QMC-Net: Data-Aware Quantum Representations for Remote Sensing Image Classification

Md Aminur Hossain; Ayush V. Patel; Biplab Banerjee

quant-phApr 10, 2026

QMC-Net: Data-Aware Quantum Representations for Remote Sensing Image Classification

Authors: Md Aminur Hossain, Ayush V. Patel, Biplab Banerjee

Organizations: Space Applications Centre, Indian Space Research Organisation, Ahmedabad, India · School of Technology, Pandit Deendayal Energy University, Gandhinagar, India · Centre of Studies in Resources Engineering, Indian Institute of Technology, Bombay, India · Centre for Machine Intelligence and Data Science, Indian Institute of Technology, Bombay, India

Abstract

Hybrid quantum-classical models offer a promising route for learning from complex data; however, their application to multi-band remote sensing imagery often relies on generic, data-agnostic quantum circuits that fail to account for channel-specific statistical variability. In this work, we propose a data-driven framework that maps band-level statistics such as Shannon Entropy, Variance, Spectral Flatness, and Edge Density to the hyperparameters of customized quantum circuits. Building on this framework, we introduce QMC-Net, a hybrid architecture that processes six data channels using band-specific quantum circuits, enabling adaptive quantum feature encoding and transformation across channels. Experiments on the EuroSAT and SAT-6 datasets demonstrate that QMC-Net achieves accuracies of 93.80 % and 99.34 %, respectively, while a residual-enhanced variant further improves performance to 94.69 % and 99.39 %. These results consistently outperform strong classical baselines and monolithic hybrid quantum models, highlighting the effectiveness of data-aware quantum circuit design under NISQ constraints.

Figures & tables

Figure 1: Visualization of the six-channel data (RGB, EVI, NDVI, & Entropy) derived from the EuroSAT dataset for representative samples of all 10 classes.

Hyperparameter	Driving Metric(s)	Design Heuristic
Qubits ( $Q$ )	Entropy ( $H$ ), Flatness ( $F$ )	$Q\gtrsim H$ , strict if $F$ is high
Depth ( $D$ )	Entropy ( $H$ )	Low $\to$ 1-2L; Med $\to$ 2-4L; High $\to$ 4+L ∗
Entanglement ( $E$ )	Edge Density ( $ED$ )	Low $\to$ Linear; Med $\to$ Ring; High $\to$ Multi-scale
Parameters ( $P$ )	Variance ( $\sigma^{2}$ )	Low $\to$ 1/Q; Med $\to$ 3/Q; High $\to$ 3/Q + Re-upload
∗ Increased depth is applied cautiously to mitigate barren plateaus.

Table 1: The Metric-Driven Design Rubric. This table links each quantum hyperparameter to its primary classical data driver and the corresponding design heuristic.

Dataset	Band	$\boldsymbol{H}$	$\boldsymbol{\sigma^{2}}$	$\boldsymbol{F}$	$\boldsymbol{ED}$
EuroSAT	Red	6.4017	4015.0316	0.2109	0.5775
	Green	6.6672	4022.5922	0.2332	0.5898
	Blue	6.7971	4019.2067	0.2401	0.5598
	NDVI	7.0133	2911.9893	0.8526	0.3098
	EVI	2.9599	6512.0394	0.2438	0.3368
	Entropy	7.1627	1888.4368	0.9420	0.1369

Table 2: Classical metrics for EuroSAT and SAT-6 datasets

Figure 2: QMC-Net Architecture: A hybrid quantum-classical deep learning framework for processing multi-band remote sensing imagery (RGB, EVI, NDVI, and Entropy).

Figure 3: Band-specific Quantum Circuits designed for EuroSAT bands by following the design rubrics - a) RGB Circuit, b) NDVI Circuit, c) EVI Circuit, & d) Entropy Circuit

Table 6

Ansatz for All Bands	Accuracy (%)	Precision	Recall	F1
No Quantum Circuits in QMC-Net	84.54	84.40	84.50	84.30
Ry Circuit	91.87	92.10	91.90	91.80
Bellman Circuit	90.98	91.30	91.00	90.90
Real Amplitudes	92.04	92.00	92.00	92.00
Band-Specific (Ours)	93.80	93.80	93.80	93.80

Table 5: Architectural Ablation: Band-Specific vs. Monolithic Ansatze.

Bands Included						Accuracy (%)
R	G	B	EVI	NDVI	Entropy	Result
✓	✓	✓				80.86
✓	✓	✓	✓			88.41
✓	✓	✓		✓		90.88
✓	✓	✓			✓	85.65
✓	✓	✓	✓	✓		90.39

Table 6: Additive Ablation Study on Band Contribution.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure S1: Band-specific quantum circuits for SAT-6 dataset, a) For RGB, NDVI, & Entropy bands, b) For EVI band

Figure S2: Placement of Residual block in our QMC-Net Architecture

Figure S3: Accuracy and Loss curves on a) EuroSAT dataset & b) SAT-6 dataset

Figure S4: t-SNE 2D plot of our trained model on a) EuroSAT dataset & b) SAT-6 dataset

Panel of Figure 13 — Figure S5: Confusion matrices for (a) EuroSAT and (b) SAT-6 datasets.

Hybrid Quantum-Classical ML

Hybrid quantum-classical machine learning combines quantum circuits or annealers with classical models for learning, prediction, generation, and optimization.

See all 179 papers

HQ-UNet: A Hybrid Quantum-Classical U-Net with a Quantum Bottleneck for Remote Sensing Image Segmentation
Apr 29, 2026Md Aminur Hossain, Ayush V. Patel, Ikshwaku Vanani +1Remote Sensing Image Segmentation Hybrid Quantum-Classical ML
Space Applications Centre, Indian Space Research Organisation (ISRO), India · New-Brunswick Honors College, Rutgers University, New Jersey · Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay
HQF-Net: A Hybrid Quantum-Classical Multi-Scale Fusion Network for Remote Sensing Image Segmentation
Apr 8, 2026Md Aminur Hossain, Ayush V. Patel, Siddhant Gole +2Remote Sensing Image Segmentation Multi-Scale Feature Fusion
Space Applications Centre, ISRO, India · Indian Institute of Technology Bombay
Evolving Hybrid Quantum-Classical Architectures for Image Classification
Oct 2, 2026Devroop Kar, Daniel Krutz, Travis DesellHybrid Quantum-Classical ML Quantum Architecture Search
Rochester Institute of Technology Rochester, NY 14623, USA

Apr 29, 2026J/K move · Enter open · S save

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

Space Applications Centre, Indian Space Research Organisation (ISRO), India · New-Brunswick Honors College, Rutgers University, New Jersey · Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay

Remote Sensing Image Segmentation Hybrid Quantum-Classical ML Quantum Neural Networks

Open paper AskPDF arXiv

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.

Abstract

Hybrid Quantum-Classical ML

Explore similar work

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

HQF-Net: A Hybrid Quantum-Classical Multi-Scale Fusion Network for Remote Sensing Image Segmentation

Evolving Hybrid Quantum-Classical Architectures for Image Classification