Pith. sign in

REVIEW 2 major objections 2 minor 38 references

In-sensor computing on the IMX500 vision sensor processes Earth observation images at 96.68 percent accuracy while fitting inside 8 MB and using 14.19 mJ per inference.

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 · grok-4.3

2026-06-28 16:58 UTC pith:DVPBKDDE

load-bearing objection The paper supplies IMX500-specific energy and latency numbers on EuroSAT but does not test downlink reduction or orbital conditions. the 2 major comments →

arxiv 2606.01271 v1 pith:DVPBKDDE submitted 2026-05-31 cs.CV

Exploiting In-Sensor Computing for Energy-Efficient Earth Observation

Luigi Capogrosso , Pietro Bonazzi , Loris Hoxhaj , Michele Magno This is my paper
classification cs.CV
keywords in-sensor computingTinyMLEarth observationIMX500energy efficiencysatellite dataconvolutional neural networksEuroSAT
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 establishes that shifting convolutional neural network inference directly onto the sensor chip can address the mismatch between massive satellite data collection and limited downlink bandwidth. It adapts several efficient models to the Sony IMX500 platform and reports that they still reach 96.68 percent accuracy on land-cover classification after the necessary memory and quantization changes. The resulting system runs at 17.40 frames per second with low latency and energy draw, moving the first stage of filtering away from the main onboard computer. Readers would care because satellites currently transmit large amounts of low-value imagery; early rejection at the sensor could free bandwidth and power for more useful observations.

Core claim

The central claim is that an end-to-end Earth Observation pipeline built by combining TinyML techniques with the Sony IMX500 Intelligent Vision Sensor maintains 96.68 percent accuracy on the EuroSAT dataset while respecting the sensor's 8 MB memory limit, delivering 17.40 FPS throughput, 27.43 ms latency, 14.19 mJ energy per inference, and 42.26 GMAC/J efficiency, thereby showing that in-sensor processing can offload the primary embedded device and reduce transmission of noisy or irrelevant data.

What carries the argument

The in-sensor computing framework that places optimized ConvNet inference directly on the IMX500 sensor to perform early filtering before data reaches the main processor.

Load-bearing premise

That performance measured on the EuroSAT dataset after platform-specific optimizations will translate to real orbital imagery without extra thermal, radiation, or power constraints from actual satellite integration.

What would settle it

Evaluating the deployed models on a set of actual orbital satellite images and measuring whether accuracy stays near 96.68 percent and energy stays near 14.19 mJ, or attempting full hardware integration and checking for thermal or radiation failures.

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

If this is right

  • Computation is offloaded from the primary embedded device to the sensor itself.
  • Transmission of noisy or irrelevant data over limited downlink bandwidth is reduced.
  • Energy efficiency reaches 42.26 GMAC/J, making the approach viable under satellite power budgets.
  • Standard efficient models remain usable after the memory and quantization steps required by the 8 MB platform limit.

Where Pith is reading between the lines

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

  • If the energy numbers hold under flight conditions, similar sensor-level filtering could be applied to other remote-sensing tasks such as change detection or anomaly spotting.
  • The same approach might reduce the required size or power of the main onboard computer in future satellite designs.
  • Extending the method to multi-spectral or higher-resolution sensors would require repeating the quantization and memory-mapping steps for each new hardware target.

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 / 2 minor

Summary. The manuscript introduces an in-sensor computing framework for Earth Observation that integrates TinyML-optimized ConvNets (SqueezeNet, ShuffleNetV2, MCUNetV1) with the Sony IMX500 sensor. Processing is shifted to the sensor to offload the primary embedded device and reduce downlink of noisy or irrelevant data. On the EuroSAT land-cover classification task the optimized models achieve 96.68% accuracy while fitting within the sensor's 8 MB limit, delivering 17.40 FPS average throughput, 27.43 ms latency, 14.19 mJ per inference, and 42.26 GMAC/J efficiency.

