A Nonparametric Estimator for Coherent Change Detection: The Permutational Change Detection

Abstract

Nowadays, Synthetic Aperture Radar (SAR) is widely used in heterogeneous fields with aims strictly dependent on the objectives of the application. One of the most common is the exploitation of the Interferometric-SAR (InSAR) to measure millimeter movements on the Earth’s surface, aiming to monitor failures (e.g. landslides) or to measure the health state of infrastructures (e.g. mining assets, bridges, buildings, etc). In this context, developing algorithms to detect temporal and spatial changes in the radar targets becomes very important. This paper focuses on the temporal change detection framework, proposing a non-parametric Coherent Change Detection (CCD) algorithm called Permutational Change Detection (PCD), a purely statistical algorithm whose core is the Permutation Test. The PCD estimates the temporal Change Points (CPs) of a radar target recognizing <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">blocks structure</i> in the coherence matrix, namely new radar objects. The algorithm has been fine-tuned for small SAR datasets, with the specific aim of prioritizing the analysis of the latest changes. A rigorous mathematical derivation of the algorithm is carried out, explaining how some limits have been addressed. Then, the performance analysis on simulated data is deeply accomplished, carried out for the stand-alone PCD and for the PCD compared with a parametric CCD algorithm based on the Generalized Likelihood Ratio Test (GLRT), and with the Omnibus and REACTIV detectors. The comparison with these other algorithms and the stand-alone performance analysis point out the robustness of the PCD in dealing with very noisy environments, even in the case of a single block. Finally, the PCD is validated by processing two Sentinel I data stacks, ascending and descending geometry, of the 2016 Central Italy earthquake.

Topics & Concepts