Pith. sign in

REVIEW 2 major objections 6 minor 31 references

The paper claims that a hybrid inference-engine configuration—ExecuTorch with XNNPACK for VCA and PyTorch for SUnSAL—cuts the end-to-end execution time of the HSB-SV hyperspectral-multispectral fusion pipeline on a Raspberry Pi 5 from 527.1

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 03:28 UTC pith:FZNRAXDE

load-bearing objection Modest, internally consistent speedup for the authors' own HSB-SV fusion code on a Raspberry Pi 5, but the 'quality fully preserved' claim rests only on component-level MAE from one clean synthetic benchmark. the 2 major comments →

arxiv 2607.23136 v1 pith:FZNRAXDE submitted 2026-07-25 eess.IV cs.PF

Optimized Embedded Implementation of Hyperspectral-Multispectral Image Fusion on Raspberry Pi

Salah Eddine Brezini , Okba Bekhelifi , Oussama Mezouar , Chams Eddine Choucha , Sarra Boukhacheba , Fethi Abdelatif Dali This is my paper
classification eess.IV cs.PF
keywords hyperspectral-multispectral fusionhyperspectral super-resolutionembedded systemsRaspberry Pi 5VCASUnSALedge inference optimizationspectral variability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper is trying to establish that a computationally heavy hyperspectral-multispectral fusion algorithm can be made practical on a low-cost embedded platform by splitting its two main stages across different optimized execution engines. It targets the HSB-SV pipeline, which extracts spectral endmember bundles with VCA and then performs sparse unmixing with SUnSAL. On a Raspberry Pi 5, the authors report that using ExecuTorch with the XNNPACK backend for VCA and PyTorch for SUnSAL reduces total runtime from 527.1 ms to 356.7 ms—a 1.48x speedup—with numerical differences from the original NumPy implementation on the order of 1e-6 or lower. A sympathetic reader would care because this suggests that UAV-based hyperspectral remote sensing, which must operate under tight size, weight, and power constraints, could move fusion processing onto the platform itself instead of relying on ground-station computers.

Core claim

The central claim is that the most efficient embedded implementation of HSB-SV is a hybrid: VCA runs through ExecuTorch with the XNNPACK backend, while SUnSAL runs through PyTorch. On the Raspberry Pi 5, this configuration achieves 356.7 ms mean end-to-end execution time versus 527.1 ms for the original NumPy implementation, a 1.48x speedup and a 32.3% reduction. At the component level, ExecuTorch+XNNPACK is fastest for VCA (24.05 ms vs 33.69 ms) and PyTorch is fastest for SUnSAL (170.53 ms vs 261.40 ms). The authors further claim that numerical fidelity is preserved across all conversions, with mean absolute errors of 1e-6 or lower, and they attribute the hybrid's success to XNNPACK's optim

What carries the argument

The key mechanism is the hybrid execution architecture that assigns each algorithmic stage to the accelerator best suited to its computational pattern. VCA is dominated by large matrix multiplications and projections, which map well onto XNNPACK's low-level ARM SIMD kernels, whereas SUnSAL is an iterative sparse-regression solver whose repeated updates benefit from staying inside PyTorch's tensor runtime rather than crossing engine boundaries on every iteration. The paper also introduces an algorithmic simplification inside VCA: it hard-codes the signal-to-noise ratio to 30 dB instead of estimating it, removing the SNR-estimation overhead under the assumption that UAV imagery has higher and

Load-bearing premise

The speedup and accuracy claims rely on VCA's SNR being safely fixed at 30 dB, which the authors justify by assuming UAV imagery has higher and more stable SNR than satellite imagery; if real UAV data deviates from that assumption, VCA's endmember extraction changes and the final fused image may degrade.

