Luo et al. (2026) PSiam-HDSFNet: A Pseudo-Siamese Hybrid Dilation Spiral Feature Network for Flood Inundation Change Detection Based on Heterogeneous Remote Sensing Imagery

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-04
• Authors: Yichuang Luo, Xunqiang Gong, Yuanxin Ye, Pengyuan Lv, Shuting Yang, Ailong Ma, Yanfei Zhong
• DOI: 10.3390/rs18050788

Research Groups

[Information not explicitly available in the provided text.]

Short Summary

This paper proposes a novel pseudo-Siamese hybrid dilation spiral feature network (PSiam-HDSFNet) to improve flood change detection accuracy from heterogeneous SAR and optical remote sensing images, specifically addressing challenges in distinguishing small ground objects from actual inundated regions. The method significantly enhances change detection accuracy, with F1 scores improving by up to 7.704% compared to suboptimum methods.

Objective

• To improve the accuracy of flood area change detection from heterogeneous SAR and optical remote sensing imagery, particularly in distinguishing small ground objects within the background from actual inundated regions.

Study Configuration

• Spatial Scale: Pixel-level detection within remote sensing images.
• Temporal Scale: Event-driven (comparison of pre- and post-flood event imagery).

Methodology and Data

• Models used: Pseudo-Siamese hybrid dilation spiral feature network (PSiam-HDSFNet), which includes:
  • Enhanced Deep Residual Blocks
  • Residual Dense Blocks
  • Hybrid Dilated Pyramid (HDP) module with a sawtooth wave-like dilated coefficient
  • Spiral Feature Pyramid (SFP) module
  • Galerkin-type attention with linear complexity
  • Align OPT-SAR (AlignOS) method
• Data sources: Synthetic Aperture Radar (SAR) images and optical images.

Main Results

• PSiam-HDSFNet significantly improves change detection accuracy by effectively extracting and processing depth features from heterogeneous SAR and optical images without requiring image domain translation.
• The proposed method achieved F1 score improvements of 7.704%, 7.664%, 4.353%, and 1.111% across four flood coverage categories detection tasks compared to the suboptimum method.
• The Hybrid Dilated Pyramid (HDP) module successfully enhances multi-scale semantics of deep features, selectively reinforcing semantic features in flood areas and weakening noise from small ground objects.
• The Spiral Feature Pyramid (SFP) module makes the deep features of SAR and optical images more consistent in spatial structure and numerical distribution patterns, further suppressing noise from small ground objects.
• Galerkin-type attention efficiently reconstructs abstract semantic information into interpretable flood features in the decoder.
• The Align OPT-SAR (AlignOS) method effectively aligns SAR and optical image features for subsequent flood area detection.

Contributions

• Proposes PSiam-HDSFNet, a novel deep learning network for robust flood change detection using heterogeneous SAR and optical remote sensing imagery.
• Introduces a Hybrid Dilated Pyramid (HDP) module with a sawtooth wave-like dilated coefficient to enhance multi-scale semantic feature extraction, specifically targeting flood areas and suppressing noise from small ground objects.
• Designs a Spiral Feature Pyramid (SFP) module to improve the spatial and numerical consistency of deep features between SAR and optical images, further suppressing noise.
• Integrates Galerkin-type attention with linear complexity into the decoder for rapid and efficient reconstruction of abstract flood semantic information.
• Develops the Align OPT-SAR (AlignOS) method for effective feature alignment between SAR and optical images, crucial for heterogeneous data fusion.
• Demonstrates significant quantitative improvements in flood change detection accuracy (up to 7.704% F1 score improvement) without relying on image domain translation.

Funding

[Information not explicitly available in the provided text.]

Citation

@article{Luo2026PSiamHDSFNet,
  author = {Luo, Yichuang and Gong, Xunqiang and Ye, Yuanxin and Lv, Pengyuan and Yang, Shuting and Ma, Ailong and Zhong, Yanfei},
  title = {PSiam-HDSFNet: A Pseudo-Siamese Hybrid Dilation Spiral Feature Network for Flood Inundation Change Detection Based on Heterogeneous Remote Sensing Imagery},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18050788},
  url = {https://doi.org/10.3390/rs18050788}
}