Urban Vegetation Estimation Derived from QuickBird Based on Object-oriented Method

Urban vegetation is considered a crucial factor for realizing environmental quality goals and foreseeing sustainable local development in the framework of Local Agenda21 guiding environmental politics. High-resolution image offer abundance function in improving method of urban environment and guideline of evaluation. The method of Object-oriented classification, considered both spectral characters and structural information, can provide high-resolution images´ classification precision and reduce large data redundancy compared with the traditional pixel-based classification methods. In this paper, we use object-oriented method to extract different green space based on various urban environmental land use, such as commonality green, street green and so on.. The QuickBird image is segmented to a set of different scale image, we have make many experiment in order to obtain the optimal segmentation of every level, so the image is segmented by scale 6, and then each object is classified into non-vegetation and 3 kinds of vegetation classes based on the nearest neighbour classifier (NN), Due to the effect of mixed pixels, the second segmentation level scale 5 is carried out into the non-vegetation class and several omitted vegetation objects are derived. Finally, we use a GIS mapping data combined with the results of vegetation classification to separate urban vegetation area into different usages (i.e. commonality green, street green...). Through analyse to the result of classification, the area of the vegetation is 597.63 hm², the urban garden fraction is 25.38%. This method offers a new application approach for classifying the high-resolution remote sensing image and the result of accuracy can reach 98%.