Significance. If the reported hardware metrics generalize beyond EuroSAT, the work would supply concrete, directly measured evidence that in-sensor inference is feasible under tight memory and energy budgets typical of satellite payloads. The absence of fitted parameters or circular reductions in the evaluation strengthens the empirical grounding. The primary value lies in the platform-specific optimization results rather than in any demonstrated end-to-end downlink savings.

major comments (2)
  1. [Abstract] Abstract: the claim that the framework 'effectively mitigating the downlink transmission of noisy or irrelevant data' is unsupported by any experiment; the evaluation consists solely of EuroSAT classification accuracy and IMX500 platform metrics with no measurement of frame filtering, downlink volume reduction, or robustness to orbital degradations.
  2. [Experimental results paragraph] Experimental results paragraph: the headline 96.68% accuracy is presented without error bars, ablation results on the platform-specific optimizations, or explicit confirmation that post-optimization accuracy was measured on held-out data, weakening the assertion of competitive performance under the stated constraints.
minor comments (2)
  1. A diagram clarifying the placement of in-sensor processing relative to the satellite OBC and downlink path would improve readability of the proposed pipeline.
  2. [Abstract] The abstract should state which model attains the 96.68% figure and whether this is the best or an average across the three networks.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and will revise the manuscript accordingly where the concerns are valid.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the framework 'effectively mitigating the downlink transmission of noisy or irrelevant data' is unsupported by any experiment; the evaluation consists solely of EuroSAT classification accuracy and IMX500 platform metrics with no measurement of frame filtering, downlink volume reduction, or robustness to orbital degradations.

    Authors: We agree the claim is unsupported by direct experiments in this work. The evaluation demonstrates in-sensor inference feasibility on EuroSAT but does not measure downlink savings or orbital robustness. We will revise the abstract to remove the unsupported phrasing and instead note that in-sensor processing conceptually enables selective data handling, with end-to-end downlink evaluation identified as future work. revision: yes

  2. Referee: [Experimental results paragraph] Experimental results paragraph: the headline 96.68% accuracy is presented without error bars, ablation results on the platform-specific optimizations, or explicit confirmation that post-optimization accuracy was measured on held-out data, weakening the assertion of competitive performance under the stated constraints.

    Authors: The 96.68% figure is the post-optimization accuracy on the standard held-out EuroSAT test split. We will add error bars from repeated training runs with varied seeds, include an ablation table quantifying accuracy impact of each optimization step (quantization, memory fitting), and explicitly state the test-set evaluation protocol in the revised results section. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical measurements with no derivation chain

full rationale

The paper reports direct hardware measurements (accuracy 96.68%, 17.40 FPS, 27.43 ms latency, 14.19 mJ/inference) of pre-existing ConvNets (SqueezeNet, ShuffleNetV2, MCUNetV1) after platform-specific optimizations on the public EuroSAT dataset. No equations, fitted parameters, predictions, or first-principles derivations appear in the provided text. No self-citations are invoked to justify uniqueness or load-bearing premises. The central claims rest on experimental results rather than any reduction to inputs by construction. This matches the reader's assessment of score 1.0 and contains none of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Work is purely empirical; no mathematical derivation, new constants, or postulated entities are introduced. Model selection and platform constraints are the only implicit choices.

pith-pipeline@v0.9.1-grok · 5796 in / 1009 out tokens · 21534 ms · 2026-06-28T16:58:08.723878+00:00 · methodology

