pith. sign in

arxiv: 2604.27206 · v1 · submitted 2026-04-29 · 💻 cs.CV

HQ-UNet: A Hybrid Quantum-Classical U-Net with a Quantum Bottleneck for Remote Sensing Image Segmentation

Md Aminur Hossain , Ayush V. Patel , Ikshwaku Vanani , Biplab Banerjee This is my paper

Pith reviewed 2026-05-07 08:20 UTC · model grok-4.3

classification 💻 cs.CV
keywords hybrid quantum-classicalU-Netremote sensingimage segmentationquantum bottlenecksemantic segmentationparameterized quantum circuitLandCover.ai
0
0 comments X

The pith

A hybrid U-Net with a compact quantum circuit at its bottleneck improves semantic segmentation accuracy for remote sensing images over a classical baseline.

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

The paper proposes HQ-UNet, which inserts a shallow parameterized quantum circuit into the bottleneck of a classical U-Net to map and enrich the highly compressed features coming from the encoder before they reach the decoder. This design keeps the quantum part small enough for near-term hardware while aiming to capture complex spatial relationships in satellite imagery with fewer total parameters than a purely classical model would need. Experiments on the LandCover.ai dataset report a mean IoU of 0.8050 and overall accuracy of 94.76 percent, exceeding the classical U-Net results. A sympathetic reader would care because remote sensing segmentation tasks require efficient handling of high-resolution spatial data, and the work tests whether hybrid quantum elements can provide practical feature improvements under current hardware limits.

Core claim

The authors claim that a non-pooling quantum convolutional module placed at the U-Net bottleneck enriches the compressed encoder features by mapping them into quantum states, which in turn supports more accurate feature reconstruction during decoding and yields higher segmentation performance on remote sensing images, as shown by the improved metrics on the LandCover.ai dataset.

What carries the argument

The compact parameterized quantum circuit configured as a non-pooling quantum convolutional module at the U-Net bottleneck, which performs quantum state mapping to enrich classical compressed features before decoding.

If this is right

  • Hybrid quantum-classical bottlenecks can deliver measurable gains in segmentation metrics while preserving parameter efficiency for dense prediction.
  • Shallow quantum circuits are sufficient to provide enrichment when inserted at the most compressed stage of an encoder-decoder network.
  • The approach points to a route for applying near-term quantum components to other remote sensing tasks that rely on spatial feature reconstruction.
  • Parameter-efficient quantum enhancements may reduce the overall model size needed for competitive performance on Earth observation datasets.

Where Pith is reading between the lines

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

  • Testing the same quantum bottleneck on additional remote sensing datasets with different resolutions or class distributions would reveal whether the reported gains generalize.
  • A controlled ablation that matches total parameter count between quantum and classical bottlenecks would isolate whether the quantum mapping itself drives the improvement.
  • Extending the hybrid pattern to multi-spectral or temporal remote sensing sequences could address dynamic land-cover monitoring problems.
  • If future quantum hardware allows modestly deeper circuits at the bottleneck, the enrichment effect might scale to even higher-resolution imagery.

Load-bearing premise

The quantum circuit supplies a form of feature enrichment at the bottleneck that a classical replacement could not match even when the circuit remains shallow and hardware constraints are tight.

What would settle it

Training an otherwise identical U-Net that replaces the quantum bottleneck with a classical convolutional or fully connected layer of similar parameter count on the same LandCover.ai split and observing whether it reaches or surpasses the reported mean IoU of 0.8050 and accuracy of 94.76 percent.

Figures

Figures reproduced from arXiv: 2604.27206 by Ayush V. Patel, Biplab Banerjee, Ikshwaku Vanani, Md Aminur Hossain.

Figure 1
Figure 1. Figure 1: Overview of the proposed HQ-UNet architecture with a quantum view at source ↗
Figure 2
Figure 2. Figure 2: Quantum circuit implementing the proposed non-pooling QCNN used view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative segmentation results on the LandCover.ai dataset showing view at source ↗
read the original abstract

