Synergistic Use Of WorldView-2 Imagery And Airborne LiDAR Data For Urban Impervious Surface Extracting Method

High spatial resolution optical imagery provides detailed information of urban objects for impervious surface extraction,but there are lots of uncertainties resulting in extraction errors,including same objects with different spectral signatures,similar spectra of different objects,shadows of buildings,and roads covered by large tree crowns.Synergistic use of multi-source remote sensing data is considered as an effective approach to reducing these limitations.Compared to optical imagery,airborne Light Detection and Ranging(LiDAR)data records spatial three dimension information of urban objects in the form of discrete point clouds,and by which we can generate accurate height image and avoid the shadows in the high spatial resolution images.However,Li DAR lack the spatial features of urban land covers.Therefore,Synergistic use of high resolution optical images and airborne Li DAR data achieve the complementary advantages and finer implement impervious surfaces extraction.In our research,three sites in rural,rural-urban and urban subsets,located on Pingdingshan City,Henan Province,were selected as the study areas.A seris of approaches for extracting impervious surfaces from WorldView-2 imagery and airborne LiDAR data were proposed,including pixel-based Support Vector Machine(SVM)hierarchical classification and threshold-based impervious surfaces extraction method,object-oriented SVM hierarchical extraction,and hierarchical extraction based on the modal of three dimension Convolutional Neural Network(CNN).The purpose of our research are exploring the potential of improving impervious surfaces extraction combined using of high resolution optical imagery and airborne LiDAR data,excavating the adventages of object-oriented classification and CNN modal in impervious surfaces mapping in the different types of urban areas.We will provide some noval ideas and scientific guidances for impervious surfaces extractions using multi-source remote sensing data.According to proposed methods,our research made three major comparisons,including:(1)extraction from WorldView-2 images alone and WorldView-2+airborne LiDAR data;non-hierarchical and hierarchical classification;(2)pixel-based and object-oriented hierarchical extraction;(3)impervious surfaces extraction using Support Vector Machine(SVM)and CNN model.The major results and findings of our research show that:(1)Compare with high resolution optical imagery alone,high resolution optical and airborne Li DAR fused data can solve the problems of same objects with different spectral signatures,similar spectra of different objects in buildings and roads extraction.(2)Impervious surfaces extraction method based on spectral and height thresholds can identify the impervious surfaces in the shadowsfrom tall buildings,optimize roads extraction and detect roads coverd by large tree crowns.The approach improves the accuracy of impervious surfaces extraction.(3)Compared with pixel-based extraction,object-oriented hierarchical extraction method takes full advantages of some merits of LiDAR height image,including the orthographic,non-shadows images,and the buildings objects can be segmented exactly.Using our method,the shape of buildings can be described accurately and the imperviousness and vegtations in shadows are extracted robustly and not relied on thresholds.The problem of boundary of tree crowns are misclassified to buildings because of the inconsistence of sensors between optical image and LiDAR data is solved based object-oriented hierarchical method.(4)Compared with SVM classification,the extraction using three dimensions CNN modal improve obviously the confusion between imperviousness and bare soils and distinguish automatically the roads and vegatations in the shadows.