Pith. sign in

REVIEW 5 major objections 6 minor 29 references

Spiking Neural Networks for SAR Interferometric Phase Unwrapping: A Theoretical Framework for Energy-Efficient Processing

T0 review · 5 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 This paper proposes the first spiking-neural-network framework for InSAR phase unwrapping, with a claimed local-minimum convergence guarantee and 30–100x lower processing energy.

desk verdict A genuine first proposal for SNN-based phase unwrapping, but the central reformulation is undefined and the convergence theorem is about the wrong objective; the paper is not ready for peer review. read the letter →

arxiv 2506.20782 v1 pith:PI4VOENX submitted 2025-06-25 cs.NE cs.ETcs.LGeess.SP

Marc Bara This is my paper · ORCID
classification cs.NEcs.ETcs.LGeess.SP
keywords spikingneuralnetworksSARinterferometryphaseunwrappingneuromorphiccomputingenergy-efficientprocessingInspikeencodingconvergenceanalysis
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

Phase unwrapping—deciding how many $2\pi$ cycles to add at each pixel of a radar interferogram—is normally a two-dimensional optimization problem solved by graph or deep-learning methods. The paper argues that it can instead be recast as a temporal sequence task, with unwrapping decisions propagating through the interferogram along a defined traversal, so that the natural dynamics of a spiking neural network do the work. It proposes three spike encodings (rate, temporal, population) for phase, gradient, and coherence, and a three-layer leaky integrate-and-fire architecture with lateral spatial connections and a competitive decision layer. On that basis it claims the network converges to a local minimum of the unwrapping error and that event-driven processing could cut energy use by 30–100×, which matters as missions like NISAR push Earth-observation data volumes toward hundreds of petabytes.

What carries the argument

The load-bearing object is the hierarchical spiking architecture: an encoding layer of $3MN$ neurons producing spike trains for phase, gradient, and coherence; a processing layer of leaky integrate-and-fire neurons with lateral Gaussian connections $w_{ij}=w_0\exp(-d_{ij}^2/2\sigma^2)h(\gamma_i,\gamma_j)$ that enforce spatial continuity; and a decision layer in which five competing neurons per pixel select the wrap count. The convergence mechanism is the combined learning rule $\Delta w_{ij}=\eta_1\,\mathrm{STDP}(\Delta t_{ij})+\eta_2 e_{ij}$, with the error term $e_{ij}$ back-propagated through a surrogate gradient, driving the energy $E_{\mathrm{SNN}}=\sum \|k_{\mathrm{SNN}}-k_{\mathrm{true}}\|^2+\lambda R(w)$ down to a local minimum.

What would settle it

Take a synthetic interferogram with a known pair of residues requiring a branch cut, feed it through the proposed encoding and network, and compare the selected wrap counts with the true ones. If wrap-count error grows as the traversal length increases, or if the network cannot reproduce the branch cut without a specified path, the claimed equivalence and Theorem 1 are falsified. A direct proxy is to check whether the SNN's decisions decrease the energy in equation (2) on paths of increasing length.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that phase unwrapping's spatial continuity constraint can be carried by SNN temporal dynamics instead of an explicit optimization loop. Section 3.1 reformulates the standard gradient-error energy $E=\sum(\nabla_x\phi_a-\nabla_x\phi_w)^2+(\nabla_y\phi_a-\nabla_y\phi_w)^2$ into a temporal sequence modeling task, and the rest of the paper builds the machinery that makes that task run: rate-coded spikes for phase magnitude, temporally coded spikes for gradients, population-coded spikes for coherence, recurrent lateral connections whose weights decay with distance and scale with coherence, and a winner-take-all decision layer that outputs the wrap count $k(x,y)\in\{-2,-1,0,1,2\}$. Theorem 1 states that under bounded inputs and suitable learning rates the architecture converges to a local minimum of the unwrapping error, with a proof sketch based on contraction-mapping and Lyapunov arguments.

Load-bearing premise

The framework stands or falls on the reformulation of two-dimensional phase unwrapping as a one-dimensional temporal sequence of unwrapping decisions; the paper never specifies the traversal pattern or shows that errors at residues do not accumulate, and if that equivalence fails, the convergence theorem is not about phase unwrapping.

