Detecting urban vegetation from IKONOS data using an object-oriented approach

Urban vegetation plays a very important role in urban planning, environmental protecting, and sustainable development policy making. High resolution remote sensing imageries like IKONOS can provide such information economically and timely. However, the traditional classifying method based on pixel results in poor accuracy for the increased resolution causes not only in inter-class spectral variability but also intra-class spectral variability. Thus, developing a new method to extract thematic information from high resolution remotely sensed data might become one of the most challenging tasks of remote sensing within the coming years. This research proposed an object-oriented method to obtain the distribution of grass and tree in urban environment. The whole process included two steps: first, the original IKONOS data and the derived data of NDVI and VI were segmented at scale 5 and then, the objects obtained from this finer segmentation process were classified into two categories (vegetation and non-vegetation) using the feature space composed of mean NDVI, ratio near- infra, mean VI, ratio green band, and mean red band; second, the data of IKONOS, NDVI and VI within the vegetation area were segmented at scale 100, and then the objects from this lager scale segmentation were classified by the feature of Stddev red- infra into grass and tree. The result of accuracy assessment showed that the proposed method had produced distributed vegetation with over 97% overall accuracy.