What would settle it

Run the optimized HSB-SV pipeline on real UAV or airborne hyperspectral data and compare the fixed-30 dB VCA variant against the original SNR-estimating VCA: if the endmember or abundance mean absolute error grows by more than roughly 1e-6, or if the fused image's reconstruction error against a reference image worsens, the hard-coded SNR premise is false. A simpler test is to sweep the fixed SNR from 15 dB to 45 dB on the existing synthetic dataset and observe whether the end-to-end output error stays flat; an error spike at lower SNR would falsify the claim that the simplification causes no m

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the reported timings hold, the HSB-SV fusion of a 100x100-pixel, 144-band image can complete in about 357 ms on a Raspberry Pi 5, making per-frame real-time or near-real-time fusion on UAV payloads plausible.
  • The hybrid pattern—using a lightweight SIMD-optimized backend for matrix-heavy stages and a native tensor runtime for iterative solvers—can be applied to other linear-algebra-dominated remote sensing pipelines, not just HSB-SV.
  • The numerical fidelity claim (MAE = 1e-6) implies that users can replace NumPy-based reference implementations with inference engines without re-tuning the fusion parameters or degrading output quality.
  • The paper identifies the spectral response function (SRF) estimation step as the next computational bottleneck; reducing or approximating that step would yield further speedups beyond the reported 1.48x, possibly approaching true real-time operation.
  • The fact that PyTorch wins on an Intel PC while the hybrid wins on ARM suggests that embedded optimization results are architecture-specific and should be benchmarked per target CPU rather than assumed portable.

Where Pith is reading between the lines

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

  • The fixed 30 dB SNR assumption is a testable bet: running VCA with the original SNR estimator on real UAV hyperspectral data and comparing the endmember bundles and fused output against the fixed-30 dB version would reveal whether the simplification holds outside the synthetic benchmark.
  • The reported MAE compares the optimized implementation against the original NumPy implementation, not against ground-truth fusion quality; a stronger validation would measure error against a reference sharpened image, as is standard in Wald's protocol.
  • Because SUnSAL still consumes about 170 ms of the 357 ms total, further gains could come from quantizing SUnSAL's internal operations to float16 or from early-exit criteria on the ADMM iterations, although the paper does not explore those paths.
  • The hybrid decomposition is a general design principle: when porting scientific Python code to edge devices, computationally heterogeneous pipelines should be partitioned so that iterative solvers stay in a native tensor runtime and stateless matrix-heavy subroutines go to a SIMD-optimized inference backend.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper presents an optimized embedded implementation of the HSB-SV hyperspectral/multispectral fusion algorithm on a Raspberry Pi 5. The optimization strategy consists of porting the two most computationally intensive components, VCA and SUnSAL, to PyTorch and to edge inference frameworks (ONNX Runtime, ExecuTorch with XNNPACK), and evaluating six configurations on a synthetic Wald-protocol benchmark. The authors report component-level mean absolute errors around 1e-6 between the original NumPy code and the accelerated versions, and a full-pipeline reduction on the Raspberry Pi 5 from 527.1 ms to 356.7 ms (1.48x speedup) for the best hybrid configuration (ExecuTorch+XNNPACK for VCA, PyTorch for SUnSAL). They conclude that fusion quality is fully preserved and that the implementation is suitable for UAV-based edge processing.

Significance. If the claims hold, this is a modest but practically useful engineering contribution: it quantifies the benefit of off-the-shelf inference frameworks for a non-neural linear-algebra remote-sensing pipeline on an ARM edge platform. The speedup arithmetic is internally consistent, the comparison across six configurations and two machines is informative, and the component-level MAE values are encouragingly small. The main limitation is that the paper's strongest qualitative claim—'fully preserving fusion quality'—is supported only by internal component errors and not by any end-to-end quality metric on the fused product, and the fixed-SNR modification to VCA is validated only on a clean synthetic image. These gaps are fixable and do not invalidate the measured speedups, but they do prevent the quality-preservation claim from being accepted as stated.

