Study On Extraction Methods Of Urban Green Space Information Based On Decision Tree And Decomposition Of Mixed Pixels

As the regulator of the urban ecological balance, Green space plays an important role in construction of ecological city. It can beautify the environment, clean air and improve the climate. In addition to these ecological functions, it also can provide entertainment and sightseeing places for residents, and meet the needs of the people’s spiritual and cultural life. Green space can improve the naturalness of city landscape, and promote the harmonious development of the urban residents and nature, therefore, green space has become an important symbol of urban modernization and civilization. In recent years, China’s urbanization process is accelerated, and land contradiction is obvious, resulting in a sharp reduction of green space. These lead to the emergence of a variety of the city problem, such as "urban heat island effect", heavy rain, dust storms and other disharmonious natural phenomenon. Therefore, there is an urgent problem that how to protect and reasonably distribute the green space.Remote sensing technology provides important technical means for obtaining accurate green space information on account of its merits in its macro in its cyclical, economic, and objectivity aspects. In this paper, TM remote sensing image is applied, and the combination decision tree and mixed pixel decomposition is used to extract urban green space information. We combine water index, building index,vegetation index, the principal component analysis and tasseled cap transformation to build decision tree and extract the green space information (grassland,coniferous vegetation and broad-leaved vegetation); And then the MNF transform and PPI are used to purify endmember, and N dimensional scatter plot is applied to determine endmember. Finally, abundance value of green space obtained by linear decomposition transformation multiply by a single pixel area, we get the green area.The combination of decision tree and decomposition of mixed pixel is used to extract the green information, and accuracy is above87%. Compared to decision tree,classification accuracy is improved by more than4%,showing superior to conventional classification. The results show that:the combination of decision tree and decomposition of mixed pixel is used to obtain information for Moderate-Resolution remote sensing data, which improves the monitoring precision and provides objective reference for urban planning.