Editorial extensions

If this is right

  • Trained on scenes with known ground-truth wrap counts, the network would settle at a local minimum of the unwrapping error, giving a principled spike-based alternative to SNAPHU and minimum-cost-flow solvers.
  • At the claimed 30–100× energy reduction, NISAR-scale InSAR processing could shift from 300–400 W GPUs to milliwatt-scale neuromorphic chips, changing data-center energy budgets for Earth observation.
  • Because cost scales with spike count rather than pixel count, scenes with sparse coherent regions become proportionally cheaper to unwrap.
  • The sequential formulation opens a streaming mode in which large interferograms are unwrapped without loading the full scene into memory.
  • The framework suggests hybrid designs in which SNN proposals are corrected by classical network-flow refinement.

Reading between the lines

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

  • An extension the paper leaves implicit: the one-dimensional traversal is the crux, and a concrete test would be to run the architecture on synthetic interferograms with known residue pairs and measure whether wrap-count errors accumulate along the path; the paper neither specifies the traversal nor bounds this error.
  • The 30–100× energy figure probably depends on scene coherence statistics; dense fringe patterns or widespread low coherence would raise spike activity, so the realistic saving should be benchmarked scene by scene rather than quoted as a single range.
  • The same spike encodings could be carried into multi-baseline and time-series InSAR, where temporal spike codes might directly represent deformation sequences, though the paper does not develop that extension.
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

5 major / 6 minor

Summary. The manuscript proposes a theoretical framework for applying spiking neural networks (SNNs) to interferometric SAR phase unwrapping. It introduces three spike-encoding schemes for wrapped phase, gradients, and coherence; a three-layer hierarchical SNN architecture (encoding, processing, decision); a supervised learning rule combining STDP with surrogate gradients; and theoretical analyses of convergence, computational complexity, and energy efficiency. The central claims are that the architecture converges to a local minimum of the phase unwrapping error function (Theorem 1, Section 4.1) and that it offers energy savings of 30–100× (or 100–1000× in Section 4.3) over conventional processing.

Significance. If rigorously established, the paper would open a new application area for neuromorphic computing in remote sensing and could motivate energy-efficient InSAR processing at scale. The claimed convergence guarantee and the specific reformulation of 2D phase unwrapping as a temporal sequence are potentially important, and the paper correctly identifies a gap in the literature on neural phase unwrapping. However, the significance is contingent on the theoretical claims being properly derived; as they stand, they are asserted rather than demonstrated. The paper also correctly points to real energy concerns in Earth observation data processing, which gives the topic practical relevance.

major comments (5)
  1. [§3.1 and §4.1] The reformulation of the 2D spatial phase unwrapping problem in Eq. (2) as a temporal sequence modeling task is asserted but never defined. The text states that unwrapping decisions propagate 'following a defined traversal pattern' but no traversal pattern is specified, no argument is given that a causal 1D sequence can represent the 2D continuation problem (including residues and branch cuts), and no connection is established between the energies in Eq. (2) and Eq. (12). Because Theorem 1 is stated for E_SNN in Eq. (12), a supervised prediction error, the claimed convergence does not address the phase unwrapping objective unless this missing equivalence is supplied.
  2. [§3.2.1, Eq. (3)] The rate coding formula r_phi(x,y) = r_max * |phi_w(x,y) + pi| / (2*pi) does not preserve phase continuity near the ±pi boundary. For phase values just below pi the encoding rate approaches r_max, while values just above -pi approach zero, so a small physical phase change across the wrap boundary produces a maximal change in spike rate. This directly contradicts the statement that 'this encoding ensures phase continuity near ±pi boundaries' and undermines the proposed encoding's suitability for wrapped phase data.
  3. [§4.1, Theorem 1] Theorem 1 is not proven. The proof sketch asserts that the network dynamics constitute a contraction mapping and that E_SNN decreases monotonically under the update rule, but no Lyapunov function, contraction argument, or explicit conditions on learning rates and input bounds are given. The proof delegates to Chakraborty & Mukhopadhyay [2], which concerns spatio-temporal classification and does not address phase unwrapping or the specific energy E_SNN. Consequently, the convergence guarantee, which is the paper's central theoretical claim, is unsupported.
  4. [§4.3 and Abstract] The energy efficiency claims are inconsistent and underived. The Abstract, Introduction, and Conclusion state potential savings of 30–100×, while Section 4.3 gives a numerical example claiming 100–1000× reduction, and Section 5.1 cites 4–16× for general signal processing. No analytical derivation ties these numbers to the proposed architecture, and the assertion that 'typical interferograms with 30% coherent pixels' produce the required sparsity is not justified with data or a model. These inconsistencies make the energy-efficiency claims unreliable.
  5. [§4.2] The computational complexity argument relies on the assumption that the average spike rate r satisfies r << r_max because of sparse coherence patterns, but no evidence or mechanism is provided that the proposed encoding and network would actually produce such sparsity for phase unwrapping. Without that condition, the claimed complexity advantage over SNAPHU and minimum-norm methods does not follow.
minor comments (6)
  1. [§3.2.1] The notation r_max is used without a definition or typical range in the text; the statement 'typically 100-200 Hz' appears in the text but should be part of the mathematical definition.
  2. [§4.1] The phrase 'appropriate learning rates' and 'bounded input conditions' in Theorem 1 are never made precise, making the theorem unverifiable as stated.
  3. [§5.1] The claim that 'recent benchmarks show 4-16× improvements' is not accompanied by a specific comparison to the proposed architecture, and the relevance to phase unwrapping is unclear.
  4. [References] Reference [2] is cited as an example of 'recent unsupervised SNN methods' and as the basis for the convergence proof, but the cited paper is about supervised spatio-temporal classification, not unsupervised learning or phase unwrapping.
  5. [General] The paper contains no experimental results, which is acceptable for a theoretical contribution only if the theory is fully developed; given the gaps in the theoretical derivations, the lack of empirical validation further weakens the support for the claimed advantages.
  6. [§7 (not present)] A short 'Limitations' subsection is missing; the Discussion mentions implementation challenges, but a formal statement of the assumptions and scope of the theoretical results would improve clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's centrality claims rest on literature review and architectural proposals, not on equations that reduce to their own inputs.

