Studying On Change Detection Of Buildings In High Resolution Remote Sensing Image Based On Pseudo-orthorectification

Using remote sensing images for change detection can provide powerful means for environmental monitoring,resource exploration,disaster assessment and geographic database update.In recent years,with the rapid development of earth observation technology,a large number of optical remote sensing imaging satellites with high spatial resolution have emerged at home and abroad,such as World View and GJ-1,etc.,but they also bring new challenges to the processing of remote sensing images.For change detection,using high resolution remote sensing image change detection is the developing direction of the change detection technology,however,change detection technology is still under the influence of various factors and make the detection accuracy is not guaranteed,is one of the main problems in making is unable to get high precision DSM direct change detection,it is impossible to effectively solve the phenomenon that the image point displacement caused by satellite side-view imaging results in a large amount of elevation information of the building,which leads to false detection.However,the reality is that most of the high resolution satellite image information will be kept a lot of building elevation information.In the case that real projective images cannot be generated,how to effectively avoid the false detection caused by image point displacement is an urgent problem to be solved.Therefore,in view of the above problems,this paper conducts the following research:(1)Aiming at the problem of false detection of change detection caused by perspective distortion in high-resolution image,an approximate orthorectification algorithm based on pseudo-orthorectification is proposed to eliminate the building elevation information in remote sensing image.First of all,this paper defines the concept of pseudoorthorectification,and then describes the technical process of pseudo-orthorectification in detail.In this paper,the exact length of building elevation contour is used as the displacement length of building points,and the roof of each building is translated according to its displacement length,then pseudo-orthorectification image can be generated and building elevation information can be eliminated.(2)For the problem that the registration accuracy of high-resolution remote sensing images is not ideal,a global-local registration method based on Delaunay triangular secondorder polynomial transformation is proposed.Firstly,use IMAS(Image Matching by Affine Simulation)algorithm to extract the initial matching points at different times images;then use a matching algorithm based on triangulation photometric and geometric constraints to obtain accurate matching points for global registration;finally,after triangulation optimization and merging,uses second-order polynomial transformation for local registration to obtain accurate multi-temporal registration images.The experimental results show that the accuracy of the registration method can reach less than 1 pixel,and it can well solve the distortion in the registration process.(3)Method for building changes detecting in remote sensing image based on pseudo orthorectification.First,a method for extracting buildings from high-resolution remote sensing images based on Grab-Cut with automatic sample selection is proposed.The specific method steps are as follows.First,by using the image's own RPC file,the image point displacement direction of a single image is calculated by following the projection trajectory method,so that the image is rotated to make the image point displacement in the horizontal direction.Secondly,the image is denoised,the edge of the image is extracted by the Canny,and the elevation contour is extracted by combining with morphological operation and Hough transformation.Thirdly,combining the corner information and object space positioning consistency method,the elevation contour is selected accurately.Finally,using the building elevation contour as the constraint and combining with the geometric rules,a building extraction method based on Grab-cut with automatic sample selection is used to extract the accurate roof of the building.The experimental results show that the average precision of this method can reach 86%.Finally,the building information is combined with NDVI,NDWI and MSI to obtain the change information through Local Coregistration Adjustment algorithm,and the real remote sensing image is used for verification.The experimental results show that the accuracy of building change detection can reach 91.94%%,which is higher than that of the method without pseudo-orthorectification.