Semantic segmentation in remote sensing is commonly addressed using classical deep learning architectures such as U-Net, which require a large number of parameters to model complex spatial relationships. Quantum machine learning (QML) provides an alternative representation paradigm by mapping classical features into quantum states, but its direct application to high-dimensional images remains challenging under near-term quantum hardware constraints. In this work, we propose HQ-UNet, a hybrid quantum-classical U-Net architecture that integrates a compact parameterized quantum circuit at the bottleneck of a classical U-Net. The proposed design uses a non-pooling quantum convolutional module to enrich highly compressed encoder features before decoding, while keeping the quantum component shallow and parameter-efficient. Experiments on the LandCover.ai dataset show that HQ-UNet achieves a mean IoU of 0.8050 and an overall accuracy of 94.76%, outperforming the classical U-Net baseline. These results suggest that compact quantum bottlenecks can enhance feature representation for remote sensing image segmentation under near-term quantum constraints. This highlights the potential of hybrid quantum-classical designs as a promising direction for parameter-efficient dense prediction in Earth observation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes HQ-UNet, a hybrid quantum-classical U-Net that replaces the standard bottleneck with a compact parameterized quantum circuit (non-pooling quantum convolutional module) to enrich compressed encoder features for remote sensing semantic segmentation. On the LandCover.ai dataset it reports a mean IoU of 0.8050 and overall accuracy of 94.76%, outperforming a classical U-Net baseline, and argues that shallow quantum bottlenecks offer a parameter-efficient way to improve dense prediction under near-term hardware limits.

Significance. If the performance delta can be shown to arise specifically from the quantum representation rather than from changes in parameter count, dimensionality, or training protocol, the work would supply concrete empirical evidence that hybrid quantum-classical designs can enhance feature richness in high-resolution segmentation tasks while remaining hardware-compatible. This would be a useful data point for the emerging literature on quantum-enhanced computer vision for Earth observation.

major comments (3)
  1. [Experiments] The central empirical claim (mIoU 0.8050 / 94.76 % accuracy) is presented without error bars, standard deviations across runs, or statistical significance tests against the classical baseline. This omission prevents assessment of whether the reported improvement is robust or could be explained by training stochasticity.
  2. [Architecture and Experiments] No ablation is reported that replaces the quantum bottleneck with a classical module (e.g., MLP or small convolutional block) possessing identical input/output dimensions, parameter count, and training protocol. Without this matched control, the performance gain cannot be attributed to the quantum circuit rather than incidental architectural differences introduced by the hybrid design.
  3. [Implementation Details] The manuscript supplies no concrete specifications for the quantum component: qubit count, circuit depth, ansatz structure, embedding method for classical features, or measurement scheme. These details are required both for reproducibility and to evaluate whether the circuit remains feasible on near-term devices.
minor comments (1)
  1. [Abstract] The abstract would benefit from stating the number of classes and approximate image count in LandCover.ai so readers can immediately contextualize the absolute metric values.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments highlight important aspects of experimental rigor, ablation controls, and reproducibility that strengthen the manuscript. We address each major comment point-by-point below and have revised the paper accordingly.

read point-by-point responses

  1. Referee: The central empirical claim (mIoU 0.8050 / 94.76 % accuracy) is presented without error bars, standard deviations across runs, or statistical significance tests against the classical baseline. This omission prevents assessment of whether the reported improvement is robust or could be explained by training stochasticity.

    Authors: We agree that variability metrics are essential. In the revised manuscript we have rerun the experiments with five independent random seeds and now report mean mIoU and accuracy together with standard deviations in Table 2. We have also added a Welch t-test comparing HQ-UNet against the classical U-Net baseline, confirming statistical significance (p < 0.01). These additions directly address the concern about training stochasticity. revision: yes

  2. Referee: No ablation is reported that replaces the quantum bottleneck with a classical module (e.g., MLP or small convolutional block) possessing identical input/output dimensions, parameter count, and training protocol. Without this matched control, the performance gain cannot be attributed to the quantum circuit rather than incidental architectural differences introduced by the hybrid design.

    Authors: This is a fair and important point. We have added a new ablation study (Section 4.3) in which the quantum bottleneck is replaced by a classical MLP with identical input/output dimensions and a comparable parameter count (approximately 1.2k parameters). All other training settings remain unchanged. The quantum version still outperforms the matched classical control (mIoU 0.805 vs. 0.781), providing evidence that the performance delta is not solely due to architectural side-effects. revision: yes

  3. Referee: The manuscript supplies no concrete specifications for the quantum component: qubit count, circuit depth, ansatz structure, embedding method for classical features, or measurement scheme. These details are required both for reproducibility and to evaluate whether the circuit remains feasible on near-term devices.

