Pith. sign in

REVIEW 4 major objections 6 minor 36 references

Lightweight Cloud Masking Models for On-Board Inference in Hyperspectral Imaging

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A 1D CNN shrunk to 597 trainable parameters with a fixed 30-feature PCA input reaches 94.48% cloud accuracy and 81.44% IoU on HYPSO-1, and runs at 802 ms per image on CPU, the authors' recommended on-board model.

desk verdict Useful engineering benchmark with a missing PCA-fit-split disclosure that needs fixing before the headline accuracy claim is taken at face value. read the letter →

arxiv 2507.08052 v1 pith:M4XD7EFL submitted 2025-07-10 cs.CV cs.AIcs.LG

Mazen Ali , António Pereira , Fabio Gentile , Aser Cortines , Sam Mugel , Román Orús , Stelios P. Neophytides , Michalis Mavrovouniotis This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords cloudmaskinghyperspectralimagingon-boardinferencelightweightneuralnetworksmodelcompressionprincipalcomponentanalysissegmentationsatelliteEarthobservation
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

Cloud masking is the preprocessing step that decides whether a satellite image is worth downlinking, and doing it on board would save bandwidth and power. This paper argues that such masking can be done by very small models: it compares gradient-boosting classifiers and 1D convolutional networks on the open HYPSO-1 sea-land-cloud hyperspectral dataset, and reports that every model exceeds 93% classification accuracy. The central result is that singular-value-decomposition compression plus a fixed PCA projection from 112 spectral bands to 30 yields a CNN with only 597 trainable parameters, 94.48% cloud accuracy, and 81.44% cloud IoU, while cutting CPU inference from about 4.8 seconds to 0.8 seconds per image. The authors identify this variant, 1DJuLiNetSingularityF30, as the best overall trade-off for deployment on resource-constrained satellite hardware. If the claim holds, cloud screening becomes a routine on-board function for small satellites rather than a ground-station task.

What carries the argument

The load-bearing object is the 1DJuLiNetSingularityF30 architecture: a one-dimensional CNN whose convolutions run along the 112 spectral bands of each pixel, with layers compressed through singular value decomposition of the convolutional weight tensors, followed by a fixed PCA projection matrix that maps the input to 30 principal components as a non-trainable first layer. The PCA projection is what cuts feature dimensionality and CPU latency; the SVD compression is what reduces trainable parameters from 4563 to 597. The paper reports that the first 30 principal components retain over 99.99% of the variance, which is the quantitative premise linking the reduced input to the retained accuracy.

What would settle it

Retrain 1DJuLiNetSingularityF30 with the PCA fit carried out only on the 30 training images, then evaluate on the 5 test images; if accuracy or IoU drops materially below 94.48% and 81.44%, the fixed-projection pipeline has data leakage. A second check is to benchmark the 20 kB model on the Zynq-7030-class FPGA used by HYPSO-1 and compare measured per-image latency and power against the 802 ms CPU figure.

Watch

Extended reading notes

Core claim

The paper establishes that the spectral-only 1D CNN architecture introduced for HYPSO-1 can be shrunk by two mechanisms without losing its practical value: replacing convolutional layers with SVD-based tensorized layers, and prepending a fixed PCA layer that drops the input from 112 to 30, 18, 7, or 4 dimensions. On the test split, the 30-feature variant reaches 94.48% accuracy and 81.44% Jaccard index for the cloud class, close to the uncompressed retrained CNN's 95.38% and 82.72%, while needing only 597 trainable parameters, 20 kB of storage, and 802 ms per image on a CPU. The paper presents this variant as the best deployment trade-off among all investigated models.

Load-bearing premise

The results assume the principal-component projection was computed without looking at the test images, but the paper never states which images were used for that calculation; if test images helped shape the projection, the reported accuracies would be inflated.

Editorial extensions

If this is right

  • A 20 kB, 597-parameter model fits comfortably in the 8K-parameter on-chip cache of the Zynq-7030-class processor that the HYPSO-1 mission uses, so the claimed compression is compatible with existing flight hardware.
  • At 802 ms per image on a CPU, cloud screening can run as images are acquired, letting the satellite discard cloudy scenes before downlink.
  • The 92.85% precision on the cloud class means few clear-sky pixels are thrown away, which matters when the cost of a false cloud label is losing a useful image.
  • The trend from the 30-feature variant down to the 4-feature variant shows a smooth accuracy-versus-compute frontier, so mission designers can pick an operating point for their power budget.
  • All boosting and CNN variants exceed 93% accuracy, so the paper's conclusion is not tied to one architecture; even the 12-trainable-parameter F04 variant remains usable.