major comments (2)
  1. [Section IV and Conclusion] The central claim that 'fully preserving fusion quality' is not supported by the reported metrics. Tables I and II compare internal component outputs (VCA endmember bundles and SUnSAL abundance maps) between original and accelerated code, but no metric is reported for the final fused hyperspectral image (e.g., SAM, ERGAS, PSNR, UIQI, or similar) and no comparison against the reference image is given. Since the conclusion explicitly states that 'numerical accuracy was fully preserved, with mean absolute errors on the order of 1e-6 or lower,' the paper should include fused-image quality metrics for the original and optimized pipelines, and ideally for each of the six configurations. Without this, the claim is an extrapolation from intermediate quantities.
  2. [Section III.B] The modification of VCA to fix SNR at 30 dB is an algorithmic change, not merely a numerical port. In the original VCA, the estimated SNR selects the projection strategy, and the extracted endmembers can therefore differ when the true SNR is below the hard-coded threshold. The paper justifies the modification by the qualitative statement that UAV acquisitions have 'higher and more stable SNR' than satellite sensors, but provides no sensitivity analysis and no real-UAV validation. The synthetic benchmark is generated from a clean reference image with Gaussian spatial degradation and no explicitly injected sensor noise, so it likely operates in the high-SNR regime where the fixed and estimated branches coincide. To support the conclusion, the authors should either (a) report the sensitivity of VCA outputs, SUnSAL abundances, and the fused image to true SNR over, say, 10–40 dB, (b) validate
minor comments (6)
  1. [Abstract] Typo: 'dminution' should be 'diminution'.
  2. [Section IV, discussion of Table III] The text says the ExecuTorch+XNNPACK configuration gives '1.4x improvement' on the Pi 5, but the reported times (527.1 ms to 356.7 ms) correspond to a 1.48x speedup. Please make the numbers consistent.
  3. [Tables I and II] The row label 'RT-OT' is ambiguous and appears to be a typo for either 'ORT-ET' or 'ORT-ET' (ONNX Runtime vs ExecuTorch). Define all abbreviations in the table captions or in the text.
  4. [Section III.A] The sentence 'the former difference is due to the half precision (float32) operations of PyTorch' is inaccurate: float32 is single precision, not half precision. Half precision is float16. This matters because the authors attribute the MAE to precision.
  5. [Section IV and experimental setup] Several hyperparameters that affect the results are not reported: the SUnSAL sparsity weight λ, the ADMM internal parameters, the number of VCA runs/bundles, and the set of endmember counts. Without these values, the experiments are not fully reproducible. Please include a short reproducibility paragraph or table.
  6. [Section V] The statement that the implementation is 'well-suited for real-time remote sensing applications on UAV-based systems' is stronger than the evidence: 356.7 ms per 100×100 image is not shown to meet any real-time constraint. Either define the real-time requirement and show the margin, or soften the wording to 'closer to real-time' / 'more suitable for embedded scenarios.'

Circularity Check

0 steps flagged

No circularity found: the speedup is measured empirically, the accuracy check compares ports to the original implementation, and the fixed-SNR assumption is a stated approximation rather than a fitted target.

full rationale

The paper is an empirical porting and benchmarking study: it measures execution times for six implementation configurations and compares the optimized outputs with the original NumPy implementation using mean absolute error. The central reported results (1.48x speedup, 356.7 ms on Raspberry Pi 5) are direct experimental observations, not predictions derived from fitted parameters or from equations that presuppose the result. The only algorithmic modification is fixing VCA's SNR to 30 dB in Section III.B; this is an explicit assumption about UAV acquisition conditions, not a parameter fitted to the benchmark outputs, so it is not a fitted input disguised as a prediction. The self-citations [13] and [14] identify the baseline algorithm and the earlier Raspberry Pi implementation, but the paper's measured speedups and numerical equivalence checks do not reduce to those citations: they would stand or fall regardless of the authors' identity. The conclusion that fusion quality was 'fully preserved' is supported only by component-level MAE on a single synthetic benchmark, and the paper acknowledges the lack of real-UAV validation; this is an external-validity limitation rather than circular reasoning. No equation is defined in terms of the claim it is meant to establish, and no renamed known result is presented as a new derivation. Accordingly, no significant circularity is present.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The optimization claim rests on standard linear-algebra kernels and two algorithmic assumptions (fixed SNR, dataset representativeness). No new physical entities are introduced. The largest uncharged cost is the inference that component-level MAE preserves final image quality, plus unreported solver parameters (lambda, ADMM) and missing artifacts.