0 comments
read the original abstract

The rapid growth of the satellite industry has driven a significant increase in geospatial data acquisition, highlighting a critical bottleneck: the severe disparity between the volume of collected sensor data and the limited downlink bandwidth available to ground stations. While On-Board Computing (OBC) has helped address this by pre-processing data in orbit, this article further advances the paradigm by introducing an in-sensor computing framework. We present an optimized end-to-end Earth Observation (EO) pipeline tailored for strict computational constraints by integrating TinyML techniques with the Sony IMX500 Intelligent Vision Sensor. Specifically, our approach shifts processing directly to the sensor level, offloading the computation from the primary embedded device, and effectively mitigating the downlink transmission of noisy or irrelevant data. We evaluated several efficient Convolutional Neural Networks (ConvNets), i.e., SqueezeNet, ShuffleNetV2, and MCUNetV1, on the EuroSAT dataset. Experimental results show that, despite the optimizations required for deployment on the IMX500 platform, our models maintain a competitive 96.68% accuracy while operating within its 8 MB constraints. Specifically, the models reach an average processing throughput of 17.40 FPS with a latency of 27.43 ms. Furthermore, our system profile exhibits high energy efficiency, with a low energy footprint of 14.19 mJ per inference and an efficiency rating of 42.26 GMAC/J, demonstrating its viability for in-sensor deployment.

Figures

Figures reproduced from arXiv: 2606.01271 by Loris Hoxhaj, Luigi Capogrosso, Michele Magno, Pietro Bonazzi.

Figure 1
Figure 1. Figure 1 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] 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

38 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Telecommunication Systems for Small Satellites Operating at High Frequencies: A Review.Information, 11(5):258, 2020

    Alessandra Babuscia. Telecommunication Systems for Small Satellites Operating at High Frequencies: A Review.Information, 11(5):258, 2020

    2020

  2. [2]

    PicoSAM2: Low-Latency Segmentation In-Sensor for Edge Vision Appli- cations.IEEE Sensors Conference, 2025

    Pietro Bonazzi, Nicola Farronato, Stefan Zihlmann, Haotong Qin, and Michele Magno. PicoSAM2: Low-Latency Segmentation In-Sensor for Edge Vision Appli- cations.IEEE Sensors Conference, 2025

    2025

  3. [3]

    PicoSAM3: Real-Time In-Sensor Region-of-Interest Segmentation

    Pietro Bonazzi, Nicola Farronato, Stefan Zihlmann, Haotong Qin, and Michele Magno. PicoSAM3: Real-Time In-Sensor Region-of-Interest Segmentation.arXiv preprint arXiv:2603.11917, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  4. [4]

    Tiny- Tracker: Ultra-Fast and Ultra-Low-Power Edge Vision In-Sensor for Gaze Estima- tion.IEEE Sensors, 2023

    Pietro Bonazzi, Thomas R¨ uegg, Sizhen Bian, Yawei Li, and Michele Magno. Tiny- Tracker: Ultra-Fast and Ultra-Low-Power Edge Vision In-Sensor for Gaze Estima- tion.IEEE Sensors, 2023

    2023

  5. [5]

    TinyGLASS: Real-Time Self-Supervised In-Sensor Anomaly Detection,

    Pietro Bonazzi, Rafael Sutter, Luigi Capogrosso, Mischa Buob, and Michele Magno. TinyGLASS: Real-Time Self-Supervised In-Sensor Anomaly Detection. arXiv preprint arXiv:2603.16451, 2026

    work page arXiv 2026

  6. [6]

    The CubeSat mission FUTURE: A preliminary analysis to validate the on-board autonomous orbit determination

    Carmine Buonagura, Salvatore Borgia, Mattia Pugliatti, Alessandro Morselli, Francesco Topputo, Filippo Corradino, Pierluigi Visconti, Luca Deva, Alberto Fedele, Giuseppe Leccese, et al. The CubeSat mission FUTURE: A preliminary analysis to validate the on-board autonomous orbit determination. In12th Inter- national Conference on Guidance, Navigation & Con...

