; the relative 4.2-point gain could be tested by retraining on a field-verified label subset.
  • The spectral branch's contribution could be dissected by ablating specific EnMAP spectral regions (e.g., SWIR versus VNIR) to identify which wavelengths drive the fine-grained gains.
  • The same dual-stream design could be applied to other 30-m imaging spectrometers, such as DESIS or EMIT, to test whether the benefit generalizes beyond EnMAP.
  • A simpler unimodal baseline with denser temporal sampling, not just monthly composites, would clarify how much of the gain is truly spectral rather than a byproduct of comparing against a temporally impoverished multispectral baseline.
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 / 6 minor

Summary. The paper introduces H2Crop, a large-scale dataset of 16,344 EnMAP hyperspectral tiles paired with Sentinel-2 time series over France, featuring a four-tier crop taxonomy derived from LPIS declarations. The authors propose a dual-stream Transformer that processes hyperspectral data through a Spectral-Spatial Decoupled Vision Transformer and Sentinel-2 time series through a Video Swin Transformer, with a hierarchical classification head. The central empirical claim is that adding EnMAP hyperspectral data to Sentinel-2 time series improves average F1 by 4.2 percentage points (up to 6.3% at Level 3). The paper also reports comparisons against CNN/LSTM baselines and analyzes effects across temporal windows and crop-change scenarios.

Significance. If the central claim is properly validated, H2Crop would be a substantial contribution: it is the first large-scale benchmark combining EnMAP hyperspectral imagery with Sentinel-2 time series, and the four-tier taxonomy with over one million parcels addresses a clear gap in fine-grained agricultural remote sensing. The paper ships a dataset and code repository, which is commendable and increases reproducibility. The methodological claim that hyperspectral data provides complementary discriminative information for fine-grained crop types is plausible and practically important. However, the current experimental design does not conclusively separate the benefit of the EnMAP spectral content from the benefit of increased model capacity, and the absence of error bars undermines the stated statistical significance. The analysis of changed vs. unchanged crop areas in Section 5.1 is a thoughtful addition that partially supports the value of hyperspectral data, but it suffers from the same capacity-confounding issue.

major comments (4)
  1. [§4.1, Table 2] The headline improvement from adding hyperspectral data is confounded with added model capacity. In the S2-only configuration, the model uses only the Sentinel-2 temporal Swin Transformer; in S2+Hyper, it additionally employs the Spectral-Spatial Decoupled ViT (two extra transformer branches) plus fusion convolutions. The paper does not match parameter counts, FLOPs, or training budget across configurations, nor does it include a control where the same extra capacity receives a content-free input (e.g., random noise or spatially upsampled Sentinel-2 spectral bands). Without such a control, the 4.2% F1 gain cannot be attributed specifically to EnMAP's spectral information. I recommend adding a same-capacity S2-only variant (e.g., a second temporal branch with duplicated S2 bands) and a noise-input hyperspectral branch.
  2. [§4.1, Section 3.5] The text states that the 4.2% improvement is 'a statistically significant improvement', but no error bars, confidence intervals, number of repeated seeds, or significance tests are reported anywhere in the experimental section. Given that model selection is based on the best validation F1 across epochs, the reported numbers could reflect overfitting to the validation split. Please provide standard deviations over at least three independent training runs and, ideally, a paired significance test (e.g., Wilcoxon signed-rank test across class-level F1 scores or across test tiles).
  3. [§2.1.2] The ground truth is derived from France's LPIS farmer declarations, which are administrative data subject to misreporting, coarseness, and inconsistencies with actual cultivar-level plantings. The paper assumes these declarations are accurate enough to supervise 101 fine-grained crop types, including variety-level labels such as 'winter common soft wheat'. Without any label-quality check or error analysis, the reported F1 improvements at Levels 3-4 may reflect the structure of administrative noise rather than true spectral separability. Please include an analysis of label ambiguity (e.g., comparison with independent ground-truth data for a subset, or an assessment of parcel-level label change rates) and discuss how this affects the interpretation of the fine-grained results.
  4. [§4.2, Table 3] The comparison against 'existing deep learning approaches' is limited to three convolutional/LSTM baselines (UNet, 3DCNN, CNN-LSTM), all sharing the same hierarchical head and the same training strategy. No recent Transformer-based crop classification models (e.g., BreizhCrops-style transformers, TimeSformer, U-Net with attention, or the AgriFM foundation model cited by the authors) are compared. Given that the proposed method is a Transformer, a comparison only against CNNs does not substantiate the claim of 'higher accuracy over existing deep learning approaches' in a meaningful way, especially because the capacity difference may explain the gap. Please add at least one recent spatiotemporal-transformer baseline with matched training budget.
minor comments (6)
  1. [§2.3, Table 1] The table's row for the MDAS dataset is labeled 'MADS' in the second column; please correct the typo.
  2. [Figure 1 caption] The word 'Sentienl-2' is misspelled; it should be 'Sentinel-2'.
  3. [§3.1] The notation for the spectral and spatial transformer outputs is inconsistent: the text uses both 'X' and 'O' for the output features, and Equation (1) combines O_S and O_P without explicitly defining the channel dimensions after concatenation. Please clarify.
  4. [§3.2] The description of the Video Swin Transformer decoder is hard to follow, particularly the shape transformation of encoder features (F'_j) and the temporal dimension alignment. A table of tensor shapes at each stage would improve readability.
  5. [§4.1, Figure 6] The radar charts use the same color for the S2-only and S2+Prior+Hyper curves in the legend, making the visualization difficult to interpret; consider using dashed or dotted line styles in addition to color.
  6. [References] Several references have incomplete author lists (e.g., An et al., 2024; Chen et al., 2024a; Li et al., 2025) or missing page numbers; please verify against the publisher versions.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the 4.2% F1 gain is a held-out empirical comparison, not a fitted parameter or a self-citation chain.

