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

2026-04-21 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Remote Sensing · Depth: Expert, extended

Summary

HQF-Net, a novel hybrid quantum-classical multi-scale fusion network, has been developed for remote sensing image semantic segmentation. This architecture integrates multi-scale semantic guidance from a frozen DINOv3 ViT-L/16 backbone with a customized U-Net. Key innovations include a Deformable Multiscale Cross-Attention Fusion (DMCAF) module for aligning features, Quantum-enhanced Skip connections (QSkip) for feature refinement, and a Quantum bottleneck with Mixture-of-Experts (QMoE) that adaptively combines local, global, and directional quantum circuits. HQF-Net demonstrated consistent improvements across three remote sensing benchmarks, achieving 0.8568 mIoU and 96.87% overall accuracy on LandCover.ai, 71.82% mIoU on OpenEarthMap, and 55.28% mIoU with 99.37% overall accuracy on SeasoNet. An ablation study confirmed the individual contributions of its major components.

Key takeaway

For research scientists developing advanced remote sensing segmentation models, HQF-Net demonstrates that integrating hybrid quantum-classical components can yield superior performance. You should consider incorporating multi-scale semantic guidance and quantum-enhanced feature refinement modules, particularly a Mixture-of-Experts approach, to improve accuracy and boundary delineation in complex imagery, even under current NISQ device constraints.

Key insights

Hybrid quantum-classical networks can significantly improve remote sensing semantic segmentation by fusing multi-scale features and quantum-enhanced refinement.

Principles

• Multi-scale semantic guidance improves segmentation.
• Quantum circuits enhance feature interactions.
• Adaptive routing of quantum experts is effective.

Method

HQF-Net integrates a DINOv3 ViT-L/16 backbone with a U-Net via DMCAF, QSkip, and a QMoE bottleneck, using parameterized quantum circuits for feature enrichment and adaptive routing.

In practice

• Use DINOv3 for robust visual representations.
• Implement deformable cross-attention for feature alignment.
• Employ quantum-enhanced skip connections for refinement.

Topics

• HQF-Net
• Remote Sensing Segmentation
• Hybrid Quantum-Classical Learning
• DINOv3 Vision Transformer
• Deformable Cross-Attention

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Implicit spatial-frequency fusion of hyperspectral and lidar data via kolmogorov-arnold networks — KANs enable adaptive, non-linear fusion of HSI and LiDAR data across spatial and frequency domains.
• SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning - AI at Meta — SAM 3.1 enhances real-time video object tracking via multiplexing, doubling processing speed and reducing GPU demands.
• Feasibility of Indoor Frame-Wise Lidar Semantic Segmentation via Distillation from Visual Foundation Model — Cross-modal distillation from Visual Foundation Models enables effective, label-free indoor lidar semantic segmentation.
• Hybrid Approach for Enhancing Lesion Segmentation in Fundus Images — A hybrid model combining U-Net and SLIC significantly improves choroidal nevus segmentation accuracy and generalizability on fundus images.
• GMBFormer: An NDVI-Guided Global Memory Bank Transformer for Urban Green-Space Extraction from Ultra-High-Resolution Imagery — GMBFormer improves urban green-space extraction by decoupling NDVI and using a global memory bank for semantic reuse.
• Robust Multispectral Semantic Segmentation under Missing or Full Modalities via Structured Latent Projection — Structured latent projection in CBC-SLP preserves modality-specific information, enhancing multimodal semantic segmentation robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.