full rationale

The paper does not derive its central claims from its own fitted parameters or self-citations. The main constructive novelty, the spike encoding schemes (Eqs. 3-5) and the hierarchical SNN architecture (Eqs. 6-8), are defined independently of the phase unwrapping objective in Eq. 2. The convergence claim in Theorem 1 is supported only by a proof sketch that references a supervised prediction energy E_SNN (Eq. 12), which is not shown to equal the phase unwrapping energy of Eq. 2; this is a completeness gap, not a circularity, because the theorem's conclusion is not obtained by substituting the claimed result into its own premises. The text explicitly acknowledges the framework is theoretical and that practical implementation faces open challenges, which further supports that no fitted input is being renamed as a prediction. There are no self-citations by the author, and the cited SNN convergence literature is cited as external support rather than as an unexamined premise specific to this paper. The main risk is that the 2D-to-1D temporal reformulation in Section 3.1 is asserted rather than proven, but an unsupported reformulation is not circular reasoning; it is a missing justification. Overall, the derivation chain is not circular, though it is incomplete.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The framework rests on standard SNN machinery, plus several unverified assumptions about InSAR data and hardware. The main load-bearing assumption is that phase unwrapping can be recast as a causal temporal sequence along an unspecified traversal path. The energy claims rely on assumed coherence sparsity and particular neuromorphic power models, and the convergence theorem borrows a general SNN convergence result without deriving the required conditions.