full rationale

The paper's central claim—adding EnMAP hyperspectral data to Sentinel-2 time series improves average F1 by 4.2%—is measured on a stratified, spatially de-overlapped test split (Section 2.2, Table 2) and is not a fitted quantity. The S2-only and S2+Hyper configurations share the same training protocol; the reported gain does not reduce by construction to any input, and no equation defines the benefit in terms of the data used to fit the model. Self-citations (e.g., Li et al. 2025 AgriFM as 'inspiration' for the temporal branch, and Chen et al. 2024a/2025 in future directions) are incidental design pointers, not load-bearing evidence for the headline result. The hierarchical labels derive from the external LPIS/EuroCrops HCATv3 framework, not from the model. The only in-scope weakness is Section 4.1's phrase 'statistically significant improvement' in the absence of reported error bars or repeated-seed tests; that is a missing-evidence/correctness concern, not a circular step. Accordingly, no circularity is present.

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

The central claim rests on the reliability of LPIS labels, the adequacy of the spatial split, and the correctness of the HCATv3 mapping. No new physical entities, forces, or dimensions are introduced; the model is a neural architecture using existing operators.

free parameters (3)
  • Fusion output dimension E = 128
    Both branches output features with E=128 for fusion, set by hand with no reported sweep (Section 3.4).
  • Spectral transformer pooling size = 4x4
    A 4x4 average pooling is applied before the spectral transformer; chosen without ablation (Section 3.4).
  • Number of transformer layers in spatial/spectral branches = 6
    Both branches use six transformer layers, a manually chosen truncation of ViT-base (Section 3.4).
assumptions (4)
  • domain assumption LPIS ground reference labels are accurate at parcel level for all 101 fine-grained crop types.
    Ground truth in H2Crop derives from farmer declarations; errors in crop type or variety reporting would propagate to benchmark labels (Section 2.1.2).
  • domain assumption Spatial overlap removal between train/val/test tiles is sufficient to prevent spatial autocorrelation leakage.
    The dataset validation step removes overlapping EnMAP acquisition footprints, but the threshold and method are not detailed (Section 2.2).
  • domain assumption Monthly Sentinel-2 cloud-free composites preserve the phenological signal needed for crop discrimination.
    Compositing over a month can blur fast phenological transitions; this is standard practice but an assumption (Section 2.1.1).
  • domain assumption HCATv3 mapping from French LPIS categories to 101 crop types is correct and consistent across years.
    EuroCrops taxonomy mapping is used to generate labels; mapping errors would systematically mislabel classes (Section 2.1.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fine-grained Hierarchical Crop Type Classification from Integrated Hyperspectral EnMAP Data and Multispectral Sentinel-2 Time Series: A Large-scale Dataset and Dual-stream Transformer Method." pith.science (2026). https://pith.science/paper/SQSVQGAK

@misc{pith2026250606155,
  author       = {Pith},
  title        = {Pith review of: Fine-grained Hierarchical Crop Type Classification from Integrated Hyperspectral EnMAP Data and Multispectral Sentinel-2 Time Series: A Large-scale Dataset and Dual-stream Transformer Method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SQSVQGAK}},
  note         = {Machine review of arXiv:2506.06155}
}
read the original abstract

