Object-oriented Classification Technology Of High Resolution Remote Sensing Image For Geographic National Conditions Surveying

Land cover information of geographic national conditions surveying is the basic data of geographic national conditions monitoring, extracting land cover information timely and accurately has become one of important tasks in the geographic national conditions surveying project, whose main data is very high resolution(VHR) remote sensing images. From the perspective of the key technologies and application of classification of land cover of geographic national conditions surveying, using the geospatial object based image analysis (GEOBIA) technology to carry out research on land cover classification method for geographic national conditions surveying, which aiming to draw land cover interpretation solutions suitable for geographic national conditions surveying. The main research contents and results are as followings:(1) Choosing the best segmentation scale. The multi-scale segmentation technology and multi-scale parallel segmentation technology based on MPI have been studied, meanwhile, a series of experiments are carried out to test the validity of the multi-scale parallel segmentation technology. Then, it takes very high resolution(VHR) remote sensing images as original data, uses classification feedback method to study the scale problems which existing in remote sensing. Through numbers of tests, the thesis obtains the best scale for different resolution remote sensing images at the same time.(2) Analyzing and selecting the optimal features. The spectral, shape, texture features which are commonly used in land cover classification have been quantified. The feature analysis for geographic national conditions surveying is conducted and obtains features suiting different types. In the end, through the GEOBIA classification feature analysis experiments based on typical experimental areas to get the effect of the number of features on the different classification algorithms and provide advising to feature selection and extraction of every types.(3) Comparative analysis of GEOBIA classification methods. Firstly, four GEOBIA classification algorithms for geographic national conditions surveying including semantic modeling classification, Support Vector Machine (SVM), Decision Tree (DT) and Random Forest (RF) are analyzed and discussed. Secondly, it takes three VHR images which have different resolution and cover three typical areas including mountain, plain and city as experimental data, analyzes comparatively these classification algorithms from the aspect of classification results, classification accuracy, stability and time consuming and analyze the advantages and disadvantages of every classification algorithms at the same time. At last, on the basis of application, it gets the optimal classifier combination scheme and the land cover classification discipline under different environments.