Object-oriented High-resolution Remote Sensing Image Classification And Applied Research

Using the QuickBird image data of Dr. Sun yat-sen's Mausoleum national park in 2006 and introduces object oriented classification concept and technique into landscape information obtaining field. Using mufti-scale segmentation method and then based on the analysis of spectral information of ground feature, shape feature,texture,vegetation index and then feature space of classification based on the characteristics of the study area is constructed and optimized to bring out the information getting of study area. Using the model of CASA to estimate the terrestrial net primary production(NPP) of Dr. Sun yat-sen's Mausoleum national park. The main contains and results are as follow:1.Using object oriented classification to exact information of Dr. Sun yat-sen's Mausoleum national park and using spectral information,geometric information,patial information of high-resolution image flexibility. By setting up member function to construct classification rules. This paper uses the informaition of spectrum,texture,geometry,space etc. To improve the classification accuracy greatly and get sort resoults more orderly and realistic. This shows that object oriented classification has applicability in the information extraction on high-resolution remote sensing image.2.Using mufti-scale segmentation method of relatively fit interval scale and analyzes objects of size and form basing on mufti-scale segmentation. We choose the most appropriate segmentation scale parameter by visual discrimination and extraction purposes. First using large-scale segmentation to exact and classify as vegetation,water,impermeable surface(construction and road),shadow(vegetation shadow and construction shadow). And then we exact vegetation and vegetation shadow by mask method. Using the small segmentation scale to exact and classify as farmland,bamboo forest,coniferous forest,broad-leave forest,coniferous broadleaves mixed forest. Mufti-scale segmentation uses region merging algorithm to generate and exact objects likes as people's interpretation model. This method can avoid mixing among different types of objects effectively and improve the classification accuracy.3.Based on the Light Utility Efficiency and CASA model, we set up a model to estimate the terrestrial net primary production(NPP) of study area which the QuickBird data is used as main remote sensing data. We estimate the terrestrial net primary production of Dr. Sun yat-sen's Mausoleum national park and compare the terrestrial net primary production of some vegetation preliminary.