free parameters (4)
  • VCA fixed SNR = 30 dB
    Section III.B: the SNR-estimation step of VCA is bypassed and SNR is hard-coded to 30 dB under an unverified UAV-acquisition assumption; no sensitivity analysis is provided.
  • SUnSAL sparsity weight lambda = not reported (inherited from [14])
    Eq. (7) defines the objective with lambda; the paper never states its value or selection procedure, yet it controls the sparse abundances and hence the fusion output.
  • SUnSAL ADMM internal parameters = not reported
    ADMM penalty/stopping criteria and maximum iterations are not given; these affect both runtime (the paper's headline claim) and numerical closeness, so replication requires guessing them.
  • VCA bundle/endmember counts = not reported
    AEEB/VCA outputs depend on the number of endmembers and bundles; these settings are not stated in the paper.
axioms (5)
  • domain assumption LSU model and PSF/SRF degradation relations (Eqs. 1-4)
    The fusion framework assumes pixels are linear mixtures of endmembers and that spatial/spectral degradations are captured by F and R; inherited from hyperspectral unmixing/fusion literature and from authors' prior HSB-SV.
  • standard math VCA and SUnSAL/ADMM behave as described in [20],[23]
    The optimized implementation ports these algorithms without re-deriving them; correctness of the speedup comparison presupposes these references are correctly implemented.
  • domain assumption Benchmark dataset generated under Wald's protocol from REF [11] is representative of real UAV hyperspectral fusion
    All timing/quality claims come from one 100x100, 144-band synthetic scene with seven materials; generalizing to other sensors/scenes is assumed, and the authors themselves note larger/real datasets are needed.
  • ad hoc to paper Numerical closeness of component outputs (MAE ~1e-6) implies preservation of final fusion quality
    The conclusion asserts 'fully preserved' quality from VCA/SUnSAL MAE values; no final-image quality metric is measured, so this inference is the paper's own, unsupported.
  • domain assumption UAV acquisitions have higher and more stable SNR than satellites, so SNR=30 dB is safe
    Section III.B: this premise justifies bypassing SNR estimation; no quantitative support is supplied.

pith-pipeline@v1.3.0-alltime-deepseek · 9920 in / 16150 out tokens · 145037 ms · 2026-08-01T03:28:41.959551+00:00 · methodology

0 comments
read the original abstract

Remote sensing optical images have become central to a wide range of applications. In particular, hyperspectral images, with their high spectral resolution, enable the extraction of rich information about the objects and materials present in the observed scene. Nevertheless, processing such data comes at the expense of a high computational load due to its large data volume, making real-time processing very difficult to achieve. Recently, we proposed an approach to investigate the feasibility of processing such data on a Raspberry Pi by implementing a hyperspectral super-resolution technique, namely HSB-SV. However, the implementation resulted in high computational time. To overcome this limitation, we apply computational optimization techniques based on migrating the most intensive operations to PyTorch and edge inference frameworks such as ONNX Runtime and ExecuTorch with the XNNPACK backend. The proposed optimized implementation is deployed on a Raspberry Pi 5 platform. Experimental results demonstrate a significant reduction in computational time, achieving a 1.48x overall speedup on the Raspberry Pi 5, the total execution time decreases significantly, from 527.1 ms to 356.7 ms corresponding to a 32.3% dminution, while fully preserving the fusion quality, making the approach more suitable for embedded and edge computing scenarios, particularly for UAV-based hyperspectral remote sensing applications.

Figures

Figures reproduced from arXiv: 2607.23136 by Chams Eddine Choucha, Fethi Abdelatif Dali, Okba Bekhelifi, Oussama Mezouar, Salah Eddine Brezini, Sarra Boukhacheba.