    2023

  7. [7]

    Performance Analysis of Edge and In-Sensor AI Processors: A Comparative Review.arXiv preprint arXiv:2603.08725, 2026

    Luigi Capogrosso, Pietro Bonazzi, and Michele Magno. Performance Analysis of Edge and In-Sensor AI Processors: A Comparative Review.arXiv preprint arXiv:2603.08725, 2026

    work page arXiv 2026

  8. [8]

    A Machine Learning-Oriented Survey on Tiny Machine Learning.IEEE Access, 12:23406–23426, 2024

    Luigi Capogrosso, Federico Cunico, Dong Seon Cheng, Franco Fummi, and Marco Cristani. A Machine Learning-Oriented Survey on Tiny Machine Learning.IEEE Access, 12:23406–23426, 2024

    2024

  9. [9]

    TinyML Enhances CubeSat Mission Capa- bilities.arXiv preprint arXiv:2603.20174, 2026

    Luigi Capogrosso and Michele Magno. TinyML Enhances CubeSat Mission Capa- bilities.arXiv preprint arXiv:2603.20174, 2026

    work page arXiv 2026

  10. [10]

    Opportunities and challenges of on-board AI-based image recognition for small satellite Earth observation missions.Advances in Space Re- search, 75(9):6734–6751, 2025

    Bharadwaj Chintalapati, Arthur Precht, Sougata Hanra, Rene Laufer, Marcus Li- wicki, and Jens Eickhoff. Opportunities and challenges of on-board AI-based image recognition for small satellite Earth observation missions.Advances in Space Re- search, 75(9):6734–6751, 2025. Exploiting In-Sensor Computing for Energy-Efficient Earth Observation 9

    2025

  11. [11]

    Crisp, P.C.E

    N.H. Crisp, P.C.E. Roberts, S. Livadiotti, V.T.A. Oiko, S. Edmondson, S.J. Haigh, C. Huyton, L.A. Sinpetru, K.L. Smith, S.D. Worrall, J. Becedas, R.M. Dom´ ınguez, D. Gonz´ alez, V. Hanessian, A. Mølgaard, J. Nielsen, M. Bisgaard, Y.- A. Chan, S. Fasoulas, G.H. Herdrich, F. Romano, C. Traub, D. Garc´ ıa-Almi˜ nana, S. Rodr´ ıguez-Donaire, M. Sureda, D. Ka...

    2020

  12. [12]

    Mmitigating challenges of the space environment for onboard artificial intelligence: Design overview of the imaging payload on spirit

    Miguel Ortiz Del Castillo, Jonathan Morgan, Jack McRobbie, Clint Therakam, Zaher Joukhadar, Robert Mearns, Simon Barraclough, Richard Sinnott, Andrew Woods, Chris Bayliss, et al. Mmitigating challenges of the space environment for onboard artificial intelligence: Design overview of the imaging payload on spirit. In IEEE/CVF Conference on Computer Vision a...

    2024

  13. [13]

    Review on Hardware Devices and Software Techniques Enabling Neural Network Inference Onboard Satellites.Remote Sens- ing, 16(21):3957, 2024

    Lorenzo Diana and Pierpaolo Dini. Review on Hardware Devices and Software Techniques Enabling Neural Network Inference Onboard Satellites.Remote Sens- ing, 16(21):3957, 2024

    2024

  14. [14]

    Earth+: On-Board Satellite Imagery Compression Leveraging Historical Earth Ob- servations

    Kuntai Du, Yihua Cheng, Peder Olsen, Shadi Noghabi, and Junchen Jiang. Earth+: On-Board Satellite Imagery Compression Leveraging Historical Earth Ob- servations. In30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, 2025

    2025

  15. [15]

    Advancing Earth observation: a survey on AI-powered image processing in satellites.European Journal of Remote Sensing, 58(1), 2025

    Aidan Duggan, Bruno Andrade, and Haithem Afli. Advancing Earth observation: a survey on AI-powered image processing in satellites.European Journal of Remote Sensing, 58(1), 2025

    2025

  16. [16]

    9.6 A 1/2.3inch 12.3Mpixel with On-Chip 4.97TOPS/W CNN Processor Back-Illuminated Stacked CMOS Image Sensor

    Ryoji Eki, Satoshi Yamada, Hiroyuki Ozawa, Hitoshi Kai, Kazuyuki Okuike, Ha- reesh Gowtham, Hidetomo Nakanishi, Edan Almog, Yoel Livne, Gadi Yuval, Eli Zyss, and Takashi Izawa. 9.6 A 1/2.3inch 12.3Mpixel with On-Chip 4.97TOPS/W CNN Processor Back-Illuminated Stacked CMOS Image Sensor. InIEEE Interna- tional Solid- State Circuits Conference (ISSCC), 2021