free parameters (7)
  • r_max = 100-200 Hz
    Maximum firing rate in rate coding (Eq. 3), chosen by hand; scales all spike counts and energy estimates.
  • Δt = hand-chosen
    Temporal resolution in gradient coding (Eq. 4); controls precision but value not specified.
  • N_total = hand-chosen
    Population size for coherence coding (Eq. 5); number of neurons per coherence bin.
  • σ = hand-chosen
    Interaction range of lateral connections (Eq. 7); controls spatial smoothing, no fitting or stated value.
  • w_0 = hand-chosen
    Base lateral connection weight (Eq. 7); sets strength of spatial continuity constraint.
  • λ = hand-chosen
    Regularization weight in energy function (Eq. 12); not specified.
  • η1, η2, β = hand-chosen
    Learning rates and surrogate gradient steepness in Eqs. 9-11; no values given.
assumptions (6)
  • standard math Leaky integrate-and-fire neuron model and surrogate gradient training are valid for this task.
    Used in Sections 3.3.2 and 3.4; taken from prior SNN literature without verification for phase unwrapping.
  • domain assumption Convergence analysis of generic recurrent SNNs (Chakraborty and Mukhopadhyay) applies to the proposed architecture and conditions.
    Theorem 1 proof sketch delegates to this reference without establishing bounded-input or learning-rate conditions.
  • ad hoc to paper Phase unwrapping can be solved as a causal temporal sequence along a traversal pattern.
    Section 3.1 states this reformulation; no traversal pattern is defined and no equivalence proof is given.
  • domain assumption Typical InSAR scenes are sparse in coherence, with about 30% coherent pixels and spike rate r much less than r_max.
    Sections 4.2 and 4.3 use this to justify complexity and energy savings; no data are provided.
  • domain assumption Neuromorphic hardware energy constants (23 pJ per spike, 0.1 pJ per neuron per timestep) are representative for this workload.
    Section 4.3 adopts values from Loihi 2 and SpiNNaker2 literature; applicability to the proposed large network is assumed.
  • ad hoc to paper Rate coding preserves phase continuity near the ±π boundary.
    Section 3.2.1 claims this, but Eq. (3) maps -π to 0 Hz and +π to r_max Hz, so the representation is discontinuous across the wrap boundary.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spiking Neural Networks for SAR Interferometric Phase Unwrapping: A Theoretical Framework for Energy-Efficient Processing." pith.science (2026). https://pith.science/paper/PI4VOENX

@misc{pith2026250620782,
  author       = {Pith},
  title        = {Pith review of: Spiking Neural Networks for SAR Interferometric Phase Unwrapping: A Theoretical Framework for Energy-Efficient Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PI4VOENX}},
  note         = {Machine review of arXiv:2506.20782}
}
read the original abstract

We present the first theoretical framework for applying spiking neural networks (SNNs) to synthetic aperture radar (SAR) interferometric phase unwrapping. Despite extensive research in both domains, our comprehensive literature review confirms that SNNs have never been applied to phase unwrapping, representing a significant gap in current methodologies. As Earth observation data volumes continue to grow exponentially (with missions like NISAR expected to generate 100PB in two years) energy-efficient processing becomes critical for sustainable data center operations. SNNs, with their event-driven computation model, offer potential energy savings of 30-100x compared to conventional approaches while maintaining comparable accuracy. We develop spike encoding schemes specifically designed for wrapped phase data, propose SNN architectures that leverage the spatial propagation nature of phase unwrapping, and provide theoretical analysis of computational complexity and convergence properties. Our framework demonstrates how the temporal dynamics inherent in SNNs can naturally model the spatial continuity constraints fundamental to phase unwrapping. This work opens a new research direction at the intersection of neuromorphic computing and SAR interferometry, offering a complementary approach to existing algorithms that could enable more sustainable large-scale InSAR processing.

Figures

Figures reproduced from arXiv: 2506.20782 by the authors.

Figure 1
Figure 1. Hierarchical SNN architecture for phase unwrapping. The system transforms wrapped phase into spike trains [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Spike encoding schemes. (a) Rate coding encodes phase magnitude as spike frequency. (b) Temporal coding [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 26 canonical work pages

  1. [2]

    Heterogeneous recurrent spiking neural network for spatio-temporal classification,

    I. Chakraborty and A. Mukhopadhyay, “Heterogeneous recurrent spiking neural network for spatio-temporal classification,”Frontiers in Neuroscience, vol. 17, p. 994517, 2023

    work page 2023

  2. [1]

    Energy efficient and low-latency spiking neural networks on embedded microcontrollers through spiking activity tuning,

    F. Barchi, G. Zanatta, E. Parisi, A. Bich, A. Acquaviva, and A. Bartolini, “Energy efficient and low-latency spiking neural networks on embedded microcontrollers through spiking activity tuning,”Neural Computing and Applications, vol. 36, pp. 18897–18917, 2024

    work page 2024

  3. [3]

    Two-dimensional phase unwrapping with use of statistical models for cost functions in nonlinear optimization,

    C. W. Chen and H. A. Zebker, “Two-dimensional phase unwrapping with use of statistical models for cost functions in nonlinear optimization,”Journal of the Optical Society of America A, vol. 18, no. 2, pp. 338–351, 2001. 7 Spiking Neural Networks for SAR Interferometric Phase Unwrapping: A Theoretical Framework for Energy-Efficient ProcessingA PREPRINT

    work page 2001

  4. [4]