Fine-grained crop type classification serves as the fundamental basis for large-scale crop mapping and plays a vital role in ensuring food security. It requires simultaneous capture of both phenological dynamics (obtained from multi-temporal satellite data like Sentinel-2) and subtle spectral variations (demanding nanometer-scale spectral resolution from hyperspectral imagery). Research combining these two modalities remains scarce currently due to challenges in hyperspectral data acquisition and crop types annotation costs. To address these issues, we construct a hierarchical hyperspectral crop dataset (H2Crop) by integrating 30m-resolution EnMAP hyperspectral data with Sentinel-2 time series. With over one million annotated field parcels organized in a four-tier crop taxonomy, H2Crop establishes a vital benchmark for fine-grained agricultural crop classification and hyperspectral image processing. We propose a dual-stream Transformer architecture that synergistically processes these modalities. It coordinates two specialized pathways: a spectral-spatial Transformer extracts fine-grained signatures from hyperspectral EnMAP data, while a temporal Swin Transformer extracts crop growth patterns from Sentinel-2 time series. The designed hierarchical classification head with hierarchical fusion then simultaneously delivers multi-level crop type classification across all taxonomic tiers. Experiments demonstrate that adding hyperspectral EnMAP data to Sentinel-2 time series yields a 4.2% average F1-scores improvement (peaking at 6.3%). Extensive comparisons also confirm our method's higher accuracy over existing deep learning approaches for crop type classification and the consistent benefits of hyperspectral data across varying temporal windows and crop change scenarios. Codes and dataset are available at https://github.com/flyakon/H2Crop.

Figures

Figures reproduced from arXiv: 2506.06155 by the authors.

Figure 1
Figure 1. Study area, spatial distribution and representative samples of the H [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Flowchart outlining the construction process of the H [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. presents key statistical insights into the H2Crop dataset’s composition [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Comprehensive four-tier hierarchical structure of agricultural taxonomy in the H [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Overall network architecture for hierarchical crop classification. Spectral and spatial transformer to extract [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Performance comparison of four experimental configurations. Radar charts showing precision/ [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Visualization of hierarchical classification results using S2 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Performance comparison of conventional deep learning architectures: Bar charts showing F1 scores across [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Comparative evaluation of classification head architectures. left bars show independent head performance at [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Temporal dynamics of multi-modal data contributions: Line plot showing F1 score evolution across temporal [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Differential effects of hyperspectral data and prior knowledge: (Top) F1 improvements from hyperspectral data without (left) and with (right) prior knowledge; (Bottom) F1 changes from prior knowledge without (left) and with (right) hyperspectral data. 5.2. Relationshi…
Figure 12
Figure 12. Figure 12: Correlation between class distribution and classification performance. Top four line plots shows the correlation [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

73 extracted references · 68 canonical work pages

  1. [1]

    write newline

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

  2. [2]

    write newline

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

  3. [3]

    write newline

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

  4. [4]

    title Pretrain a remote sensing foundation model by promoting intra-instance similarity

    author An [ X.] , author He [ W.] , author Zou [ J.] , author Yang [ G.] , author Zhang [ H.] . title Pretrain a remote sensing foundation model by promoting intra-instance similarity . journal IEEE Transactions on Geoscience and Remote Sensing year 2024

    work page 2024

  5. [5]

    title 220 band aviris hyperspectral image data set: June 12, 1992 indian pine test site 3

    author Baumgardner [ M.] , author Biehl [ L.] , author Landgrebe [ D.] . title 220 band aviris hyperspectral image data set: June 12, 1992 indian pine test site 3 . journal (No Title) year 2015

    work page 1992

  6. [6]

    title Mapping of crop types and crop sequences with combined time series of sentinel-1, sentinel-2 and landsat 8 data for germany

    author Blickensd \"o rfer [ L.] , author Schwieder [ M.] , author Pflugmacher [ D.] , author Nendel [ C.] , author Erasmi [ S.] , author Hostert [ P.] . title Mapping of crop types and crop sequences with combined time series of sentinel-1, sentinel-2 and landsat 8 data for germany . journal Remote sensing of environment year 2022 ; volume 269 : pages 112831

    work page 2022

  7. [7]

    title Random forests

    author Breiman [ L.] . title Random forests . journal Machine learning year 2001 ; volume 45 : pages 5--32

    work page 2001

  8. [8]

    title The enmap spaceborne imaging spectroscopy mission: Initial scientific results two years after launch

    author Chabrillat [ S.] , author Foerster [ S.] , author Segl [ K.] , author Beamish [ A.] , author Brell [ M.] , author Asadzadeh [ S.] , author Milewski [ R.] , author Ward [ K.J.] , author Brosinsky [ A.] , author Koch [ K.] , et al. title The enmap spaceborne imaging spectroscopy mission: Initial scientific results two years after launch . journal Rem...

    work page 2024

Show all 73 references
  1. [9]

    title Adversarial instance augmentation for building change detection in remote sensing images

    author Chen [ H.] , author Li [ W.] , author Shi [ Z.] . title Adversarial instance augmentation for building change detection in remote sensing images . journal IEEE Transactions on Geoscience and Remote Sensing year 2021 ; volume 60 : pages 1--16

    2021

  2. [10]

    title Rsmamba: Remote sensing image classification with state space model

    author Chen [ K.] , author Chen [ B.] , author Liu [ C.] , author Li [ W.] , author Zou [ Z.] , author Shi [ Z.] . title Rsmamba: Remote sensing image classification with state space model . journal IEEE Geoscience and Remote Sensing Letters year 2024 a

    2024

  3. [11]

    title Dynamicvis: An efficient and general visual foundation model for remote sensing image understanding

    author Chen [ K.] , author Liu [ C.] , author Chen [ B.] , author Li [ W.] , author Zou [ Z.] , author Shi [ Z.] . title Dynamicvis: An efficient and general visual foundation model for remote sensing image understanding . journal arXiv preprint arXiv:250316426 year 2025

    2025

  4. [12]

    title Rsprompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model

    author Chen [ K.] , author Liu [ C.] , author Chen [ H.] , author Zhang [ H.] , author Li [ W.] , author Zou [ Z.] , author Shi [ Z.] . title Rsprompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model . journal IEEE Transactions on...

