Extraction and classification of urban land cover based on multisource data and their applications in Chongming

Abstract

The distribution and allocation of different land cover classes is related with resource exploitations, environmental pollution controlment and human habitat environment quality. In this paper, the multi-source data including high resolution RS imagery-FORMOSAT-2 image and digital topographic maps are applied to acquire the information of urban land cover by taking Chongming as a case study. Firstly, the overall framework is proposed to apply multi-source data to extract and classify urban land covers. Then, some classes of land cover are extracted and the high resolution RS imagery is classified based on C5.0 decision tree classifier. In the feature library of different urban land covers established, there are three features: spectral feature, texture feature, and shape information. Spectral and texture features are acquired from the RS imagery, and shape information is computed from digital vector maps using ArcGIS. Based on multi-feature, the classification model via C5.0 decision tree is constructed to realize the urban land cover classification and extract different land cover classes. Finally, classification accuracy and results are compared between this method and other conventional classification methods. This method proves to improve the classification accuracy more effectively.