    Improved phase unwrapping method based on unscented Kalman filter and D-LinkNet,

    Z. Chen, Y . Wang, and H. Zhang, “Improved phase unwrapping method based on unscented Kalman filter and D-LinkNet,”Signal, Image and Video Processing, vol. 18, pp. 6131–6142, 2024

    work page 2024

  5. [5]

    A novel phase unwrapping method based on network programming,

    M. Costantini, “A novel phase unwrapping method based on network programming,”IEEE Transactions on Geoscience and Remote Sensing, vol. 36, no. 3, pp. 813–821, 1998

    work page 1998

  6. [6]

    Backpropagation-based learning tech- niques for deep spiking neural networks: A survey,

    M. Dampfhoffer, S. Shrestha, T. Mesquida, V . Joubert, and M. Mozafari, “Backpropagation-based learning tech- niques for deep spiking neural networks: A survey,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, pp. 11906–11921, 2023

    work page 2023

  7. [7]

    Advancing neuromorphic computing with Loihi: A survey of results and outlook,

    M. Davies, A. Wild, G. Orchard, Y . Sandamirskaya, G. A. F. Guerra, P. Joshi, P. Plank, and S. R. Risbud, “Advancing neuromorphic computing with Loihi: A survey of results and outlook,”Proceedings of the IEEE, vol. 109, no. 5, pp. 911–934, 2021

    work page 2021

  8. [8]

    D. C. Ghiglia and M. D. Pritt,Two-dimensional phase unwrapping: Theory, algorithms, and software. New York: Wiley-Interscience, 1998

    work page 1998

Show all 29 references
  1. [9]

    Satellite radar interferometry: Two-dimensional phase un- wrapping,

    R. M. Goldstein, H. A. Zebker, and C. L. Werner, “Satellite radar interferometry: Two-dimensional phase un- wrapping,”Radio Science, vol. 23, no. 4, pp. 713–720, 1988

    1988

  2. [10]

    SpiNNaker2: A large-scale neuromorphic system for event-based and asynchronous machine learning,

    H. A. Gonzalez, C. Mayr, C. Liu, J. Thommes, Z. Brodeur, T. Ellguth, H. Trensch, M. R ¨uckerl, S. H ¨oppner, and J. Partzsch, “SpiNNaker2: A large-scale neuromorphic system for event-based and asynchronous machine learning,” arXiv:2401.04491, 2024

    2024 arXiv

  3. [11]

    An introduction to neuromorphic computing and its potential impact for unattended ground sensors,

    M. Hill and C. Vineyard, “An introduction to neuromorphic computing and its potential impact for unattended ground sensors,” Sandia National Laboratories, Tech. Rep. SAND2021-10263, 2021

    2021

  4. [12]

    Directly training temporal spiking neural network with sparse surrogate gradient,

    Y . Li, W. Dong, M. Xie, G. Liu, and D. Xu, “Directly training temporal spiking neural network with sparse surrogate gradient,”Neural Networks, vol. 179, p. 106499, 2024

    2024

  5. [13]

    Spike-based signal processing with temporal coding in spiking neural networks,

    Z. Liu, L. Wang, Y . Zhang, and X. Chen, “Spike-based signal processing with temporal coding in spiking neural networks,”Frontiers in Neuroscience, vol. 17, p. 1266003, 2023

    2023

  6. [14]

    Deep learning-enabled spatial phase unwrapping for 3D measurement,

    C. Luo, X. Wang, and H. Li, “Deep learning-enabled spatial phase unwrapping for 3D measurement,” arXiv:2208.03524, 2022

    2022 arXiv

  7. [15]

    Networks of spiking neurons: The third generation of neural network models,

    W. Maass, “Networks of spiking neurons: The third generation of neural network models,”Neural Networks, vol. 10, no. 9, pp. 1659–1671, 1997

    1997

  8. [16]

    Roadmap to neuromorphic computing with emerging technologies,

    A. Mehonic, A. J. Kenyon, M. S. Munde, W. H. Ng, M. Buckwell, L. Montesi, K. Zarudnyi, A. Rabelo Montes, Z. Cheng, C. Luo, I. Valov, R. A. John, O. Simeone, A. Shluger, M. Lanza, H. Abbas, M. Kozicki, and T. Prodromakis, “Roadmap to neuromorphic computing with emerging technol...