    2021

  17. [17]

    Artificial intelligence for satellite com- munication: A review.Intelligent and Converged Networks, 2(3):213–243, 2021

    Fares Fourati and Mohamed-Slim Alouini. Artificial intelligence for satellite com- munication: A review.Intelligent and Converged Networks, 2(3):213–243, 2021

    2021

  18. [18]

    Convolutional Neural Networks for On-Board Cloud Screening.Remote Sensing, 11(12):1417, 2019

    Sina Ghassemi and Enrico Magli. Convolutional Neural Networks for On-Board Cloud Screening.Remote Sensing, 11(12):1417, 2019

    2019

  19. [19]

    CloudScout: A Deep Neural Net- work for On-Board Cloud Detection on Hyperspectral Images.Remote Sensing, 12(14):2205, 2020

    Gianluca Giuffrida, Lorenzo Diana, Francesco de Gioia, Gionata Benelli, Gabriele Meoni, Massimiliano Donati, and Luca Fanucci. CloudScout: A Deep Neural Net- work for On-Board Cloud Detection on Hyperspectral Images.Remote Sensing, 12(14):2205, 2020

    2020

  20. [20]

    The Phi-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation.IEEE Transactions on Geoscience and Remote Sensing, 60:1– 14, 2022

    Gianluca Giuffrida, Luca Fanucci, Gabriele Meoni, Matej Batic, Leonie Buckley, Aubrey Dunne, Chris van Dijk, Marco Esposito, John Hefele, Nathan Vercruyssen, Gianluca Furano, Massimiliano Pastena, and Josef Aschbacher. The Phi-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation.IEEE Transactions on Geoscience...

    2022

  21. [21]

    On-Board Image Com- pression using Convolutional Autoencoder: Performance Analysis and Application Scenarios

    Giorgia Guerrisi, Giovanni Schiavon, and Fabio Del Frate. On-Board Image Com- pression using Convolutional Autoencoder: Performance Analysis and Application Scenarios. InIEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2023

    2023

  22. [22]

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 12(7):2217–2226, 2019. 10 L. Capogrosso et al

    2019

  23. [23]

    SqueezeNet: AlexNet-level accuracy with 50x fewer pa- rameters and ¡ 0.5 MB model size

    Forrest N Iandola, Song Han, Matthew W Moskewicz, Khalid Ashraf, William J Dally, and Kurt Keutzer. SqueezeNet: AlexNet-level accuracy with 50x fewer pa- rameters and ¡ 0.5 MB model size. InInternational Conference on Learning Rep- resentations (ICLR), 2017

    2017

  24. [24]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. Adam: A Method for Stochastic Optimization. InInternational Conference for Learning Representations (ICLR), 2015

    2015

  25. [25]

    Vu, and George Goussetis

    Oltjon Kodheli, Eva Lagunas, Nicola Maturo, Shree Krishna Sharma, Bhavani Shankar, Jesus Fabian Mendoza Montoya, Juan Carlos Merlano Duncan, Danilo Spano, Symeon Chatzinotas, Steven Kisseleff, Jorge Querol, Lei Lei, Thang X. Vu, and George Goussetis. Satellite Communications in the New Space Era: A Survey and Future Challenges.IEEE Communications Surveys ...

