Beyond Backscatter: InSAR coherence from detected SAR images

2026-06-05 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

A deep learning framework is proposed for InSAR coherence regression directly from detected SAR images, eliminating the need for accurate coregistration. This method utilizes a Residual U-Net, trained on coherence maps derived from precisely coregistered Sentinel-1 SLC data, specifically 12-day SLC pairs, to learn the relationship between backscatter magnitudes and coherence. Experimental results demonstrate that this approach achieves high-resolution coherence regression with improved accuracy compared to existing intensity-based methods. The network exhibits strong generalization capabilities across diverse geographical locations and even to different temporal baselines not encountered during training. Its ability to operate on globally available analysis-ready data, such as ground range detected data distributed via Google Earth Engine, facilitates large-scale applications in mission design, change monitoring, and various mapping tasks.

Key takeaway

For Machine Learning Engineers developing SAR-based monitoring systems, this framework offers a path to high-resolution InSAR coherence without complex coregistration. You can utilize globally available analysis-ready data, like Google Earth Engine products, to streamline large-scale change detection and mapping tasks. This simplifies data preparation and expands the applicability of coherence analysis across diverse geographies and temporal scales.

Key insights

A Residual U-Net regresses InSAR coherence directly from detected SAR images, bypassing coregistration and improving accuracy for large-scale applications.

Principles

• Deep learning can infer coherence from backscatter.
• Coregistration is not always essential for coherence.
• Models can generalize to unseen temporal baselines.

Method

A Residual U-Net is trained on 12-day Sentinel-1 SLC coherence maps to learn backscatter-coherence relationships. It then regresses coherence from detected SAR image magnitudes, bypassing coregistration.

In practice

• Use ground range detected data for coherence.
• Apply to mission design and change monitoring.
• Facilitate large-scale mapping tasks.

Topics

• InSAR Coherence
• Deep Learning
• SAR Imaging
• Residual U-Net
• Sentinel-1
• Google Earth Engine
• Change Monitoring

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

Related on AIssential

• Beyond Backscatter: InSAR coherence from detected SAR images — Deep learning can accurately regress InSAR coherence from detected SAR images, bypassing complex coregistration.
• 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.
• WILD-SAM: Phase-Aware Expert Adaptation of SAM for Landslide Detection in Wrapped InSAR Interferograms — WILD-SAM adapts SAM for landslide detection in InSAR data by integrating phase-aware spectral alignment and wavelet-guided prompt enhancement.
• v292: TerraBytes at ICML 2025 — The TerraBytes workshop showcases diverse Earth Observation advancements, from specific applications to foundational model deployment and data optimization.
• Learned Non-Maximum Suppression for 3D Object Detection — Learned filtering modules, D2D-Rescore and GossipNet3D, enhance 3D object detection by replacing heuristic NMS with detection-to-detection relation leveraging.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.