    2024

  5. [13]

    title Time travelling pixels: Bitemporal features integration with foundation model for remote sensing image change detection

    author Chen [ K.] , author Liu [ C.] , author Li [ W.] , author Liu [ Z.] , author Chen [ H.] , author Zhang [ H.] , author Zou [ Z.] , author Shi [ Z.] . title Time travelling pixels: Bitemporal features integration with foundation model for remote sensing image change detect...

    2024

  6. [14]

    title Mapping paddy rice cropping intensity and calendar in monsoon asia at 20 m resolution between 2018 and 2021 from multi-source satellite data using a sample-free algorithm

    author Chen [ Y.] , author Liang [ S.] , author Liu [ J.] , author Ma [ H.] , author Li [ W.] , author Sucharitakul [ P.] , author Luo [ N.] , author Chen [ Z.] , author Fang [ H.] , author Zhang [ F.] , et al. title Mapping paddy rice cropping intensity and calendar in monsoo...

    2018

  7. [15]

    title Using prisma hyperspectral data for land cover classification with artificial intelligence support

    author Delogu [ G.] , author Caputi [ E.] , author Perretta [ M.] , author Ripa [ M.N.] , author Boccia [ L.] . title Using prisma hyperspectral data for land cover classification with artificial intelligence support . journal Sustainability year 2023 ; volume 15 ( number 18 )...

    2023

  8. [16]

    title Improved regularization of convolutional neural networks with cutout

    author DeVries [ T.] , author Taylor [ G.W.] . title Improved regularization of convolutional neural networks with cutout . journal arXiv preprint arXiv:170804552 year 2017

    2017

  9. [17]

    title Early season mapping of winter wheat in china based on landsat and sentinel images

    author Dong [ J.] , author Fu [ Y.] , author Wang [ J.] , author Tian [ H.] , author Fu [ S.] , author Niu [ Z.] , author Han [ W.] , author Zheng [ Y.] , author Huang [ J.] , author Yuan [ W.] . title Early season mapping of winter wheat in china based on landsat and sentinel...

    2020

  10. [18]

    title An image is worth 16x16 words: Transformers for image recognition at scale

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

    2020

  11. [19]

    title A comprehensive review of rice mapping from satellite data: Algorithms, product characteristics and consistency assessment

    author Fang [ H.] , author Liang [ S.] , author Chen [ Y.] , author Ma [ H.] , author Li [ W.] , author He [ T.] , author Tian [ F.] , author Zhang [ F.] . title A comprehensive review of rice mapping from satellite data: Algorithms, product characteristics and consistency ass...

    2024

  12. [20]

    title Generating an annual 30m rice cover product for monsoon asia (2018-2023) using harmonized landsat and sentinel-2 data and the nasa-ibm geospatial foundation model

    author Fang [ H.] , author Liang [ S.] , author Li [ W.] , author Chen [ Y.] , author Ma [ H.] , author Xu [ J.] , author Ma [ Y.] , author He [ T.] , author Tian [ F.] , author Zhang [ F.] , et al. title Generating an annual 30m rice cover product for monsoon asia (2018-2023)...

    2018

  13. [21]

    title Crop type classification by desis hyperspectral imagery and machine learning algorithms

    author Farmonov [ N.] , author Amankulova [ K.] , author Szatm \'a ri [ J.] , author Sharifi [ A.] , author Abbasi-Moghadam [ D.] , author Nejad [ S.M.M.] , author Mucsi [ L.] . title Crop type classification by desis hyperspectral imagery and machine learning algorithms . jou...

    2023

  14. [22]

    title 3d cnns with adaptive temporal feature resolutions

    author Fayyaz [ M.] , author Bahrami [ E.] , author Diba [ A.] , author Noroozi [ M.] , author Adeli [ E.] , author Van Gool [ L.] , author Gall [ J.] . title 3d cnns with adaptive temporal feature resolutions . In: booktitle Proceedings of the IEEE/CVF Conference on Computer ...

    2021

  15. [23]

    title Cat: Center attention transformer with stratified spatial-spectral token for hyperspectral image classification

    author Feng [ J.] , author Wang [ Q.] , author Zhang [ G.] , author Jia [ X.] , author Yin [ J.] . title Cat: Center attention transformer with stratified spatial-spectral token for hyperspectral image classification . journal IEEE Transactions on Geoscience and Remote Sensing...

