| Urban Master Planning of Chengdu City(2016-2035)put forward the requirement of constructing urban ecological security barrier and the new model of exploring the urban green development including the Western High-tech Zone which has raised the new requirements to the management and protection of urban green space system.And how to correctly identify the spatial distribution law of urban greening tree species is the basic prerequisite for scientific management of urban green space system.The rapidly development of high-resolution remote sensing technology provides new technical methods for the information construction of urban green space and the classification of green space resources.When using high-resolution remote sensing images to identify tree species,selecting appropriate classification features and scientific classification methods has important influence on the classification results and its accuracy.This paper takes the remote sensing data of the Western High-tech Zone,Pleiades,as the research object,carrying on the pre-treatment of radiometric calibration,atmospheric correction,image fusion preprocessing and 8 kinds of main tree species measured in research area spatial information,using ENVI 5.1 to extract and analyze the spectral characteristics,vegetation index and texture features of 8 kinds of selected objects,then finally contrast the classification results and accuracy under the Maximum Likelihood Classification method and Support Vector Machine Classification method.The main research results are as follows:(1)The best band combination of Pleiades’ s remote sensing images is RGB431.This combination can reduce the inter band interference and preserve the image data information to the best extent while providing the best visual effect.After analyzing 4 fusion methods which are Gram-Scmidt transform,Pan sharpening transform,Principal Components Analyze(PCA)and Brovey transform from two aspects of qualitative and quantitative comparison,the optimum method for the study of Pleiades image fusion is Gram-Scmidt transform,which can retain and reflect the image spatial information and spectral information to the best.(2)Whether under the SVM or the MLC,combining the spectral characteristics,the vegetation index with texture features can achieve better effect than the classification results of using single one or others.This method shows that the appropriate combination of each feature helps to improve the accuracy of species identification.(3)Regardless of using the image spectral characteristics or combining with a variety of classification features,SVM behaves better than MCL,with a classification accuracy up to a maximum of 85.29%(Kappa coefficient 0.827).This shows that SVM can effectively excavate the effective classification features and behaves more stably.Therefore,the SVM is the best classification method in this study.(4)The highest producers accuracy of the Broussonetia papyrifera,Magnolia grandiflora,Ficus virens,Salix babylonica,Metasequoia glyptostroboides,Cinnamomum japonicum,Ficus benjaminaand Ginkogo biloba are 71.77%,74.42%,86.11%,75.90%,82.73%,84.54%,78.45% and 84.29%,with the highest user accuracy 79.05% 75.96%,84.07%,78.37%,87.29%,78.24%,84.02% and 88.79%,which means the classification effect is ideal.The classification accuracy of Ficus virens,Metasequoia glyptostroboides and Ginkogo biloba are the best whose producer accuracy and user accuracy are above 80%. |
