Beyond Backscatter: InSAR coherence from detected SAR images

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Environmental Science & Earth Systems, Engineering & Applied Sciences, Mathematics & Computational Sciences · Depth: Advanced, extended

Summary

A deep learning framework is proposed for coherence regression directly from detected Synthetic Aperture Radar (SAR) images, eliminating the need for accurate coregistration. This Residual U-Net model, trained on 12-day Sentinel-1 Single-Look Complex (SLC) pairs, learns the relationship between backscatter magnitudes and coherence. The framework demonstrates strong generalization across diverse datasets, including different temporal baselines (0, 6, 12, and 36 days), geographic locations (e.g., San Francisco Bay, Atacama Desert), polarizations (VV to VH), and even operative frequencies (Sentinel-1 C-band to ALOS L-band). It consistently outperforms existing intensity-based methods, achieving an RMSE of 0.108 on a 12-day Sentinel-1 test over the San Francisco Bay area. The method also successfully adapts to globally available analysis-ready Sentinel-1 Ground Range Detected (GRD) products from Google Earth Engine, enabling large-scale applications in change monitoring and mapping.

Key takeaway

For Machine Learning Engineers developing SAR applications, if you require interferometric coherence maps but want to avoid computationally intensive SLC coregistration, this deep learning framework offers a viable alternative. You can utilize detected SAR products, including Google Earth Engine GRD data, to generate high-resolution coherence, significantly accelerating large-scale change monitoring and mapping tasks. Consider fine-tuning for specific sensor or frequency domains to optimize performance.

Key insights

Deep learning can accurately regress InSAR coherence from detected SAR images, bypassing complex coregistration.

Principles

• Coherence estimation can bypass subpixel coregistration.
• Paired backscatter observations contain coherence cues.
• Deep learning models generalize across SAR acquisition conditions.

Method

A Residual U-Net is trained on coregistered SLC coherence maps, using detected backscatter pairs as input, then fine-tuned for GRD products with spatial multilooking and radiometric normalization.

In practice

• Use detected SAR for rapid coherence screening.
• Apply to Google Earth Engine GRD for large-scale monitoring.
• Extend coherence-based analysis to non-nominal SAR data.

Topics

• InSAR Coherence Estimation
• Deep Learning
• SAR Image Analysis
• Sentinel-1 GRD
• Google Earth Engine
• Residual U-Net

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

Related on AIssential

• Beyond Backscatter: InSAR coherence from detected SAR images — A Residual U-Net regresses InSAR coherence directly from detected SAR images, bypassing coregistration and improving accuracy for large-scale applications.
• When Less Is More: Simplicity Beats Complexity for Physics-Constrained InSAR Phase Unwrapping — Simpler convolutional architectures outperform complex attention-based models for physics-constrained geophysical regression tasks.
• Global Offshore Wind Infrastructure: Deployment and Operational Dynamics from Dense Sentinel-1 Time Series — A new global Sentinel-1 SAR dataset enables high-resolution monitoring of offshore wind infrastructure deployment and operations.
• Google Maps is about to get a big dose of AI — Google's new generative AI features enhance enterprise geospatial analysis and visualization across its mapping platforms.
• Simera Sense to offer larger cameras and enhanced autonomy — On-orbit reconfigurable hyperspectral payloads with sub-meter resolution are emerging for diverse Earth observation applications.
• Semantic Smoothing via Novel View Synthesis for Robust SAR Image Classification — Semantic smoothing uses generative models to create semantically consistent views for robust classification, outperforming isotropic noise.

Open in AIssential →

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