    2024

  16. [24]

    title Three-dimensional singular spectrum analysis for precise land cover classification from uav-borne hyperspectral benchmark datasets

    author Fu [ H.] , author Sun [ G.] , author Zhang [ L.] , author Zhang [ A.] , author Ren [ J.] , author Jia [ X.] , author Li [ F.] . title Three-dimensional singular spectrum analysis for precise land cover classification from uav-borne hyperspectral benchmark datasets . jou...

    2023

  17. [25]

    title In-season and dynamic crop mapping using 3d convolution neural networks and sentinel-2 time series

    author Gallo [ I.] , author Ranghetti [ L.] , author Landro [ N.] , author La Grassa [ R.] , author Boschetti [ M.] . title In-season and dynamic crop mapping using 3d convolution neural networks and sentinel-2 time series . journal ISPRS Journal of Photogrammetry and Remote S...

    2023

  18. [26]

    title Recurrent neural networks and long short-term memory networks: Tutorial and survey

    author Ghojogh [ B.] , author Ghodsi [ A.] . title Recurrent neural networks and long short-term memory networks: Tutorial and survey . journal arXiv preprint arXiv:230411461 year 2023

    2023

  19. [27]

    title Deep learning techniques for hyperspectral image analysis in agriculture: A review

    author Guerri [ M.F.] , author Distante [ C.] , author Spagnolo [ P.] , author Bougourzi [ F.] , author Taleb-Ahmed [ A.] . title Deep learning techniques for hyperspectral image analysis in agriculture: A review . journal ISPRS Open Journal of Photogrammetry and Remote Sensin...

    2024

  20. [28]

    title Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    author Guo [ D.] , author Yang [ D.] , author Zhang [ H.] , author Song [ J.] , author Zhang [ R.] , author Xu [ R.] , author Zhu [ Q.] , author Ma [ S.] , author Wang [ P.] , author Bi [ X.] , et al. title Deepseek-r1: Incentivizing reasoning capability in llms via reinforcem...

    2025

  21. [29]

    title Nesea-rice10: high-resolution annual paddy rice maps for northeast and southeast asia from 2017 to 2019

    author Han [ J.] , author Zhang [ Z.] , author Luo [ Y.] , author Cao [ J.] , author Zhang [ L.] , author Cheng [ F.] , author Zhuang [ H.] , author Zhang [ J.] , author Tao [ F.] . title Nesea-rice10: high-resolution annual paddy rice maps for northeast and southeast asia fro...

    2017

  22. [30]

    title Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? In: booktitle Proceedings of the IEEE conference on Computer Vision and Pattern Recognition

    author Hara [ K.] , author Kataoka [ H.] , author Satoh [ Y.] . title Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? In: booktitle Proceedings of the IEEE conference on Computer Vision and Pattern Recognition . year 2018 . p. pages 6546--6555

    2018

  23. [31]

    title Deep residual learning for image recognition

    author He [ K.] , author Zhang [ X.] , author Ren [ S.] , author Sun [ J.] . title Deep residual learning for image recognition . In: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition . year 2016 . p. pages 770--778

    2016

  24. [32]

    title Support vector machines

    author Hearst [ M.A.] , author Dumais [ S.T.] , author Osuna [ E.] , author Platt [ J.] , author Scholkopf [ B.] . title Support vector machines . journal IEEE Intelligent Systems and their applications year 1998 ; volume 13 ( number 4 ): pages 18--28

    1998

  25. [33]

    title Mdas: a new multimodal benchmark dataset for remote sensing

    author Hu [ J.] , author Liu [ R.] , author Hong [ D.] , author Camero [ A.] , author Yao [ J.] , author Schneider [ M.] , author Kurz [ F.] , author Segl [ K.] , author Zhu [ X.] . title Mdas: a new multimodal benchmark dataset for remote sensing. earth system science data, 1...

    2023

  26. [34]

    title Pixel shuffling is all you need: spatially aware convmixer for dense prediction tasks

    author Ibrahem [ H.] , author Salem [ A.] , author Kang [ H.S.] . title Pixel shuffling is all you need: spatially aware convmixer for dense prediction tasks . journal Pattern Recognition year 2025 ; volume 158 : pages 111068

    2025

  27. [35]

    title Remote-sensing data and deep-learning techniques in crop mapping and yield prediction: A systematic review

    author Joshi [ A.] , author Pradhan [ B.] , author Gite [ S.] , author Chakraborty [ S.] . title Remote-sensing data and deep-learning techniques in crop mapping and yield prediction: A systematic review . journal Remote Sensing year 2023 ; volume 15 ( number 8 ): pages 2014

    2023

  28. [36]

    title Deep learning

    author LeCun [ Y.] , author Bengio [ Y.] , author Hinton [ G.] . title Deep learning . journal nature year 2015 ; volume 521 ( number 7553 ): pages 436--444

