Object-oriented water extraction of PolSAR image based on target decomposition

Abstract

Synthetic Aperture Radar (SAR), with all-weather and all-time imaging capabilities, provides a stable data resource for environmental monitoring. In water related applications, obtaining precise water areas from SAR data is essential. This paper presents a high accuracy water extraction scheme based on Object-Oriented technique. First, a watershed segmentation is used for preserve waterlines. Then, a decomposition method is applied to identify water features. Afterwards, a majority voting strategy is utilized for water segment recognition. A set of Radarsat-2 data is used to verify the efficiency of the proposed method. And a pixel-based scheme and a proposed method without the majority voting step are applied for comparison. The experimental results indicate that the segmentation is able to maintain accurate waterline, while the voting strategy is capable of detecting small water area.