Figure 1
Figure 1. Figure 1: Block diagram of the proposed HSB-SV III. COMPUTATIONAL OPTIMIZATION OF HSB-SV FOR EMBEDDED DEPLOYMENT A. Accelerators PyTorch ecosystem: One factor in the success of deep learning is the continuous development of open source frameworks that provide seamless training and deployment on different devices from dedicated GPUs to consumer computers and edge CPUs. PyTorch represents the computations as a dynamic… view at source ↗

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

31 extracted references · 6 canonical work pages

  1. [1]

    A Review of Image Processing Applications based on Raspberry -Pi,

    H. K. Kondaveeti, D. Bandi, S. E. Mathe, S. Vappangi, and M. Subramanian, “A Review of Image Processing Applications based on Raspberry -Pi,” in 2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS) , 2022, pp. 22–28. doi: 10.1109/ICACCS54159.2022.9784958

    arXiv 2022

  2. [2]

    Applications of Raspberry Pi for Precision Agriculture—A Systematic Review,

    A. Joice et al. , “Applications of Raspberry Pi for Precision Agriculture—A Systematic Review,” Agriculture, vol. 15, no. 3, 2025, doi: 10.3390/agriculture15030227

    work page doi:10.3390/agriculture15030227 2025

  3. [3]

    A Review on Unmann ed Aerial Vehicle Remote Sensing: Platforms, Sensors, Data Processing Methods, and Applications,

    Z. Zhang and L. Zhu, “A Review on Unmann ed Aerial Vehicle Remote Sensing: Platforms, Sensors, Data Processing Methods, and Applications,” Drones, vol. 7, no. 6, 2023, doi: 10.3390/drones7060398

    work page doi:10.3390/drones7060398 2023

  4. [4]

    Unmanned Aerial Vehicle for Remote Sensing Applications—A Review,

    H. Yao, R. Qin, and X. Chen, “Unmanned Aerial Vehicle for Remote Sensing Applications—A Review,” Remote Sens., vol. 11, no. 12, 2019, doi: 10.3390/rs11121443

    work page doi:10.3390/rs11121443 2019

  5. [5]

    Applications of Remote Sensing in Precision Agriculture: A Review,

    R. P. Sishodia, R. L. Ray, and S. K. Singh, “Applications of Remote Sensing in Precision Agriculture: A Review,” Remote Sens.., vol. 12, no. 19, 2020, doi: 10.3390/rs12193136

    doi:10.3390/rs12193136 2020

  6. [6]

    Modeling and Unsupervised Unmixing Based on Spectral Variability for Hyperspectral Oceanic Remote Sensing Data with Adjacency Effects,

    Y. Deville et al., “Modeling and Unsupervised Unmixing Based on Spectral Variability for Hyperspectral Oceanic Remote Sensing Data with Adjacency Effects,” Remote Sens.. , vol. 15, no. 18, 2023, doi: 10.3390/rs15184583

    work page doi:10.3390/rs15184583 2023

  7. [7]

    Hyperspectral Oceanic R emote Sensing With Adjacency Effects: From Spectral-Variability-Based Modeling To Performance Of Associated Blind Unmixing Methods,

    Y. Deville et al. , “Hyperspectral Oceanic R emote Sensing With Adjacency Effects: From Spectral-Variability-Based Modeling To Performance Of Associated Blind Unmixing Methods,” in IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, 2019, pp. 282 –285. doi: 10.1109/IGARSS.2019.8898430

    arXiv 2019

  8. [8]

    Partial linear NMF-based unmixing methods for detection and area estimation of photovoltaic panels in urban hyperspectral remote sensing data,

    M. S. Karoui et al., “Partial linear NMF-based unmixing methods for detection and area estimation of photovoltaic panels in urban hyperspectral remote sensing data,” Remote Sen., vol. 11, no. 18, 2019, doi: 10.3390/rs11182164

    work page doi:10.3390/rs11182164 2019

  9. [10]

    Hyperspectral Imagery for Environmental Urban Planning,

    C. Weber et al., “Hyperspectral Imagery for Environmental Urban Planning,” in IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium , 2018, pp. 1628 –1631. doi: 10.1109/IGARSS.2018.8519085

    arXiv 2018

  10. [11]

    Hypersharpening by an NMF-Unmixing-Based Method Addressing Spectral Variability,

    S. E. Brezini, M. S. Karoui, F. Z. Benhalouche, Y. Deville, and A. Ouamri, “Hypersharpening by an NMF-Unmixing-Based Method Addressing Spectral Variability,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1 –5, 2022, doi: 10.1109/LGRS.2021.3072405

    arXiv 2022

  11. [12]