    2015

  29. [37]

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

    author Li [ J.] , author Huang [ X.] , author Tu [ L.] . title Whu-ohs: A benchmark dataset for large-scale hersepctral image classification . journal International Journal of Applied Earth Observation and Geoinformation year 2022 ; volume 113 : pages 103022

    2022

  30. [38]

    title Geographical knowledge-driven representation learning for remote sensing images

    author Li [ W.] , author Chen [ K.] , author Chen [ H.] , author Shi [ Z.] . title Geographical knowledge-driven representation learning for remote sensing images . journal IEEE Transactions on Geoscience and Remote Sensing year 2021 ; volume 60 : pages 1--16

    2021

  31. [39]

    title Agrifm: A multi-source temporal remote sensing foundation model for crop mapping

    author Li [ W.] , author Liang [ S.] , author Chen [ K.] , author Chen [ Y.] , author Ma [ H.] , author Xu [ J.] , author Ma [ Y.] , author Guan [ S.] , author Fang [ H.] , author Shi [ Z.] . title Agrifm: A multi-source temporal remote sensing foundation model for crop mappin...

    2025

  32. [40]

    title Asiawheat: The first asian 250-m annual fractional wheat cover time series (2001-2023) using convolutional neural networks and transformer models

    author Li [ W.] , author Liang [ S.] , author Chen [ Y.] , author Ma [ H.] , author Xu [ J.] , author Chen [ Z.] , author Fang [ H.] , author Zhang [ F.] . title Asiawheat: The first asian 250-m annual fractional wheat cover time series (2001-2023) using convolutional neural n...

    2001

  33. [41]

    title Advances in high-resolution land surface satellite products: A comprehensive review of inversion algorithms, products and challenges

    author Liang [ S.] , author He [ T.] , author Huang [ J.] , author Jia [ A.] , author Zhang [ Y.] , author Cao [ Y.] , author Chen [ X.] , author Chen [ X.] , author Cheng [ J.] , author Jiang [ B.] , et al. title Advances in high-resolution land surface satellite products: A ...

    2024

  34. [42]

    title Diverse hyperspectral remote sensing image synthesis with diffusion models

    author Liu [ L.] , author Chen [ B.] , author Chen [ H.] , author Zou [ Z.] , author Shi [ Z.] . title Diverse hyperspectral remote sensing image synthesis with diffusion models . journal IEEE Transactions on Geoscience and Remote Sensing year 2023 ; volume 61 : pages 1--16

    2023

  35. [43]

    title Swin transformer: Hierarchical vision transformer using shifted windows

    author Liu [ Z.] , author Lin [ Y.] , author Cao [ Y.] , author Hu [ H.] , author Wei [ Y.] , author Zhang [ Z.] , author Lin [ S.] , author Guo [ B.] . title Swin transformer: Hierarchical vision transformer using shifted windows . In: booktitle Proceedings of the IEEE/CVF in...

    2021

  36. [44]

    title Video swin transformer

    author Liu [ Z.] , author Ning [ J.] , author Cao [ Y.] , author Wei [ Y.] , author Zhang [ Z.] , author Lin [ S.] , author Hu [ H.] . title Video swin transformer . In: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . year 2022 . p...

    2022

  37. [45]

    title Fully convolutional networks for semantic segmentation

    author Long [ J.] , author Shelhamer [ E.] , author Darrell [ T.] . title Fully convolutional networks for semantic segmentation . In: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition . year 2015 . p. pages 3431--3440

    2015

  38. [46]

    title A time-weighted dynamic time warping method for land-use and land-cover mapping

    author Maus [ V.] , author C \^a mara [ G.] , author Cartaxo [ R.] , author Sanchez [ A.] , author Ramos [ F.M.] , author De Queiroz [ G.R.] . title A time-weighted dynamic time warping method for land-use and land-cover mapping . journal IEEE Journal of Selected Topics in App...

    2016

  39. [47]

    title Land use and land cover classification with hyperspectral data: A comprehensive review of methods, challenges and future directions

    author Moharram [ M.A.] , author Sundaram [ D.M.] . title Land use and land cover classification with hyperspectral data: A comprehensive review of methods, challenges and future directions . journal Neurocomputing year 2023 ; volume 536 : pages 90--113

    2023

  40. [48]

    title The land parcel identification system: A useful tool to determine the eligibility of agricultural land—but its management could be further improved

    author Owen [ P.W.] , author Milionis [ N.] , author Papatheodorou [ I.] , author Sniter [ K.] , author Viegas [ H.F.] , author Huth [ J.] , author Bortnowschi [ R.] , et al. title The land parcel identification system: A useful tool to determine the eligibility of agricultura...

