Research on Water Body Extraction from Gaofen-3 Imagery Based on Polarimetric Decomposition and Machine Learning

Abstract

The quick and accurate extraction of water bodies from images is imperative for land resources management, ecological protection, and flood disaster prevention. However, the prevalent methods for water body extraction of SAR imagery have the major issues of depending on expert experience, the poor retention of water-land boundaries, and a high false alarm rate. In this paper, focusing on these issues, we study the effectiveness of water body extraction using only simple polarimetric decomposition components and commonly used machine learning classifiers for Gaofen-3(GF-3) image. We test it using different sizes of training samples and compare it with the previous methods. The experimental results showed that the combination of polarimetric decomposition components and machine learning classifiers can achieve satisfactory accuracies in water body extraction of GF-3 image, and has strong robustness, reliability and high application value.