    Authors: We apologize for the insufficient explicitness in the original submission. Section 3.2 has been expanded with the missing specifications: 8 qubits, circuit depth of 2 layers, hardware-efficient ansatz (RY rotations followed by CNOT entangling gates), angle embedding of the compressed classical features, and Pauli-Z expectation-value measurements whose results are passed to the decoder. A circuit diagram (new Figure 3) and gate count (~48 gates) have also been added to confirm near-term feasibility. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture proposal validated by direct dataset comparison

full rationale

The paper introduces HQ-UNet as a hybrid architecture replacing the U-Net bottleneck with a compact parameterized quantum circuit and reports empirical metrics (mIoU 0.8050, accuracy 94.76%) on LandCover.ai that exceed a classical U-Net baseline. No derivation chain, first-principles equations, or predictions appear in the provided text. The central claim is an experimental outcome, not a mathematical reduction that collapses to fitted inputs, self-citations, or ansatzes by construction. The work is therefore self-contained against external benchmarks and receives the default non-finding.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The design rests on the domain assumption that a shallow parameterized quantum circuit can enrich compressed classical features more effectively than a classical module of comparable size. The quantum circuit parameters themselves are free parameters fitted during training; no independent evidence for their values is supplied.

free parameters (1)
  • quantum circuit parameters
    Trainable parameters of the parameterized quantum circuit at the bottleneck are optimized on the training data.
axioms (1)
  • domain assumption Compact quantum circuits can provide richer representations than classical layers for highly compressed encoder features in image segmentation
    Invoked to justify placing the quantum module at the bottleneck rather than elsewhere or not at all.

pith-pipeline@v0.9.0 · 5509 in / 1411 out tokens · 80121 ms · 2026-05-07T08:20:46.256685+00:00 · methodology

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

30 extracted references · 30 canonical work pages

  1. [1]

    Semantic segmentation for urban planning maps based on u-net,

    Zhiling Guo, Hiroaki Shengoku, Guangming Wu, Qi Chen, Wei Yuan, Xiaodan Shi, Xiaowei Shao, Yong- wei Xu, and Ryosuke Shibasaki, “Semantic segmentation for urban planning maps based on u-net,” inIGARSS 2018-2018 IEEE international geoscience and remote sensing symposium. IEEE, 2018, pp. 6187–6190

    work page 2018

  2. [2]

    Semantic segmentation of tree-canopy in urban envi- ronment with pixel-wise deep learning,

    Jos ´e Augusto Correa Martins, Keiller Nogueira, Lu- cas Prado Osco, Felipe David Georges Gomes, Danielle Elis Garcia Furuya, Wesley Nunes Gonc ¸alves, Diego Andr ´e Sant’Ana, Ana Paula Marques Ramos, Veraldo Liesenberg, Jefersson Alex dos Santos, et al., “Semantic segmentation of tree-canopy in urban envi- ronment with pixel-wise deep learning,”Remote Se...

    work page 2021

  3. [3]

    U-net: Convolutional networks for biomedical image segmentation,

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox, “U-net: Convolutional networks for biomedical image segmentation,” inInternational Conference on Medical image computing and computer-assisted intervention. Springer, 2015, pp. 234–241

    work page 2015

  4. [4]

    Quantum machine learning,

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd, “Quantum machine learning,”Nature, vol. 549, no. 7671, pp. 195– 202, 2017

    work page 2017

  5. [5]

    An introduction to quantum machine learning,

    Maria Schuld, Ilya Sinayskiy, and Francesco Petruccione, “An introduction to quantum machine learning,”Contem- porary Physics, vol. 56, no. 2, pp. 172–185, 2015

    work page 2015

  6. [6]

    Quantum machine learning for earth observation: A review and future prospects,

    Alessandro Sebastianelli, Francesco Mauro, Amer Delil- basic, Pietro Di Stasio, Fan Fan, Gabriele Meoni, Gabriele Cavallaro, Xiao Xiang Zhu, Paolo Gamba, and Silvia Liberata Ullo, “Quantum machine learning for earth observation: A review and future prospects,”IEEE Geoscience and Remote Sensing Magazine, 2026

    work page 2026

  7. [7]

    A comprehensive review of quantum machine learning: from nisq to fault tolerance,

    Yunfei Wang and Junyu Liu, “A comprehensive review of quantum machine learning: from nisq to fault tolerance,” Reports on Progress in Physics, vol. 87, no. 11, pp. 116402, 2024

    work page 2024

  8. [8]

