Research On The Object-oriented Method Of City Features Extraction With GF-2 Remote-sensing Imagery

With the development of the remote sensing technology of China,we can obtain sub-meter multispectral comprehensive optical remote sensing data with GF-2 satellite remote sensing data as a typical example.The maximum resolution ratio of panchromatic band of GF-2 remote sensing data reaches to 0.8 meters and the spectral resolution is as high as 3.2 meters,and moreover,it is featured with high location accuracy,high precision of radiation and fast motor responsiveness.At present,urbanization of China has a rapid development,thus acquisition of urban geographic information and timely updating of data is an important element task of urban delicacy management.The application of GF-2 information in urban ground-object recognition and sophisticated category can greatly improve extraction accuracy.Spatial information of GF data is rich,but spectral resolution is low,spectral information is weak and its spectrum statistical property is lower than stability of medium-low resolution images.Moreover,GF-2 data has few bands and spectral information is not sufficient,which are used in pixel classification method,and “different objects same image” of spectral information would affect classification results.Therefore,GF-2 urban terrain classification study should select proper method to avoid data disadvantages as far as possible and apply data advantages.This study adopted object-oriented classification method.Basic units that object-oriented classification method processes are no longer pixel but objects generated by pixel according to certain rules and sizes.Objects generated at different levels can construct a topological relation among them and as basic units,objects possess many attributive characters.A logical deduction and feature combination of attributive characters can realize discriminant analysis of spatial pattern and accurate classification of objects.The data size of GF-2 data is large,spatial characteristics are rich but spectral information is not sufficient,thus object-oriented classification can better use data geometry and structural features.Furthermore,the complex social attributes of cities contribute to the complex and diversified ground features and different land sizes,and object-oriented classification method can give full play to advantages of high spatial resolution of data,make the best of the rich spatial information and apply spatial arrangement,textural features,geometrical characteristics and spectral features of space in the classification process.Object-oriented classification method can avoid pepper salt,noise and a large number of disengagement in classification results and obtain better classification results and higher precision of classification.Object-oriented classification method can construct reasonable target extraction levels,rules and procedures according to specific data and targets to guarantee extraction effects,which can be applied in mass data once for all.Objects and procedures of the method are closer to analytical methods of human brains,the process is easy to be understood and the parameters are easy to be adjusted.Therefore,object-oriented classification method can greatly reduce artificial interpretation,and it is an automatic and intelligent method.The study sorted urban ground information based on the object-oriented classification method of GF-2 remote sensing images.The research contents mainly included: preprocessing of GF remote sensing images,object-oriented theory summary,object-oriented partition methods,object-oriented classification methods,hierarchy construction of classification rules and assessment and comparison of classification accuracy,and so on.The data involved in the study was tested by selecting the GF-2 remote sensing images of Dongcheng district,Beijing.Data preprocessing adopted ENVI software,and later,the date would be input into eCognition software to extract urban ground information with the object-oriented classification.In the end,the essay used eCognition software to derive the obtained classification results in the grid map,and stocked the vector diagram.The overall precision of extraction accuracy of urban ground information by use of the object-oriented method reached to 91.15% and the kappa coefficient is 0.8893,which has confirmed the superiority of the object-oriented method and the feasibility of the research method,and the research was meaningful.