Reading between the lines

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

  • If the PCA projection was fit using test images, the reported 94.48% accuracy would be optimistically biased; the paper does not state the data split used for the variance analysis, so the first decisive check is to refit PCA on the 30 training images only and re-measure the 5 test images.
  • The compression recipe should transfer to other hyperspectral sensors, but the PCA projection is sensor-specific: spectral band positions and calibration differ, so each new instrument would need its own fitted projection or a learned linear layer.
  • Adding spatial context through small 2D convolutions would likely raise the 81.44% cloud IoU, but would also increase latency and parameters; the paper leaves that trade-off unexplored.
  • The high confidence the CNN shows on test images hints that a confidence threshold could serve as a second gate before downlink, a metadata-driven extension the paper motivates but does not implement.
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 evaluates lightweight machine learning models for cloud and cloud shadow masking in hyperspectral satellite imagery, using the HYPSO-1 Sea-Land-Cloud dataset (38 labeled images). It compares XGBoost, LightGBM, a retrained 1D CNN baseline (1DJuLiNet), an SVD-compressed version (1DJuLiNetSingularity), and PCA feature-reduced versions with 30, 18, 7, and 4 features. The main reported result is that 1DJuLiNetSingularityF30, with 597 trainable parameters and a 30-dimensional PCA input, achieves 94.48% accuracy and 81.44% IoU on the cloud class, with 802 ms CPU inference time per image, and is declared the best overall trade-off for on-board deployment. The paper also reports model sizes, training times, inference latencies on CPU/GPU, and per-pixel FLOPS.

Significance. If the reported numbers hold, the paper demonstrates a useful practical result: a sub-600-parameter CNN can reach roughly 94% cloud-masking accuracy with runtime compatible with on-board constraints, and the comparison across boosting and CNN families is informative for edge-AI system design. The paper's strengths include the use of a public labeled dataset, the explicit reporting of parameter counts, model sizes, inference times, and FLOPS, and the hardware-feasibility analysis in Section V. The main caveat is that the PCA projection used by the feature-reduced models is not documented in terms of the data split, which is essential for unbiased evaluation. In addition, several numeric claims in the abstract and conclusions contradict the tables and must be corrected before the paper can be accepted.

major comments (4)
  1. [III.C, IV.C, Fig. 1] The PCA projection is treated as a fixed, non-trainable layer, but the manuscript does not state whether the principal components were computed on the training split only or on the full 38-image dataset. Section IV.C describes the projection matrix as non-trainable and Fig. 1 shows the variance plot without identifying the data split. If the test images were included in the PCA fit, the reported accuracy and IoU for 1DJuLiNetSingularityF30 in Table IV would be optimistically biased because the reduced coordinates would be calibrated to the exact test scenes. Please document the split used for the PCA/SVD fit, or provide the code or configuration, and confirm that all reported test-set metrics are unbiased. The paper's own note in Section IV.D that very high confidence 'may indicate a suboptimal data split' reinforces the need for this documentation.
  2. [Abstract and Section VI] The abstract states 'All boosting and CNN models achieved accuracies exceeding 93%' and Section VI repeats 'all boosting and CNN-based models performed well, achieving over 93 % classification accuracy.' This is contradicted by Table IV, where 1DJuLiNetSingularityF18 achieves 91.60% and 1DJuLiNetSingularityF04 achieves 92.93%. Please correct the abstract and conclusions, and ensure that any related claims in Section V.B (e.g., 'most of the compressed and boosting models surpassing 93%') are made consistent with the tables.
  3. [Section VI vs Table III] The Conclusions state that 1DJuLiNetSingularityF30 'improved training efficiency, completing in 16 min,' but Table III reports 31 min for this model. The Conclusions also attribute '12 training parameters' and '5 kB' to 1DJuLiNetSingularityF07, while Table III reports 63 trainable parameters and 6 kB for F07 and 12 trainable parameters and 5 kB for F04. Please reconcile the conclusion numbers with Table III, since training time and parameter count are among the criteria used to justify the 'best overall choice' claim.
  4. [Section VI, Tables IV and V] The conclusion that 1DJuLiNetSingularityF30 'emerged as the best overall choice' is not derived from any stated selection rule. Considering only the reported metrics, 1DJuLiNetSingularityF07 has 94.00% accuracy and 78.48% IoU (close to F30's 94.48% and 81.44%), yet is 2.9 times faster on CPU (278 ms vs 802 ms) and uses 63 trainable parameters and 6 kB, compared with 597 and 20 kB for F30. Please specify the multi-criteria decision rule (e.g., a required accuracy/IoU threshold and relative weighting of speed and size) or present a Pareto/dominance analysis to justify the 'best overall' claim.
minor comments (6)
  1. [Section IV.E] The text says 'times were consistent accross different images'; the word 'across' is misspelled.
  2. [Section IV.D] The sentence 'The the most populated class is "land"' contains a duplicated definite article; please fix.
  3. [Fig. 1 caption] The caption says 'Histogram of the principal components in decreasing order of amplitude'; this should be 'variance' or 'eigenvalue' rather than 'amplitude', since Fig. 1 plots variance ratio.
  4. [Section IV.E, Table V] The columns 'GPU 1st Im.' and 'GPU Rest' are not defined in the text; please state explicitly that the first image measures cold-start/warm-up time and the rest is the steady-state time.
  5. [Section V.B] The phrase 'XGBoost requires as few as 66 FLOPS and 330 floating-point comparisons' should specify that these are per-pixel values, consistent with Table VI's units.
  6. [Section IV.D] The statement that very high confidence 'may indicate a suboptimal data split' is not elaborated; please clarify whether this refers to a train/test distribution shift, the PCA fitting procedure, or something else, and provide any supporting evidence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical benchmark, and the PCA-split concern is a data-leakage issue rather than a construction-level circularity.