    2024

  9. [17]

    Compressed SAR interferometry in the big data era,

    D. H. T. Minh and Y . N. Ngo, “Compressed SAR interferometry in the big data era,”Remote Sensing, vol. 14, no. 2, p. 390, 2022

    2022

  10. [18]

    A long short-term memory for AI applications in spike-based neuro- morphic hardware,

    A. Rao, P. Plank, A. Wild, and W. Maass, “A long short-term memory for AI applications in spike-based neuro- morphic hardware,”Nature Machine Intelligence, vol. 4, no. 5, pp. 467–479, 2022

    2022

  11. [19]

    The NASA-ISRO SAR (NISAR) mission dual-band radar instrument,

    P. A. Rosen, Y . Kim, R. Kumar, T. Misra, R. Bhan, and V . R. Sagi, “The NASA-ISRO SAR (NISAR) mission dual-band radar instrument,” inIEEE International Geoscience and Remote Sensing Symposium, 2024

    2024

  12. [20]

    Phase unwrapping techniques for SAR interferometry: A comprehensive review,

    A. Saini and D. M. V . M. Ramanujam, “Phase unwrapping techniques for SAR interferometry: A comprehensive review,”International Journal of Remote Sensing, vol. 44, no. 12, pp. 3456–3489, 2023

    2023

  13. [21]

    Environmental burden of United States data centers in the artificial intelligence era,

    M. A. B. Siddik, A. Shehabi, and L. Marston, “Environmental burden of United States data centers in the artificial intelligence era,”Environmental Research Letters, vol. 19, no. 3, p. 034047, 2024

    2024

  14. [22]

    PhaseNet 2.0: Phase unwrapping of noisy data based on deep learning approach,

    G. E. Spoorthi, R. K. S. S. Gorthi, and S. Gorthi, “PhaseNet 2.0: Phase unwrapping of noisy data based on deep learning approach,”IEEE Transactions on Image Processing, vol. 29, pp. 4862–4872, 2020

    2020

  15. [23]

    Deep learning spatial phase unwrapping: A comparative review,

    K. Wang, Q. Kemao, J. Di, and J. Zhao, “Deep learning spatial phase unwrapping: A comparative review,” Advanced Photonics Nexus, vol. 1, no. 1, p. 014001, 2022

    2022

  16. [24]

    DENet: A deep neural network architecture for InSAR phase unwrapping,

    Z. Wu, T. Wang, Y . Wang, R. Adler, and C. Wang, “DENet: A deep neural network architecture for InSAR phase unwrapping,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, p. 5224514, 2021

    2021

  17. [25]

    Spiking neural networks: A survey,

    K. Yamazaki, V . K. V o-Ho, D. Bulsara, and N. Le, “Spiking neural networks: A survey,”APSIPA Transactions on Signal and Information Processing, vol. 11, no. 1, p. e16, 2022

    2022

  18. [26]

    Reconsidering the energy efficiency of spiking neural networks,

    M. Yan, R. Xiong, Y . Wang, and C. Li, “Reconsidering the energy efficiency of spiking neural networks,” arXiv:2406.06543, 2024. 8 Spiking Neural Networks for SAR Interferometric Phase Unwrapping: A Theoretical Framework for Energy-Efficient ProcessingA PREPRINT

    2024 arXiv

  19. [27]

    Artificial intelligence in interferometric synthetic aperture radar phase unwrapping: A review,

    L. Zhou, H. Yu, Y . Lan, and M. Xing, “Artificial intelligence in interferometric synthetic aperture radar phase unwrapping: A review,”IEEE Geoscience and Remote Sensing Magazine, vol. 9, no. 2, pp. 10–28, 2021

    2021

  20. [28]

    Direct training high-performance deep spiking neural networks: A review of theories and methods,

    L. Zhou, H. Yu, Y . Lan, S. Gong, and M. Xing, “Direct training high-performance deep spiking neural networks: A review of theories and methods,”Frontiers in Neuroscience, vol. 18, p. 1383844, 2024

    2024

  21. [29]

    Unwrap-Net: A deep neural network-based InSAR phase unwrapping method assisted by airborne LiDAR data,

    Y . Zhou, H. Zhang, Y . Wang, and Z. Chen, “Unwrap-Net: A deep neural network-based InSAR phase unwrapping method assisted by airborne LiDAR data,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 218(B), pp. 102–113, 2024. 9

    2024

Pith tools

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