Information Extraction Method And Application Research Of City Land Utilization Based On Multi-source Remote Sensing

With the deepening of the urbanization process of Jiangsu Province, Peixian newindustrialization, striving for garden city greening projects implemented step by step, change ofland use pattern city area within the scope of the accelerating. This paper is based on thisbackground, the remote sensing image data using the Peixian part of the region of the2011Geoeye-1multispectral and high resolution image in2010and Worldview-1high resolution,using unsupervised classification method, maximum likelihood classification, decision treeclassification, support vector machine classification and object-oriented classification method,eCognition, produced by the user comparative analysis of accuracy, kappa coefficient, theoptimal classification method was selected, and the related survey data and government planningand construction of the file, to carry on the analysis to classification results of the two images andchanges, and the changes of driving force has been discussed from economy, policy point of view,obtained the following conclusions:First, In view of city land use information extraction on the image resolution and spectralinformation requirements, through the pretreatment of the original satellite images, a new imagefusion into the amount of information is more abundant, the higher resolution, improving imagequality. So we adopt Brovey, HIS, PanSharpen, Gram-Schmid, PCA, Wavelet fusion method offusion of WorldView-1panchromatic and multispectral images based on Geoeye-1, reduction, intraditional spectral texture index, improved vegetation weight similar quantitative evaluationmethod is introduced, focusing on the image structure information and spectral and texture thedegree of balance, and using MatLab to construct the evaluation program, through thecomprehensive analysis and comparison of G_MMSIM and conventional index, that the fusioneffect of PCA fusion method and visual interpretation of surface features is relatively close,suitable for classification in the study area.Secondly，Using remote sensing image classification method is different, the remote sensingimage texture information, vegetation, water, building index feature as the assistant classificationinformation, using unsupervised classification, phase element maximum likelihood classification,decision tree classification, support vector machine classification and eCognition classificationbased on object-oriented method based on partial data, through the combination of Peixian greenspace survey in the analysis, the overall accuracy and kappa coefficient index, obtained thedifferent classification results. The overall accuracy of unsupervised classification with aminimum of78.75%and0.78kappa coefficient; the overall accuracy of maximum likelihoodclassification with84.82%and0.79of the kappa coefficient; decision tree classification resultswith an overall accuracy of85.83%and about0.80of the kappa coefficient; support vectormachine classification has been an overall accuracy of88.50%and0.82of the kappa coefficient; finally the ecognition classification the overall accuracy and kappa coefficient to obtainmaximum, respectively90.50%and0.84, the eCognition is the image classification method is themost suitable.Thirdly, According to the classification of remote sensing image data eCognition, statisticsof each kind of land use information and changes, obtained the study area of bare land, water areain2011than in2010declined slightly, land, cultivated land, forest land and construction andtraffic land rising trend results.