PolSAR Time Series Processing With Binary Partition Trees

Abstract

Abstract-This paper deals with the processing of polarimetric synthetic aperture radar (SAR) time series. Different approaches to deal with the temporal dimension of the data are considered, which are derived from different target characterizations in this dimension. These approaches are the basis for defining two different binary partition tree (BPT) structures that are employed for SAR polarimetry (PolSAR) data processing. Once constructed, the BPT is processed by a tree pruning, producing a set of spatiotemporal homogeneous regions, and estimating the polarimetric response within them. It is demonstrated that the proposed technique preserves the PolSAR information in the spatial and the temporal domains without introducing bias nor distortion. Additionally, the evolution of the data in the temporal dimension is also analyzed, and techniques to obtain BPT-based scene change maps are defined. Finally, the proposed techniques are employed to process two real RADARSAT-2 data sets.

Multi-image Segmentation: A Collaborative Approach Based on Binary Partition Trees

“…In the case of multi-date images, we may handle the land-cover evolutions of the observed territories by considering an adequate region model and merging criterion [23]. Such spatio-temporal information could be used to follow local consensus between images leading to hyper-trees where the branches model local temporal fusion decisions.…”

Target characterization by means of PolInSAR temporal evolution

“…Then, an homogeneity measure φ(X) has to be defined for this purpose, in order to test the homogeneity of the region X. Accordingly, the tree pruning is obtained by setting a pruning threshold δ p over this homogeneity measure [7] φ(X) = 1…”

“…In fact, the polarimetric evolution is contained within the sequence of Z ii matrices of their region model Z N , as described in (5). A temporal stability measure t s may be employed to determine the relevance of the changes among different acquisitions for each region [7] …”

Analysis of Min-Trees over Sentinel-1 Time Series for Flood Detection

“…Extension of such hierarchies to satellite image time series (SITS) has only been addressed in very few works. Alonso-González et al propose in [7] to build a Binary Partition Tree between two SAR images considering a temporal connectivity. Falco et al perform pixelwise change detection [8] comparing sums of differential attribute profiles (AP) in each pixel.…”