Comparison of Different Scatter Number Estimators in SAR Tomography

Inhalt:
In multi-baseline SAR tomography (TomoSAR), the technique of scatter number estimation plays a vital role due to their strong support in both peaks detection step and the internal inversion methods (i.e. TSVD, Capon, MUSIC, etc.). However, little is known comparatively about the performances of the estimators. In this paper, we compare in theoretic simulated environment and validate the estimation results generated using six widely used scatter number estimators, including theAkaike’s information-theoretic criterion (AIC), the minimum description length (MDL), the Hannan and Quinn method (HQ), the smoothed rank profile test (SRPT), the gerschgorin’s disk theorem (GDE) and the canonical correlation test (CCT) with multi-baseline Radarsat-2 images. We presentthecuvers of probility of estimation with change of signal to noise ratio (SNR) with all the estimators and results show that all hold strong robustness to SNR especially the GDE and CCT. We validate the performances of all the estimators byusing an stadium of Radarsat-2 images in TomoSAR and show that most mehods over estimate the scatter number except the GDE and CCT, and comparison with coherence images and structure of the stadium support the results of GDE a little more.