Study on the automatic extraction of water body from TM image using decision tree algorithm

Abstract

Landsat TM is the unique resource for global change research and applications in agriculture, geology, forestry, regional planning and so on. As the development of remote sensing applications, various water body extraction methods have been researched upon and developed in TM data. But there still are some limitations in the extracting water bodies which are partly confused with some residential districts. This paper concentrates on extracting water bodies by using decision tree classifiers based on TM images. Firstly, the formula TM2+TM3>TM4+TM5 was used to extract reservoirs, ponds and broad rivers. Then according to the mechanism and spectral characteristics of water body and other objects in TM data, the structure of the decision tree and the locations for each notes in the tree are determined. Finally, based on the model established using ENVI software, water area is extracted automatically, and the yield images are checked by visual and statistical accuracy assessment. The results show the application of decision tree is simple and could improve the accuracy of water body extraction. The decision tree, however, missed small water bodies at scales below the sensor resolution.