full rationale

This is an empirical benchmarking study with no derivation chain in which a predicted quantity is defined in terms of its fitted input. The closest candidate is the PCA feature-reduction step in Sections III.C and IV.C: the projection matrix is non-trainable and is built using Singularity's PCA tools, and the F30 model's 94.48% accuracy is reported on the held-out test split described in Section IV.B (30 training, 3 validation, 5 test images). The paper does not state whether the PCA fit was computed on the training split only, which is a legitimate data-leakage and correctness concern. However, circularity requires that the output be equivalent to the input by construction, and no equation or procedure in the paper defines the F30 accuracy as the PCA fit itself. The PCA transform is unsupervised, and the model parameters are separately trained on labels. Self-citations to [34] (Orus) and references to the authors' Singularity tool are not load-bearing: compression via SVD and PCA are standard operations, and no uniqueness theorem or derived prediction rests on those citations. The numeric inconsistencies between the abstract/conclusions (all models 'over 93%') and Table IV (91.60% and 92.93% for two variants) undermine general claims but are correctness errors, not circularity. The limitations paragraph explicitly acknowledges the single-dataset scope and lack of real-satellite deployment, which further supports the empirical nature of the claims. Verdict: no significant circularity.

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

The central claims rely on a correctly labeled public dataset, the exclusion of 8 bands as anomalous (inherited from prior work), and the transfer of CNN hyperparameters from [7]. The PCA fitting split is an unstated assumption that could affect the feature-reduced results. Boosting hyperparameters are fitted to the validation set; PCA component counts are chosen by variance thresholds rather than by test-set performance.

free parameters (4)
  • XGBoost hyperparameters = max_depth=5, min_child_weight=3, subsample=0.504, colsample_bytree=0.779, learning_rate=0.258, num_boost_round=695
    Tuned with Optuna on the validation set (Table I).
  • LightGBM hyperparameters = num_leaves=28, min_data_in_leaf=410, num_boost_round=140
    Tuned with Optuna on the validation set (Table II).
  • PCA component count = 30 (also 18, 7, 4 variants)
    Chosen by cumulative variance thresholds (Section III.C); the 30-component model is the recommended one.
  • SVD compression rank = not reported
    The Singularity tool compresses convolutional layers but the rank or truncation threshold is not disclosed (Section III.C).
assumptions (4)
  • domain assumption HYPSO-1 labels are correct
    The reported accuracy is computed against this ground truth; no independent verification is provided.
  • domain assumption 8 spectral bands excluded as anomalous
    The exclusion is inherited from prior studies [29,7] without re-validation.
  • domain assumption CNN hyperparameters from [7] are appropriate here
    The authors kept them identical and trained for two epochs, as in [7].
  • ad hoc to paper PCA fit split is training-only
    The paper does not state which data were used to fit the projection; the claim of unbiased accuracy implicitly assumes no test leakage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lightweight Cloud Masking Models for On-Board Inference in Hyperspectral Imaging." pith.science (2026). https://pith.science/paper/M4XD7EFL

