Water body automated extraction optimization and accuracy assessment methodology

Landscape element extraction from high resolution satellite data or orthophotos is not fully automated yet and it is often done by hand. This consumes a lot of time and that is why it is important to look for ways to do the process atleast semi-automatic and reduce the amount of manual work. The main object of this masters paper is to create the methodology of automated water body extraction and assess its possibilites and accuracy. This paper consists of content, anotation, short-phrase dictionary, literature overview, popular landscape element extraction methods overview, methodology, result analysis, accuracy assessment and conclusions. Three most popular landscape element extraction methods are reviewed in this paper – visual (manual), automated and field measurements. Four different classifications for 20 water bodies were used – visual (manual), supervised, unsupervised and with the developed software. Results and their quality assessment are presented in the last chapter, also recommendations on how to improve the software are given. At the end of the paper conclusions are presented. This analysis proved that the developed software was really accurate while extracting water body poligons from the images, meanwhile automated supervised and unsupervised classifications were less accurate. This master paper provides a strong foundation for the continuous research of automated lanscape element extraction