SAR Coherence Estimation by Composition of Subsample Estimates and Machine Learning

Abstract

Synthetic aperture radar (SAR) coherence magnitude is an essential parameter in SAR interferometry. This is the reason why current interferometric wide area ground motion services require the estimation of the coherence magnitude as accurately and computationally effectively as possible. The objective of this article is to improve the accuracy of this coherence estimation compared to known estimators, especially when estimating low coherences and working with a small, i.e., N $< $ 30, but also large number of samples, i.e., hundred or more. Precisely, this article proposes the interferometric coherence magnitude estimation by composition of subsample estimates and machine learning (ML). The principle is to partition the given sample and to estimate coherences on these independent subsamples using different coherence magnitude estimators. It results in a nonparametric and automated statistical inference. It is shown that the composite ML estimator has a high estimation quality yet without prior information, provides a deterministic estimate and is numerically efficient, it is suitable for general interferometric synthetic aperture radar applications and operational systems. Adequate computational performance results from the fact that no iteration, numerical integration, bootstrapping, or bagging are part of the composite estimator.

Keywords

• Coherence magnitude
• composite estimator
• degree of coherence
• distributed scatterer in SqueeSAR or CAESAR or phase linking
• interferometric synthetic aperture radar (InSAR)
• supervised machine learning (ML)