    QMC-Net: Data-aware quantum representations for remote sensing image classification,

    Md Aminur Hossain, Ayush V . Patel, and Biplab Baner- jee, “QMC-Net: Data-aware quantum representations for remote sensing image classification,” inProceedings of the 28th International Conference on Pattern Recognition (ICPR), Lyon, France, 2026, Springer

    work page 2026

  9. [9]

    Quan- tum convolutional neural networks,

    Iris Cong, Soonwon Choi, and Mikhail D Lukin, “Quan- tum convolutional neural networks,”Nature Physics, vol. 15, no. 12, pp. 1273–1278, 2019

    work page 2019

  10. [10]

    D-linknet: Linknet with pretrained encoder and dilated convolution for high resolution satellite imagery road extraction,

    Lichen Zhou, Chuang Zhang, and Ming Wu, “D-linknet: Linknet with pretrained encoder and dilated convolution for high resolution satellite imagery road extraction,” in Proceedings of the IEEE conference on computer vision and pattern recognition workshops, 2018, pp. 182–186

    work page 2018

  11. [11]

    Deep residual learning for image recognition,

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    work page 2016

  12. [12]

    Deepglobe 2018: A challenge to parse the earth through satellite images,

    Ilke Demir, Krzysztof Koperski, David Lindenbaum, Guan Pang, Jing Huang, Saikat Basu, Forest Hughes, Devis Tuia, and Ramesh Raskar, “Deepglobe 2018: A challenge to parse the earth through satellite images,” in Proceedings of the IEEE conference on computer vision and pattern recognition workshops, 2018, pp. 172–181

    work page 2018

  13. [13]

    A review of quantum machine learning methods for remote sensing tasks,

    Nour Aburaed, Faisal Shah Khan, and Mohammed Q Alkhatib, “A review of quantum machine learning methods for remote sensing tasks,”Quantum Machine Intelligence, vol. 8, no. 1, pp. 50, 2026

    work page 2026

  14. [14]

    Quantum machine learning: A comprehensive review of integrating ai with quantum computing for computational advance- ments,

    Raghavendra M Devadas and T Sowmya, “Quantum machine learning: A comprehensive review of integrating ai with quantum computing for computational advance- ments,”MethodsX, vol. 14, pp. 103318, 2025

    work page 2025

  15. [15]

    Quanvolutional neural networks: powering image recognition with quantum circuits,

    Maxwell Henderson, Samriddhi Shakya, Shashindra Pradhan, and Tristan Cook, “Quanvolutional neural networks: powering image recognition with quantum circuits,”Quantum Machine Intelligence, vol. 2, no. 1, pp. 2, 2020

    work page 2020

  16. [16]

    An image clas- sification algorithm based on hybrid quantum classical convolutional neural network,

    Wei Li, Peng-Cheng Chu, Guang-Zhe Liu, Yan-Bing Tian, Tian-Hui Qiu, and Shu-Mei Wang, “An image clas- sification algorithm based on hybrid quantum classical convolutional neural network,”Quantum Engineering, vol. 2022, no. 1, pp. 5701479, 2022

    work page 2022

  17. [17]

    Absence of barren plateaus in quantum convolutional neural networks,

    Arthur Pesah, Marco Cerezo, Samson Wang, Tyler V olkoff, Andrew T Sornborger, and Patrick J Coles, “Absence of barren plateaus in quantum convolutional neural networks,”Physical Review X, vol. 11, no. 4, pp. 041011, 2021

    work page 2021

  18. [18]

    Qufex: Quantum feature extraction module for hybrid quantum-classical deep neu- ral networks,

    Naman Jain and Amir Kalev, “Qufex: Quantum feature extraction module for hybrid quantum-classical deep neu- ral networks,”arXiv preprint arXiv:2501.13165, 2025

    work page arXiv 2025

  19. [19]

    Qc- net: a hybrid quantum-classical neural network model for medical image segmentation: A. wang et al.,

    Aijuan Wang, Xiangqi Li, Lusi Li, and Tiehu Li, “Qc- net: a hybrid quantum-classical neural network model for medical image segmentation: A. wang et al.,”Quantum Information Processing, vol. 24, no. 10, pp. 340, 2025

    work page 2025

  20. [20]

    Qu-net: Quantum-enhanced u-net for self supervised embedding and classification of skin cancer images,

