The Decision Tree Algorithm of Automatically Extracting Residential Information from SPOT Images
SPOT卫星影像居民地信息自动提取的决策树方法研究

Extracting residential information by remote sensing technology is significant for the loss estimation of natural disaster and the study of urban extension and environmental change.In this paper,taking Jiangning county of Nanjing as a case study area,the extraction of re sidential information from SPOT images is discussed.Firstly, the characteristics of residential areas in this area on this image are studied,secondly,spectral characteristics of residential areas and other land-use types on SPOT-4(XI) image are analyzed to find the possibility of extracting residential areas from the background according to spectral characteristics.Thirdly,a simple model of decision tree is proposed on the basis of spectral magnitude relations and some proper thresholds of residents and other land-use types.But some roads still cannot be separated from residential areas because their spectral characteristics are similar.Therefore,their shape characteristics are analyzed.The result is generalized and converted to vector coverage.Then the shape indexes of every spots are calculated and categorized choosing a threshold to remove the roads.Finally,the vector coverage of residential areas is overlaid on the original image to check the effectness of this model,and an accuracy assessment is given to the result by random samples.The results suggest that this model is simple and effective,especially for the residential areas over 10000 m2,and the analyzedfor is much higher than the supervised classification's,however,some pixels near the water-bodies and the roads are judged by mistakes.This problem can be solved by the rule judgment based on the knowledge of spatial relations.Therefore,the cities,towns and villages in the south of Yangze River where there are various types of land cover can be extracted from the images by this model,and the model is not limited by the time,only the thresholds will be changed.