Wetland information extraction of the East Dongting Lake using mean shift segmentation

Wetlands are a natural complex formed by the interaction of land and water systems. They play an irreplaceable role in biodiversity conservation, control of global climate change, water purification and mitigation of flood disaster. Thus, extracting information of wetlands has become very important. Recent years the rapid development of high spatial resolution remote sensing technology provides great potential for improvement of data sources and advancing methods for quantitative acquisition and analysis of wetland information. It is well known that object-oriented method is a relatively new technology for landscape segmentation. Although there are some reports in application of object-oriented analysis for extraction of wetland information in China, there is still a lack of studies on the impacts of used segmentation techniques on accuracy of classification. In this study, an excellent image region segmentation method which appeared in the recent years, called mean shift segmentation algorithm, was used to extract the information of wetland in the East Dongting Lake of China and the obtained results were compared with those from a conventional segmentation algorithm provided by ENVI EX. The assessment of the results was conducted using four kinds of quantitative indicators and based on the accuracy of interpretation. The results showed that the conventional segmentation algorithm was unable to provide the accurate segmentation results in delineation of wetland areas. Integrating the edge detection information of NDVI and the mean shift segmentation algorithm not only could make it possible segmentation of shallow water bodies, but also could lead to much better classification results than using the traditional method and the mean shift segmentation alone.