    A Penalization-Based NMF Approach for Hyperspectral Unmixing Addressing Spectral Variability with an Additively - Tuned Mixing Model,

    S. E. Brezini, Y. Deville, M. S. Karoui, F. Z. Benhalouche, and A. Ouamri, “A Penalization-Based NMF Approach for Hyperspectral Unmixing Addressing Spectral Variability with an Additively - Tuned Mixing Model,” in 2021 IEEE International Ge oscience and Remote Sensing Symposium IGARSS , 2021, pp. 3841 –3844. doi: 10.1109/IGARSS47720.2021.9553366

    arXiv 2021

  12. [13]

    Implementation of a Lightweight Hyperspectral and Multispectral Image Fusion Method for Earth Observation on Raspberry,

    Brezini, Salah Eddine, M. Kadiri, A. Meche, Dahmani, Mohammed, and W. Louchene, “Implementation of a Lightweight Hyperspectral and Multispectral Image Fusion Method for Earth Observation on Raspberry,” presented at the 2026 International Conference on Speech, Multimodal and Advanced Communication Systems (ICSMACS), Algiers, Algeria, 2026. Proceedings forthcoming

    2026

  13. [14]

    Hyperspectral and Multispectral Image Fusion with Au tomated Extraction of Image -Based Endmember Bundles and Sparsity -Based Unmixing to Deal with Spectral Variability,

    S. E. Brezini and Y. Deville, “Hyperspectral and Multispectral Image Fusion with Au tomated Extraction of Image -Based Endmember Bundles and Sparsity -Based Unmixing to Deal with Spectral Variability,” Sensors, vol. 23, no. 4, 2023, doi: 10.3390/s23042341

    work page doi:10.3390/s23042341 2023

  14. [15]

    Improving Hypersharpening for WorldView-3 Data,

    M. Selva, L. Santurri, and S. Baronti, “Improving Hypersharpening for WorldView-3 Data,” IEEE Geoscience and Remote Sensing Letters , vol. 16, no. 6, pp. 987 –991, 2019, doi: 10.1109/LGRS.2018.2884087

    arXiv 2019

  15. [16]

    Hyper-sharpening: A first approach on SIM -GA data,

    M. Selva, B. Aiazzi, F. Butera, L. Chiarantini, and S. Baronti, “Hyper-sharpening: A first approach on SIM -GA data,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. , vol. 8, no. 6, pp. 3008 – 3024, 2015, doi: 10.1109/JSTARS.2015.2440092

    arXiv 2015

  16. [17]

    Hyperspectral and Mutlispectral Image Fusion Based on Spectral Library and Sparse Unmixing to Address Spectral Variability,

    S. E. Brezini, “Hyperspectral and Mutlispectral Image Fusion Based on Spectral Library and Sparse Unmixing to Address Spectral Variability,” in 2025 Joint Urban Remote Sensing Event (JURSE), 2025, pp. 1 –4. doi: 10.1109/JURSE60372.2025.11075971

    arXiv 2025

  17. [18]

    Spectral Variability in Hyperspectral Data Unmixing: A comprehensive review,

    R. A. Borsoi et al. , “Spectral Variability in Hyperspectral Data Unmixing: A comprehensive review,” IEEE Geosci. Remote Sens. Mag., vol. 9, no. 4, pp. 223–270, 2021, doi: 10.1109/MGRS.2021.3071158

    arXiv 2021

  18. [19]

    Advances in hyperspectral image unmixing: From algorithmic frameworks to practical applications,

    L. Ren, Z. Han, L. Gao, T. Zhang, R. Wu, and H. Zhang, “Advances in hyperspectral image unmixing: From algorithmic frameworks to practical applications,” Information Geography , vol. 2, no. 1, p. 100035, 2026, doi: https://doi.org/10.1016/j.infgeo.2025.100035

    arXiv 2026

  19. [20]

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

    J. M. P. Nascimento and J. M. Bioucas -Dias, “Vertex component analysis: A fast algorithm to unmix hyperspectral data,” IEEE Transactions on Geoscience and Remote Sensing , vol. 43, no. 4, pp. 898–910, 2005, doi: 10.1109/TGRS.2005.844293

    arXiv 2005

  20. [21]