    Khidhr Halab, Nabil Marzoug, Othmane El Mes- louhi, Zouhair Elamrani Abou Elassad, and Moulay A Akhloufi, “Qu-net: Quantum-enhanced u-net for self supervised embedding and classification of skin cancer images,”Big Data and Cognitive Computing, vol. 10, no. 1, pp. 12, 2025

    work page 2025

  21. [21]

    Towards efficient quantum hybrid diffusion models,

    Francesca De Falco, Andrea Ceschini, Alessandro Se- bastianelli, Bertrand Le Saux, and Massimo Panella, “Towards efficient quantum hybrid diffusion models,” arXiv preprint arXiv:2402.16147, 2024

    work page arXiv 2024

  22. [22]

    Hqf-net: A hybrid quantum-classical multi-scale fusion network for remote sensing image segmentation,

    Md Aminur Hossain, Ayush V Patel, Siddhant Gole, Sanjay K Singh, and Biplab Banerjee, “Hqf-net: A hybrid quantum-classical multi-scale fusion network for remote sensing image segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2026

    work page 2026

  23. [23]

    Landcover. ai: Dataset for automatic mapping of build- ings, woodlands, water and roads from aerial imagery,

    Adrian Boguszewski, Dominik Batorski, Natalia Ziemba- Jankowska, Tomasz Dziedzic, and Anna Zambrzycka, “Landcover. ai: Dataset for automatic mapping of build- ings, woodlands, water and roads from aerial imagery,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 1102–1110

    work page 2021

  24. [24]

    Quantum machine learning: A hands-on tutorial for machine learning practitioners and researchers, 2025

    Yuxuan Du, Xinbiao Wang, Naixu Guo, Zhan Yu, Yang Qian, Kaining Zhang, Min-Hsiu Hsieh, Patrick Reben- trost, and Dacheng Tao, “Quantum machine learning: A hands-on tutorial for machine learning practitioners and researchers,”arXiv preprint arXiv:2502.01146, 2025

    work page arXiv 2025

  25. [25]

    Revisiting evaluation met- rics for semantic segmentation: Optimization and evalu- ation of fine-grained intersection over union,

    Zifu Wang, Maxim Berman, Amal Rannen-Triki, Philip Torr, Devis Tuia, Tinne Tuytelaars, Luc V Gool, Jiaqian Yu, and Matthew Blaschko, “Revisiting evaluation met- rics for semantic segmentation: Optimization and evalu- ation of fine-grained intersection over union,”Advances in Neural Information Processing Systems, vol. 36, pp. 60144–60225, 2023

    work page 2023

  26. [26]

    Land cover classification from sentinel-2 images with quantum- classical convolutional neural networks,

    Fan Fan, Yilei Shi, and Xiao Xiang Zhu, “Land cover classification from sentinel-2 images with quantum- classical convolutional neural networks,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024

    work page 2024

  27. [27]

    Remote sensing classification using quantum image processing,

    Hrithik Kumar, Teymoor Ali, Chris J Holder, A Stephen McGough, and Deepayan Bhowmik, “Remote sensing classification using quantum image processing,” inArti- ficial Intelligence and Image and Signal Processing for Remote Sensing XXX. SPIE, 2024, vol. 13196, pp. 157– 169

    work page 2024

  28. [28]

    Unet++: A nested u- net architecture for medical image segmentation,

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang, “Unet++: A nested u- net architecture for medical image segmentation,” in International workshop on deep learning in medical image analysis. Springer, 2018, pp. 3–11

    work page 2018

  29. [29]

    U-net with spatial pyramid pooling mod- ule for segmenting oil palm plantations,

    Siti Raihanah Abdani, Mohd Asyraf Zulkifley, and Ma- zlina Mamat, “U-net with spatial pyramid pooling mod- ule for segmenting oil palm plantations,” in2020 IEEE 2nd international conference on artificial intelligence in engineering and technology (IICAIET). IEEE, 2020, pp. 1–5

    work page 2020

  30. [30]

    Diresunet: Ar- chitecture for multiclass semantic segmentation of high resolution remote sensing imagery data,

    Priyanka, Sravya N, Shyam Lal, J Nalini, Chintala Sud- hakar Reddy, and Fabio Dell’Acqua, “Diresunet: Ar- chitecture for multiclass semantic segmentation of high resolution remote sensing imagery data,”Applied Intel- ligence, vol. 52, no. 13, pp. 15462–15482, 2022

    work page 2022