@misc{pith2026250708052,
  author       = {Pith},
  title        = {Pith review of: Lightweight Cloud Masking Models for On-Board Inference in Hyperspectral Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M4XD7EFL}},
  note         = {Machine review of arXiv:2507.08052}
}
read the original abstract

Cloud and cloud shadow masking is a crucial preprocessing step in hyperspectral satellite imaging, enabling the extraction of high-quality, analysis-ready data. This study evaluates various machine learning approaches, including gradient boosting methods such as XGBoost and LightGBM as well as convolutional neural networks (CNNs). All boosting and CNN models achieved accuracies exceeding 93%. Among the investigated models, the CNN with feature reduction emerged as the most efficient, offering a balance of high accuracy, low storage requirements, and rapid inference times on both CPUs and GPUs. Variations of this version, with only up to 597 trainable parameters, demonstrated the best trade-off in terms of deployment feasibility, accuracy, and computational efficiency. These results demonstrate the potential of lightweight artificial intelligence (AI) models for real-time hyperspectral image processing, supporting the development of on-board satellite AI systems for space-based applications.

Figures

Figures reproduced from arXiv: 2507.08052 by the authors.

Figure 1
Figure 1. Histogram of the principal components in decreasing [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Confusion matrices for different models. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison of (a) Ground truth, (b) XGBoost, (c) 1DJuLiNet, (d) 1DJuLiNetSingularity, (e) 1DJuLiNetSingularityF07, and (f) 1DJuLiNetSingularityF30 on the sample representing Qatar. of supporting the required workload. In Table VI, the outcome of this evaluation is reported. From Table VI, it is evident how the compression and feature-reduction layers notably reduce the complexity of the model without significantly a… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of (a) Ground truth, (b) XGBoost, (c) 1DJuLiNet, (d) 1DJuLiNetSingularity, (e) 1DJuLiNetSingularityF07, and (f) 1DJuLiNetSingularityF30 on the sample representing Argentina. storage of up to 8K model parameters when encoded in 4- byte format. They demonstrat…
Figure 5
Figure 5. Figure 5: Comparison of (a) Ground truth, (b) XGBoost, (c) 1DJuLiNet, (d) 1DJuLiNetSingularity, (e) 1DJuLiNetSingularityF07, and (f) 1DJuLiNetSingularityF30 on the sample representing China. floating-point comparisons (see Table VI). These results do not only confirm the feasibi…
Figure 6
Figure 6. Figure 6: Comparison of (a) Ground truth, (b) XGBoost, (c) 1DJuLiNet, (d) 1DJuLiNetSingularity, (e) 1DJuLiNetSingularityF07, and (f) 1DJuLiNetSingularityF30 on the sample representing Iran. 1DJuLiNetSingularityF07 model, counting only 12 training parameters and requiring just 5 …
Figure 7
Figure 7. Figure 7: Comparison of (a) Ground truth, (b) XGBoost, (c) 1DJuLiNet, (d) 1DJuLiNetSingularity, (e) 1DJuLiNetSingularityF07, and (f) 1DJuLiNetSingularityF30 on the sample representing Spain. cantly improved inference speed on CPUs (800 ms per image) while remaining fast on GPUs.…
Figure 8
Figure 8. Figure 8: Diagram of prediction quality of studied models [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages

  1. [7]

    Semantic Segmentation in Satellite Hyperspectral Imagery by Deep Learning

    Jon Alvarez Justo et al. Semantic Segmentation in Satel- lite Hyperspectral Imagery by Deep Learning . 2024. arXiv: 2310.16210 [cs.CV]. URL: https://arxiv.org/ abs/2310.16210

    work page Pith review arXiv 2024

  2. [29]

    An Open Hyperspectral Dataset with Sea-Land-Cloud Ground-Truth from the Hypso- 1 Satellite

    Jon A Justo et al. “An Open Hyperspectral Dataset with Sea-Land-Cloud Ground-Truth from the Hypso- 1 Satellite”. In: 2023 13th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS) . IEEE. 2023, pp. 1–5

    work page 2023

  3. [1]

    Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward

    Devis Tuia et al. “Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward”. In: IEEE Geoscience and Re- mote Sensing Magazine (2024), pp. 2–25. DOI: 10.1109/ MGRS.2024.3425961

    work page arXiv 2024

  4. [2]

    An Introduction to Machine and Deep Learning Methods for Cloud Masking Applications

    Anna Anzalone, Antonio Pagliaro, and Antonio Tu- tone. “An Introduction to Machine and Deep Learning Methods for Cloud Masking Applications”. In: Applied Sciences 14.7 (2024). ISSN : 2076-3417. DOI: 10.3390/ app14072887. URL: https://www.mdpi.com/2076-3417/ 14/7/2887

    work page 2024

  5. [3]

    Sen2Cor for Sentinel-2

    Magdalena Main-Knorn et al. “Sen2Cor for Sentinel-2”. In: Image and Signal Processing for Remote Sensing XXIII. Ed. by Lorenzo Bruzzone. V ol. 10427. Interna- tional Society for Optics and Photonics. SPIE, 2017, p. 1042704. DOI: 10 . 1117 / 12 . 2278218.URL: https : //doi.org/10.1117/12.2278218

    doi:10.1117/12.2278218 2017

  6. [4]

    Fmask 4.0: Im- proved cloud and cloud shadow detection in Landsats 4–8 and Sentinel-2 imagery

    Shi Qiu, Zhe Zhu, and Binbin He. “Fmask 4.0: Im- proved cloud and cloud shadow detection in Landsats 4–8 and Sentinel-2 imagery”. In: Remote Sensing of Environment 231 (2019), p. 111205. ISSN : 0034-4257. DOI: https : / / doi . org / 10 . 1016 / j . rse . 2019 . 05 . 024. URL: https://www.sciencedirect.com/science/article/ pii/S0034425719302172

    work page 2019

  7. [5]

    Improving Cloud Detection with Machine Learning

    Anze Zupanc. Improving Cloud Detection with Machine Learning. Accessed: 2025-04-01. 2017. URL: https:// medium.com/sentinel-hub/improving-cloud-detection- with-machine-learning-c09dc5d7cf13

    work page 2025

  8. [6]

    CloudS2Mask: A novel deep learning approach for improved cloud and cloud shadow masking in Sentinel-2 imagery

    Nicholas Wright et al. “CloudS2Mask: A novel deep learning approach for improved cloud and cloud shadow masking in Sentinel-2 imagery”. In: Remote Sensing of Environment 306 (2024), p. 114122. ISSN : 0034-4257. DOI: https : / / doi . org / 10 . 1016 / j . rse . 2024 . 114122. URL: https://www.sciencedirect.com/science/article/ pii/S0034425724001330

    work page 2024

Show all 36 references
  1. [8]

    Advances in Hyperspectral Im- age and Signal Processing: A Comprehensive Overview of the State of the Art

    Pedram Ghamisi et al. “Advances in Hyperspectral Im- age and Signal Processing: A Comprehensive Overview of the State of the Art”. In: IEEE Geoscience and Remote Sensing Magazine 5.4 (2017), pp. 37–78. DOI: 10.1109/MGRS.2017.2762087

    2017

  2. [9]

    Cloud/shadow detection based on spectral indices for multi/hyperspectral optical remote sensing imagery

    Han Zhai et al. “Cloud/shadow detection based on spectral indices for multi/hyperspectral optical remote sensing imagery”. In: ISPRS Journal of Photogramme- try and Remote Sensing 144 (2018), pp. 235–253. ISSN : 0924-2716. DOI: https://doi.org/10.1016/j.isprsjprs. 2018 . 07 . ...

    2018 doi

  3. [10]

    Comprehensive analysis of Cubesat electrical power systems for efficient energy management

    Usama Shakoor, Mohanad Alayedi, and Ebrahim E. Elsayed. “Comprehensive analysis of Cubesat electrical power systems for efficient energy management”. In: Discover Energy 5.1 (Apr. 2025), p. 9. ISSN : 2730-

    2025

  4. [11]

    Reducing En- ergy Consumption in CubeSat Missions: The Integrated Antenna Approach

    Gulama-Garip Alisher E. Ibrayev et al. “Reducing En- ergy Consumption in CubeSat Missions: The Integrated Antenna Approach”. In: Engineered Science 33 (2025), p. 1315. DOI: 10.30919/es1315

