Multilevel hierarchical segmentation method for polarimetric SAR data based on scattering behaviour and histograms

Abstract

One of the main applications of polarimetric synthetic aperture radar (POLSAR) is the classification of different land cover. Classification of polarimetric SAR data can be pixel-based or segment-based. Various segmentation methods for polarimetric SAR data, such as region growing and split-merge, have been proposed recently. This paper proposes a new multilevel divisive hierarchical segmentation methodology based on the separation of the scattering mechanism space produced by polarimetric SAR decomposition methods. This methodology is segment specific, i.e., several segmentation levels can be created for segments selected by the user. In the first and second levels, segmentation is performed based on the dominant scattering mechanism and the second most significant scattering mechanism. In the subsequent levels, data are further segmented using a histogram-based segmentation algorithm. The methodology was applied to two datasets based on Pauli and Freeman–Durden analysis images. Three segmentation levels are implemented and compared with segmentation results based on the k-means clustering. Advanced Land Observing Satellite (ALOS) full polarimetric SAR data are used for a study area located in the southwest of the United Kingdom. The proposed methodology leads to promising results and can be adjusted to user needs and existing knowledge of the target area.