Signal Models for Changes in Polarimetric SAR Data

Synthetic aperture radar (SAR) polarimetry can improve change detection in terms of detection capabilities. In this work, we are proposing to extend the idea of target decomposition to changes affecting partial targets. This will allow the separation of polarimetric-dependent changes, providing extra information that can be used to better understand the processes affecting the targets. Three models for changes are proposed and compared. The methodologies are based on Lagrangian optimizations of distinct operators built using quadratic forms for a power ratio and a power difference. The optimizations can be accomplished by diagonalizations of specific matrices derived from polarimetric covariance matrices. These are, therefore, spectral decompositions of an appropriate matrix which we define as change matrix. The theoretical validity of the models is assessed using Monte Carlo simulations. Additionally, we perform real data validation exploiting L-band quad-polarimetric data from the E-SAR (DLR) SARTOM 2006 campaign and ALOS PALSAR (JAXA) acquisitions in Morecombe Bay (U.K.). We observed that the two algorithms based on power difference allow to decompose the change into the minimal set of scattering mechanisms (SMs) that have been added or removed from the scene. The two algorithms differ on the initial assumption on the change. On the other hand, the ratio operator provides a better detection performance although the eigenvalues do not correspond to meaningful SMs. A combination of the three methodologies can, therefore, improve detection and classification of changes.