    2025 doi

  5. [12]

    Energy budgeting for CubeSats with an inte- grated FPGA

    Scott Sterling Arnold, Ryan Nuzzaci, and Ann Gordon- Ross. “Energy budgeting for CubeSats with an inte- grated FPGA”. In: 2012 IEEE Aerospace Conference . 2012, pp. 1–14. DOI: 10.1109/AERO.2012.6187240

    2012

  6. [13]

    Optimizing the CubeSat On-Board Computer Power Consumption Un- der Hard Real-Time Constraints

    Oleksandr Liubimov and Ihor Turkin. “Optimizing the CubeSat On-Board Computer Power Consumption Un- der Hard Real-Time Constraints”. In: Integrated Com- puter Technologies in Mechanical Engineering - 2023 . Ed. by Mykola Nechyporuk, V olodymir Pavlikov, and Dmytro Krytskyi. Ch...

    2023

  7. [14]

    The Phi-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation

    Gianluca Giuffrida et al. “The Phi-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation”. In: IEEE Transactions on Geoscience and Remote Sensing 60 (2022), pp. 1–14. DOI: 10.1109/TGRS.2021.3125567

    2022

  8. [15]

    Phi-Sat-2: Onboard AI Apps for Earth Observation

    Alessandro Marin et al. “Phi-Sat-2: Onboard AI Apps for Earth Observation”. en. In: (2021)

    2021

  9. [16]

    Simulation of Multispectral and Hyperspectral EO Products for Onboard Machine Learning Application

    Nicolas Long ´ep´e et al. “Simulation of Multispectral and Hyperspectral EO Products for Onboard Machine Learning Application”. In: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 17 (2024), pp. 17651–17665. DOI: 10 . 1109 / JSTARS.2024.3434437

    2024

  10. [17]

    Artificial Intelligence Based On-Board Image Compression for the Phi-Sat-2 Mission

    Giorgia Guerrisi, Fabio Del Frate, and Giovanni Schi- avon. “Artificial Intelligence Based On-Board Image Compression for the Phi-Sat-2 Mission”. In: IEEE Jour- nal of Selected Topics in Applied Earth Observations and Remote Sensing 16 (2023), pp. 8063–8075. DOI: 10.1109/JSTAR...

    2023

  11. [18]

    CloudScout: A Deep Neural Network for On-Board Cloud Detection on Hyperspec- tral Images

    Gianluca Giuffrida et al. “CloudScout: A Deep Neural Network for On-Board Cloud Detection on Hyperspec- tral Images”. In: Remote Sensing 12.14 (2020). ISSN : 2072-4292. DOI: 10 . 3390 / rs12142205. URL: https : / / www.mdpi.com/2072-4292/12/14/2205

    2020

  12. [19]

    An FPGA-Based Hardware Accelerator for CNNs Inference on Board Satellites: Benchmarking with Myriad 2-Based Solution for the CloudScout Case Study

    Emilio Rapuano et al. “An FPGA-Based Hardware Accelerator for CNNs Inference on Board Satellites: Benchmarking with Myriad 2-Based Solution for the CloudScout Case Study”. In: Remote Sensing 13.8 (2021). ISSN : 2072-4292. DOI: 10 . 3390 / rs13081518. URL: https://www.mdpi.com/...

    2021

  13. [20]

    An Improved Fmask Method for Cloud Detection in GF-6 WFV Based on Spectral- Contextual Information

    Xiaomeng Yang et al. “An Improved Fmask Method for Cloud Detection in GF-6 WFV Based on Spectral- Contextual Information”. In: Remote Sensing 13.23 (2021). ISSN : 2072-4292. DOI: 10 . 3390 / rs13234936. URL: https://www.mdpi.com/2072-4292/13/23/4936

    2021

  14. [21]

    An Effective Cloud Detection Method for Gaofen-5 Images via Deep Learning

    Junchuan Yu et al. “An Effective Cloud Detection Method for Gaofen-5 Images via Deep Learning”. In: Remote Sensing 12.13 (2020). ISSN : 2072-4292. DOI: 10 . 3390 / rs12132106.URL: https : / / www. mdpi . com / 2072-4292/12/13/2106

    2020

  15. [22]

    Cloud Detection in Hyperspectral Images with Atmospheric Column Water Vapor: Appli- cation to PRISMA and A VIRIS-NG Images