    Hyperspectral Unmixing With Spectral Variability Using Adaptive Bundles and Double Sparsity,

    T. Uezato, M. Fauvel, and N. Dobigeon, “Hyperspectral Unmixing With Spectral Variability Using Adaptive Bundles and Double Sparsity,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 6, pp. 39 80–3992, 2019, doi: 10.1109/TGRS.2018.2889256

    arXiv 2019

  21. [22]

    Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion,

    N. Yokoya, T. Yairi, and A. Iwasaki, “Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion,” IEEE Transactions on Geoscience and Remote Sensing , vol. 50, no. 2 , pp. 528 –537, 2012, doi: 10.1109/TGRS.2011.2161320

    arXiv 2012

  22. [23]

    Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing,

    J. M. Bioucas-Dias and M. A. T. Figueiredo, “Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing,” in 2nd Workshop on Hyperspectral Image and Si gnal Processing: Evolution in Remote Sensing, WHISPERS 2010 - Workshop Program, Reykjavik, Iceland, 2010, pp. 1–4. doi: 10.1109/WHISPERS.2010.5594963

    arXiv 2010

  23. [24]

    Rasperry Pi 5 Linpack Benchmark

    “Rasperry Pi 5 Linpack Benchmark.” https://www.sci- pi.org.uk/bench/linpack.html

  24. [25]

    PyTorch: An Imperative Style, High - Performance Deep Learning Library,

    A. Paszke et al., “PyTorch: An Imperative Style, High - Performance Deep Learning Library,” in Advances in Neural Information Processing Systems, New York, NY, USA: Curran Associates Inc, Dec. 2019. [Online]. Ava ilable: http://arxiv.org/abs/1912.01703

    Pith/arXiv arXiv 2019

  25. [26]

    PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation,

    J. Ansel et al., “PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation,” in Proceedings of the 29th ACM Internat ional Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2, New York, NY, USA: ACM, Apr. 2024, pp. 929–947. doi: 10.1145/3620665.3640366

    arXiv 2024

  26. [27]

    ExecuTorch -- A Unified PyTorch Solution to Run AI Models On -Device,

    A. Podkopaev et al., “ExecuTorch -- A Unified PyTorch Solution to Run AI Models On -Device,” arXiv Prepr. arXiv2605.08195, 2026, doi: 10.48550/arXiv.2605.08195

    Pith/arXiv arXiv doi:10.48550/arxiv.2605.08195 2026

  27. [28]

    XNNPACK: High -efficiency floating -point neural network inference operators for mobile, server, and Web

    Google, “XNNPACK: High -efficiency floating -point neural network inference operators for mobile, server, and Web.” 2019. [Online]. Available: https://github.com/google/XNNPACK

    2019

  28. [29]

    ONNX Runtime

    Microsoft, “ONNX Runtime”, [Online]. Available: https://onnxruntime.ai

  29. [30]

    On Optimizing Deep Neural Networks Inference on CPUs for Brain-Computer Interfaces using Inference Engines,

    O. Bekhelifi and N. -E. Berrached, “On Optimizing Deep Neural Networks Inference on CPUs for Brain-Computer Interfaces using Inference Engines,” in 2024 IEEE International Symposium on Circuits and Systems (ISCAS), Singapore: IEEE, May 2024, pp. 1–5. doi: 10.1109/ISCAS58744.2024.10558617

    arXiv 2024

  30. [31]

    Insights into resource utilization of code small language models serving with runtime engines and execution providers,

    F. Durán, M. Martinez, P. Lago, and S. Martínez -Fernández, “Insights into resource utilization of code small language models serving with runtime engines and execution providers,” J. Syst. Softw., vol. 230, p. 112574, Dec. 2025, doi: 10.1016/j.jss.2025.112574

    arXiv 2025

  31. [32]

    C. L. Lawson and R. J. Hanson, Solving Least Squares Problems. Society for Industrial and Applied Mathematics, 1995. doi: 10.1137/1.9781611971217

    doi:10.1137/1.9781611971217 1995