Ma et al. (2026) SGCAD: A SAR-Guided Confidence-Gated Distillation Framework of Optical and SAR Images for Water-Enhanced Land-Cover Semantic Segmentation

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

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-23
• Authors: Junjie Ma, Zheng Wang, Yubo Yuan, Fengming Hu
• DOI: 10.3390/rs18060962

Research Groups

Not specified in the provided text.

Short Summary

This paper introduces SAR-guided class-aware knowledge distillation (SGCAD) to resolve fusion conflicts in multimodal SAR and optical semantic segmentation, particularly for critical categories like water bodies, by leveraging SAR as a water-expert teacher and enhancing boundary continuity.

Objective

• To address fusion conflicts and under-optimization of critical categories (e.g., water bodies) in multimodal semantic segmentation using synthetic aperture radar (SAR) and optical imagery.

Study Configuration

• Spatial Scale: Regional, based on a self-built dataset from specific satellite imagery (GF-1 optical and LuTan-1 SAR).
• Temporal Scale: Snapshot-based, using near-synchronous satellite acquisitions.

Methodology and Data

• Models used: SAR-guided class-aware knowledge distillation (SGCAD), SAR-only HRNet (water-expert teacher), LightMCANet (lightweight multimodal student model), SAR edge guidance module (SEGM).
• Data sources: Self-built dataset from GF-1 optical imagery and LuTan-1 SAR imagery.

Main Results

• SGCAD significantly improves targeted category learning (e.g., water bodies) while maintaining stable performance across other classes.
• Experiments demonstrate higher overall accuracy compared to representative baselines.
• The method yields more coherent water and road predictions than existing approaches.

Contributions

• Proposes SGCAD, a novel method for multimodal semantic segmentation that addresses fusion conflicts by leveraging SAR as a class-aware guide.
• Introduces a SAR-only HRNet trained as a water-expert teacher to learn discriminative backscattering and boundary priors for water extraction.
• Develops a class-aware distillation strategy that transfers teacher knowledge only within high-confidence water regions, suppressing noisy supervision.
• Integrates a SAR edge guidance module (SEGM) in the decoder to enhance boundary continuity for slender structures like water bodies and roads.

Funding

Not specified in the provided text.

Citation

@article{Ma2026SGCAD,
  author = {Ma, Junjie and Wang, Zheng and Yuan, Yubo and Hu, Fengming},
  title = {SGCAD: A SAR-Guided Confidence-Gated Distillation Framework of Optical and SAR Images for Water-Enhanced Land-Cover Semantic Segmentation},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18060962},
  url = {https://doi.org/10.3390/rs18060962}
}