    2016

  41. [49]

    title A twenty-year dataset of high-resolution maize distribution in china

    author Peng [ Q.] , author Shen [ R.] , author Li [ X.] , author Ye [ T.] , author Dong [ J.] , author Fu [ Y.] , author Yuan [ W.] . title A twenty-year dataset of high-resolution maize distribution in china . journal Scientific Data year 2023 ; volume 10 ( number 1 ): pages 658

    2023

  42. [50]

    title 2018 ieee grss data fusion challenge – fusion of multispectral lidar and hyperspectral data

    author Prasad [ S.] , author Le Saux [ B.] , author Yokoya [ N.] , author Hansch [ R.] . title 2018 ieee grss data fusion challenge – fusion of multispectral lidar and hyperspectral data . year 2020 . https://dx.doi.org/10.21227/jnh9-nz89. :10.21227/jnh9-nz89

    2018 doi

  43. [51]

    title Maps of cropping patterns in china during 2015--2021

    author Qiu [ B.] , author Hu [ X.] , author Chen [ C.] , author Tang [ Z.] , author Yang [ P.] , author Zhu [ X.] , author Yan [ C.] , author Jian [ Z.] . title Maps of cropping patterns in china during 2015--2021 . journal Scientific data year 2022 ; volume 9 ( number 1 ): pages 479

    2015

  44. [52]

    title End-to-end learned early classification of time series for in-season crop type mapping

    author Ru wurm [ M.] , author Courty [ N.] , author Emonet [ R.] , author Lef \`e vre [ S.] , author Tuia [ D.] , author Tavenard [ R.] . title End-to-end learned early classification of time series for in-season crop type mapping . journal ISPRS Journal of Photogrammetry and ...

    2023

  45. [53]

    title Breizhcrops: A time series dataset for crop type mapping

    author Ru wurm [ M.] , author Pelletier [ C.] , author Zollner [ M.] , author Lef \`e vre [ S.] , author K \"o rner [ M.] . title Breizhcrops: A time series dataset for crop type mapping . journal arXiv preprint arXiv:190511893 year 2019

    2019

  46. [54]

    title Eurocrops: The largest harmonized open crop dataset across the european union

    author Schneider [ M.] , author Schelte [ T.] , author Schmitz [ F.] , author K \"o rner [ M.] . title Eurocrops: The largest harmonized open crop dataset across the european union . journal Scientific Data year 2023 ; volume 10 ( number 1 ): pages 612

    2023

  47. [55]

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

    author Segarra [ J.] , author Buchaillot [ M.L.] , author Araus [ J.L.] , author Kefauver [ S.C.] . title Remote sensing for precision agriculture: Sentinel-2 improved features and applications . journal Agronomy year 2020 ; volume 10 ( number 5 ): pages 641

    2020

  48. [56]

    title Agrisen-cog, a multicountry, multitemporal large-scale sentinel-2 benchmark dataset for crop mapping using deep learning

    author Selea [ T.] . title Agrisen-cog, a multicountry, multitemporal large-scale sentinel-2 benchmark dataset for crop mapping using deep learning . journal Remote Sensing year 2023 ; volume 15 ( number 12 ): pages 2980

    2023

  49. [57]

    title Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    author Shi [ W.] , author Caballero [ J.] , author Husz \'a r [ F.] , author Totz [ J.] , author Aitken [ A.P.] , author Bishop [ R.] , author Rueckert [ D.] , author Wang [ Z.] . title Real-time single image and video super-resolution using an efficient sub-pixel convolutiona...

    2016

  50. [58]

    title Meta-weight-net: Learning an explicit mapping for sample weighting

    author Shu [ J.] , author Xie [ Q.] , author Yi [ L.] , author Zhao [ Q.] , author Zhou [ S.] , author Xu [ Z.] , author Meng [ D.] . title Meta-weight-net: Learning an explicit mapping for sample weighting . journal Advances in neural information processing systems year 2019 ...

    2019

  51. [59]

    title The enmap imaging spectroscopy mission towards operations

    author Storch [ T.] , author Honold [ H.P.] , author Chabrillat [ S.] , author Habermeyer [ M.] , author Tucker [ P.] , author Brell [ M.] , author Ohndorf [ A.] , author Wirth [ K.] , author Betz [ M.] , author Kuchler [ M.] , et al. title The enmap imaging spectroscopy missi...

    2023

  52. [60]

    title A sentinel-2 multiyear, multicountry benchmark dataset for crop classification and segmentation with deep learning

    author Sykas [ D.] , author Sdraka [ M.] , author Zografakis [ D.] , author Papoutsis [ I.] . title A sentinel-2 multiyear, multicountry benchmark dataset for crop classification and segmentation with deep learning . journal IEEE Journal of Selected Topics in Applied Earth Obs...

    2022

  53. [61]

    title Crop mapping from image time series: Deep learning with multi-scale label hierarchies

    author Turkoglu [ M.O.] , author D'Aronco [ S.] , author Perich [ G.] , author Liebisch [ F.] , author Streit [ C.] , author Schindler [ K.] , author Wegner [ J.D.] . title Crop mapping from image time series: Deep learning with multi-scale label hierarchies . journal Remote S...

