High Resolution Remote Sensing Change Detection Of Urban Typical Features Considering Parallax And Shadow

The rapid development of urbanization makes a large number of people gather in the city,which leads to a series of urban problems,such as the continuous increase of construction land,the arbitrary occupation of cultivated land and forest land,and the deterioration of ecological environment are attracting people's continuous attention.Remote sensing change detection technology can provide scientific decision-making basis for mastering urban land use change timely and accurately,containing and preventing of blind urban expansion.In urban remote sensing change detection,buildings and vegetation are often regarded as typical "target" features of change analysis.The detection results can effectively reflect the process of urban development,and provide favorable basis for urban planning and sustainable development.However,how to extract the change information of urban typical features accurately is facing many difficulties.The difference of radiometric images from different temporal and different sensors makes it difficult to identify the real change information;the building target in the city area is scattered,and the parallax in the simultaneous interpreting images is different;hazy haze and various shadows seriously affect the accurate identification of the target objects.The image quality and content of these specific environments have a direct impact on the accuracy of change detection of urban typical features,which restricts the accurate extraction of change information.Therefore,this dissertation focuses on the prominent problems in the change detection of urban typical features,and studies the change detection methods from four aspects: image relative radiation correction,building parallax analysis of two-phase image,removal of urban vegetation shadow area and change detection of urban typical features based on information fusion.The purpose of this dissertation is to make use of a variety of image information and intelligent methods to reduce the influence of error in the process of preprocessing and detection,so as to enhance the accuracy and rationality of the change detection results of typical urban features,build a new change detection method,and provide theoretical support for the application of change detection technology and production practice.The main research work and innovations are as follows:(1)When there are many changes in two-phase change detection,the histogram mean and variance of DN value of remote sensing images from the same source and different source are quite different,which results in pseudo invariant feature(PIF)points in the process of relative radiometric correction PIF is easy to be wrongly selected.Based on the Dempster Shafer(DS)evidence theory,a decision fusion algorithm of space-spectrum multi feature PIF point selection is proposed to carry out the research of relative radiometric correction of two-phase remote sensing image considering the category of ground features.This method uses histogram of oriented gradient(HOG),pixel shape index(PSI),Structural Similarity(SSIM)three kinds of features that are not sensitive to the brightness difference between multi-source remote sensing images,combined with DS evidence theory decision fusion and correlation coefficient method to select PIF points,because of the use of relatively stable image features and the advantages of two-level analysis scale,the reliability of the selected PIF points is better.(2)Aiming at the problem of misdetection of building changes caused by parallax in the high-resolution remote sensing image of urban area,a method of building change detection based on parallax analysis(PA)is proposed.This method starts from the analysis of the geometric relationship between the shadow and the building.Based on the properties of the geometric characteristic triangle of the building,the relationship between the architectural shadow and the roof,the architectural shadow and the height of the building,the architectural shadow and the parallax is deduced.This relationship is applied to the change detection of urban high-rise buildings,which effectively solves the problem of disparity uncertainty caused by the height change of buildings in the two-phase image,and then reduces the false detection of changes caused by parallax.(3)Aiming at the problem of spectral uncertainty caused by haze and shadow,and the problem of insufficient use of spatial information based on pixel analysis,a method of urban vegetation information processing based on vegetation information enhancement and snds segmentation is proposed.In terms of vegetation information enhancement:(1)haze removal through improved HSV transformation.(2)A parallel edge shadow removal(PESR)method is proposed to solve the problem of irrationality in the selection of reference area of conventional methods.Information post-processing: through the information post-processing method of Spatial Neighborhood Density Segmentation(SNDS),the extracted vegetation information is further processed and spatial information is added to enhance the reliability of the extraction results.(4)In order to improve the universality of change detection methods for typical urban features,a change detection method based on DS multi-feature decision fusion and multi-scale uncertainty analysis(DS-MUA)is proposed.Two decision level fusion methods are introduced to fuse the change intensity.In the aspect of feature fusion,the weighted DS evidence fusion is used to fuse the structural similarity of each feature with multi-source change intensity information to reduce the uncertainty of detection results.The test results are output as changed,unchanged and uncertain.In view of the uncertain image spots,the multi-scale uncertainty analysis is used,and the sequence from large scale to small scale is used to get the relatively reliable change information.The method is applied to the change detection of urban vegetation and buildings,and the effectiveness of the method is verified.Further research on Optimization of change detection results based on method fusion is carried out.In the aspect of shadow removal,three methods,PESR,SNDS and DS-MUA,are used to detect the change of urban vegetation,so as to enhance the accuracy of vegetation information extraction;in the aspect of parallax reduction,the advantages of PA and DS-MUA are combined to achieve the elimination of parallax effect while taking into account a variety of building types.The experimental results show that this method combines the features of the target features to form complementary advantages;at the same time,it uses the scale context information to avoid the shortcomings of the multi-scale fusion method which processes the results of different scales separately;and it can be applied to the change detection of a variety of urban typical features.