Hybrid Quantum-Classical Network HQF-Net Proposed for Remote Sensing Image Segmentation
A novel hybrid architecture known as HQF-Net has been unveiled for semantic segmentation in remote sensing. This model merges a static DINOv3 ViT-L/16 backbone with a tailored U-Net design via a Deformable Multiscale Cross-Attention Fusion module. To enhance feature refinement, it employs quantum-enhanced skip connections alongside a Quantum bottleneck utilizing Mixture-of-Experts. This method integrates complementary local, global, and directional quantum circuits through an adaptive routing system. The study tackles the shortcomings of traditional encoder-decoder frameworks like U-Net, which often fail to fully leverage global semantics and structured feature interactions. Effective remote sensing semantic segmentation necessitates models that can capture intricate spatial details and overarching semantic context in complex environments. The findings were published on arXiv under identifier 2604.06715v2 as a replacement cross announcement.
Key facts
- HQF-Net is a hybrid quantum-classical multi-scale fusion network for remote sensing image segmentation
- The model integrates multi-scale semantic guidance from a frozen DINOv3 ViT-L/16 backbone
- It uses a customized U-Net architecture with a Deformable Multiscale Cross-Attention Fusion module
- Quantum-enhanced skip connections (QSkip) are introduced for feature refinement
- A Quantum bottleneck with Mixture-of-Experts (QMoE) combines complementary quantum circuits
- The approach addresses limitations of classical encoder-decoder architectures like U-Net
- Remote sensing semantic segmentation requires capturing both fine details and high-level context
- The research was announced on arXiv with identifier 2604.06715v2
Entities
Institutions
- arXiv