    2021

  54. [62]

    title Worldcereal: a dynamic open-source system for global-scale, seasonal, and reproducible crop and irrigation mapping

    author Van Tricht [ K.] , author Degerickx [ J.] , author Gilliams [ S.] , author Zanaga [ D.] , author Battude [ M.] , author Grosu [ A.] , author Brombacher [ J.] , author Lesiv [ M.] , author Bayas [ J.C.L.] , author Karanam [ S.] , et al. title Worldcereal: a dynamic open-...

    2023

  55. [63]

    title Attention is all you need

    author Vaswani [ A.] , author Shazeer [ N.] , author Parmar [ N.] , author Uszkoreit [ J.] , author Jones [ L.] , author Gomez [ A.N.] , author Kaiser [ .] , author Polosukhin [ I.] . title Attention is all you need . journal Advances in neural information processing systems y...

    2017

  56. [64]

    title Remote sensing for agriculture in the era of industry 5.0--a survey

    author Victor [ N.] , author Maddikunta [ P.K.R.] , author Mary [ D.R.K.] , author Murugan [ R.] , author Chengoden [ R.] , author Gadekallu [ T.R.] , author Rakesh [ N.] , author Zhu [ Y.] , author Paek [ J.] . title Remote sensing for agriculture in the era of industry 5.0--...

    2024

  57. [65]

    title Hyper-es2t: efficient spatial--spectral transformer for the classification of hyperspectral remote sensing images

    author Wang [ W.] , author Liu [ L.] , author Zhang [ T.] , author Shen [ J.] , author Wang [ J.] , author Li [ J.] . title Hyper-es2t: efficient spatial--spectral transformer for the classification of hyperspectral remote sensing images . journal International Journal of Appl...

    2022

  58. [66]

    title Timesen2crop: A million labeled samples dataset of sentinel 2 image time series for crop-type classification

    author Weikmann [ G.] , author Paris [ C.] , author Bruzzone [ L.] . title Timesen2crop: A million labeled samples dataset of sentinel 2 image time series for crop-type classification . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing ye...

    2021

  59. [67]

    title Remote sensing for agricultural applications: A meta-review

    author Weiss [ M.] , author Jacob [ F.] , author Duveiller [ G.] . title Remote sensing for agricultural applications: A meta-review . journal Remote sensing of environment year 2020 ; volume 236 : pages 111402

    2020

  60. [68]

    title The 10-m crop type maps in northeast china during 2017--2019

    author You [ N.] , author Dong [ J.] , author Huang [ J.] , author Du [ G.] , author Zhang [ G.] , author He [ Y.] , author Yang [ T.] , author Di [ Y.] , author Xiao [ X.] . title The 10-m crop type maps in northeast china during 2017--2019 . journal Scientific data year 2021...

    2017

  61. [69]

    title Deep learning in environmental remote sensing: Achievements and challenges

    author Yuan [ Q.] , author Shen [ H.] , author Li [ T.] , author Li [ Z.] , author Li [ S.] , author Jiang [ Y.] , author Xu [ H.] , author Tan [ W.] , author Yang [ Q.] , author Wang [ J.] , et al. title Deep learning in environmental remote sensing: Achievements and challeng...

    2020

  62. [70]

    title Sanet: A self-attention network for agricultural hyperspectral image classification

    author Zhang [ B.] , author Chen [ Y.] , author Li [ Z.] , author Xiong [ S.] , author Lu [ X.] . title Sanet: A self-attention network for agricultural hyperspectral image classification . journal IEEE Transactions on Geoscience and Remote Sensing year 2023 ; volume 62 : pages 1--15

    2023

  63. [71]

    title A review on edge large language models: Design, execution, and applications

    author Zheng [ Y.] , author Chen [ Y.] , author Qian [ B.] , author Shi [ X.] , author Shu [ Y.] , author Chen [ J.] . title A review on edge large language models: Design, execution, and applications . journal ACM Computing Surveys year 2024

    2024

  64. [72]

    title Spatiotemporal fusion in 3d cnns: A probabilistic view

    author Zhou [ Y.] , author Sun [ X.] , author Luo [ C.] , author Zha [ Z.J.] , author Zeng [ W.] . title Spatiotemporal fusion in 3d cnns: A probabilistic view . In: booktitle proceedings of the IEEE/CVF conference on computer vision and pattern recognition . year 2020 . p. pa...

    2020

  65. [73]

    title Deep learning in remote sensing: A comprehensive review and list of resources

    author Zhu [ X.X.] , author Tuia [ D.] , author Mou [ L.] , author Xia [ G.S.] , author Zhang [ L.] , author Xu [ F.] , author Fraundorfer [ F.] . title Deep learning in remote sensing: A comprehensive review and list of resources . journal IEEE geoscience and remote sensing m...

    2017

Pith tools

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