Interferometric SAR Coherence Magnitude Estimation by Machine Learning

Current interferometric wide area ground motion services require the estimation of coherence magnitude as accurately and computationally effectively possible. However, a precise at same time efficient method is missing. Therefore, objective this article to improve empirical Bayesian in terms accuracy computational cost. Precisely, proposes by Machine Learning (ML). It results non-parametric automated statistical inference. applying ML context not straightforward. The number domain possible input processes infinite it train all signals. shown that expected channel amplitudes phase cause redundancies signals allowing solve issue. Similar methods, single parameter for maximum underlaying used model prior. no prior or any shape probability easy implement within framework. reports on bias, standard deviation root mean square error (RMSE) developed estimators. was found estimators RMSE from small samples ( $2 N < 30$ ) compared conventional Bayes For three notation="LaTeX">$N = 3$ zero magnitude, bias related sample estimator improves 0.53 0.39 27.8%. Assuming 0.6, reduced 33.0% 0.36 less strict 45.5% 0.29 are suitable recommended operational InSAR systems. estimation, extremely fast evaluated because iteration, numeric integration Bootstrapping needed.