Covariance Symmetries Classification in Multitemporal/Multipass PolSAR Images

A polarimetric synthetic aperture radar (PolSAR) system, which uses multiple images acquired with different polarizations in both transmission and reception, has the potential to improve the description and interpretation of the observed scene. This is typically achieved by exploiting the polarimetric covariance or coherence matrix associated with each pixel, which is processed to meet a specific goal in Earth observation. This paper presents a design framework for selecting the structure of the polarimetric covariance matrix that accurately reflects the symmetry associated with the analyzed pixels. The proposed methodology leverages both polarimetric and temporal information from multipass PolSAR images to enhance the retrieval of information from the acquired data. To accomplish this, it is assumed that the covariance matrix (of the overall acquired data) is given as the Kronecker product of the temporal and polarimetric covariances. An alternating maximization algorithm, known as the flip-flop method, is then developed to estimate both matrices while enforcing the symmetry constraint on the polarimetric covariance. Subsequently, the symmetry structure classification is formulated as a multiple hypothesis testing problem, which is solved using model order selection techniques. The proposed approach is quantitatively assessed on simulated data, showing its advantages over its competitor, which does not exploit temporal correlations. For example, it reaches accuracies of 94.6% and 92.0% for the reflection and azimuth symmetry classes, respectively, while the competitor achieves 72.5% and 72.6% under the same simulation conditions. Finally, the effectiveness of the proposed framework is further demonstrated using measured RADARSAT-2 data, corroborating the results obtained from the simulations.