    2021

  26. [26]

    Satellite Edge Com- puting for Real-Time and Very-High Resolution Earth Observation.IEEE Trans- actions on Communications, 71(10):6180–6194, 2023

    Israel Leyva-Mayorga, Marc Martinez-Gost, Marco Moretti, Ana P´ erez-Neira, Miguel ´Angel V´ azquez, Petar Popovski, and Beatriz Soret. Satellite Edge Com- puting for Real-Time and Very-High Resolution Earth Observation.IEEE Trans- actions on Communications, 71(10):6180–6194, 2023

    2023

  27. [27]

    Douglas Liddle, Antony P

    J. Douglas Liddle, Antony P. Holt, Susan J. Jason, Kathryn A. O’Donnell, and Edward J. Stevens. Space science with CubeSats and nanosatellites.Nature As- tronomy, 4(11):1026–1030, 2020

    2020

  28. [28]

    MCUNet: Tiny Deep Learning on IoT Devices

    Ji Lin et al. MCUNet: Tiny Deep Learning on IoT Devices. InAdvances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  29. [29]

    Machine Learning in Earth Observation Operations: A review

    Pablo Miralles, Antonio Fulvio Scannapieco, Nitya Jagadam, Prerna Baranwal, Bhavin Faldu, Ruchita Abhang, Sahil Bhatia, Sebastien Bonnart, Ishita Bhatnagar, Beenish Batul, et al. Machine Learning in Earth Observation Operations: A review. In72nd International Astronautical Congress (IAC), 2021

    2021

  30. [30]

    Hardware platforms enabling edge ai for space applications: A critical review.IEEE Access, 13:143939–143956, 2025

    Gabriela Mystkowska, Matteo Monopoli, Pietro Nannipieri, Luca Zulberti, David Merodio Codinachs, and Luca Fanucci. Hardware platforms enabling edge ai for space applications: A critical review.IEEE Access, 13:143939–143956, 2025

    2025

  31. [31]

    Earth Observing System Data and Information System (EOSDIS) Handbook, 2018

    National Aeronautics and Space Administration (NASA). Earth Observing System Data and Information System (EOSDIS) Handbook, 2018. Accessed: 2025-10-25

    2018

  32. [32]

    PyTorch: An Imperative Style, High-Performance Deep Learn- ing Library

    Adam Paszke et al. PyTorch: An Imperative Style, High-Performance Deep Learn- ing Library. InAdvances in Neural Information Processing Systems (NeurIPS), 2019

    2019

  33. [33]

    The Raspberry Pi AI Camera

    Raspberry Pi Documentation. The Raspberry Pi AI Camera. https://www.raspberrypi.com/documentation/accessories/ai-camera.html. Ac- cessed: 2026-03-05

    2026

  34. [34]

    Sony Semiconductor Solutions. IMX500. https://developer.sony.com/imx500. Ac- cessed: 2026-03-30

    2026

  35. [35]

    Model Compression Toolkit (MCT)

    Sony Semiconductor Solutions. Model Compression Toolkit (MCT). https://github.com/SonySemiconductorSolutions/mct-model-optimization. Ac- cessed: 2026-03-05

    2026

  36. [36]

    High-Performance On-Orbit Intelligent Computing and Real-Time Services for Remote Sensing Satellites Based on Large-Scale Computing Power in Space

    Guoping Wang, Gang Wan, Zhijuan Su, Yang Wang, Yutong Jia, Gong Li, and Shi Liang. High-Performance On-Orbit Intelligent Computing and Real-Time Services for Remote Sensing Satellites Based on Large-Scale Computing Power in Space. IEEE Access, 13:92114–92133, 2025

    2025

  37. [37]

    ShuffleNet: An Ex- tremely Efficient Convolutional Neural Network for Mobile Devices

    Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, and Jian Sun. ShuffleNet: An Ex- tremely Efficient Convolutional Neural Network for Mobile Devices. InProceedings of the IEEE conference on computer vision and pattern recognition, 2018

    2018

  38. [38]

    Expandable On-Board Real-Time Edge Computing Architecture for Luojia3 Intel- ligent Remote Sensing Satellite.Remote Sensing, 14(15):3596, 2022

    Zhiqi Zhang, Zhuo Qu, Siyuan Liu, Dehua Li, Jinshan Cao, and Guangqi Xie. Expandable On-Board Real-Time Edge Computing Architecture for Luojia3 Intel- ligent Remote Sensing Satellite.Remote Sensing, 14(15):3596, 2022

    2022