    Alexandre Alakian. “Cloud Detection in Hyperspectral Images with Atmospheric Column Water Vapor: Appli- cation to PRISMA and A VIRIS-NG Images”. In: IEEE Transactions on Geoscience and Remote Sensing 62 (2024), pp. 1–24. DOI: 10.1109/TGRS.2024.3369401

    2024

  16. [23]

    Transfer-learning-based cloud detec- tion for Zhuhai-1 satellite hyperspectral imagery

    Wang Jian et al. “Transfer-learning-based cloud detec- tion for Zhuhai-1 satellite hyperspectral imagery”. In: Frontiers in Environmental Science V olume 10 - 2022 (2022). ISSN : 2296-665X. DOI: 10.3389/fenvs.2022. 1039249. URL: https://www.frontiersin.org/journals/ environmen...

    2022 doi

  17. [24]

    Hyperspectral Image Seg- mentation for Optimal Satellite Operations: In-Orbit Deployment of 1D-CNN

    Jon Alvarez Justo et al. “Hyperspectral Image Seg- mentation for Optimal Satellite Operations: In-Orbit Deployment of 1D-CNN”. In: Remote Sensing 17.4 (2025). ISSN : 2072-4292. DOI: 10 . 3390 / rs17040642. URL: https://www.mdpi.com/2072-4292/17/4/642

    2025

  18. [25]

    XGBoost: A Scal- able Tree Boosting System

    Tianqi Chen and Carlos Guestrin. “XGBoost: A Scal- able Tree Boosting System”. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . KDD ’16. Association for Computing Machinery, 2016, 785–794. ISBN : 9781450342322. DOI: 10 . ...

    2016

  19. [26]

    Ac- cessed: 2025-04-01

    Machine Learning Challenge Winning Solutions . Ac- cessed: 2025-04-01. URL: https : / / github. com / dmlc / xgboost/tree/master/demo#machine-learning-challenge- winning-solutions

    2025

  20. [27]

    Accessed: 2025-04-01

    XGBoost. Accessed: 2025-04-01. URL: https://xgboost. readthedocs.io/

    2025

  21. [28]

    Accessed: 2025-04-01

    LightGBM. Accessed: 2025-04-01. URL: https : / / lightgbm.readthedocs.io/

    2025

  22. [30]

    The SVD of Convolutional Weights: A CNN Interpretability Framework

    Brenda Praggastis et al. The SVD of Convolutional Weights: A CNN Interpretability Framework . 2022. arXiv: 2208.06894 [cs.CV]. URL: https://arxiv.org/ abs/2208.06894

    2022 arXiv

  23. [31]

    Optuna: A Next-generation Hyper- parameter Optimization Framework

    Takuya Akiba et al. Optuna: A Next-generation Hyper- parameter Optimization Framework. 2019. arXiv: 1907. 10902 [cs.LG]. URL: https : / / arxiv. org / abs / 1907 . 10902

    2019

  24. [32]

    HYPSO-1 CubeSat: First Images and In-Orbit Characterization

    Sivert Bakken et al. “HYPSO-1 CubeSat: First Images and In-Orbit Characterization”. In: Remote Sensing 15.3 (2023). ISSN : 2072-4292. DOI: 10 . 3390 / rs15030755. URL: https://www.mdpi.com/2072-4292/15/3/755

    2023

  25. [33]

    Design of a hyperspectral im- ager using COTS optics for small satellite applications

    Elizabeth Prentice et al. “Design of a hyperspectral im- ager using COTS optics for small satellite applications”. In: June 2021. DOI: 10.1117/12.2599937

    2021 doi

  26. [34]

    A practical introduction to tensor net- works: Matrix product states and projected entangled pair states

    Rom ´an Or ´us. “A practical introduction to tensor net- works: Matrix product states and projected entangled pair states”. In: Annals of Physics 349 (2014), pp. 117–

    2014

  27. [158]

    DOI: https://doi.org/10.1016/j

    ISSN : 0003-4916. DOI: https://doi.org/10.1016/j. aop.2014.06.013. URL: https://www.sciencedirect.com/ science/article/pii/S0003491614001596

    2014 doi

  28. [7719]

    URL: https: //doi.org/10.1007/s43937-025-00069-5

    DOI: 10.1007/s43937-025-00069-5. URL: https: //doi.org/10.1007/s43937-025-00069-5

    doi

Pith tools

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