A Quick Orthographic Rectification Approach For Optical Remote Sensing Satelite Imagery Based On CUDA Architecture

With the advancement of science and technology,optical remote sensing satellite to ground observation technology is booming,has entered the data acquisition way diversification and information eruption of the new era.In recent years,high-resolution optical satellites,represented by domestic satellites ZY3,GF1,GF2 and foreign satellites IKONOS,GeoEye series,Pleiades,WorldView series,have been launched,and most high-resolution satellites have reached the sub-meter level.The image product size is designed to tens of thousands of pixels,therefor image process of data acquisition,product production,information extraction and practical applications must be extremely time-consuming.On the other hand,the real-time nature of satellite data processing is rigorously required in certain areas.The entire process from image data to final utility product must be completed in minutes to provide reliable and effective information for subsequent intelligence analysis,and so on.If the product can not be completed in time and useful information cannot be extracted,it will not be able to play the rapid,quantitative and comprehensive effect of remote sensing technology,resulting in the backlog of remote sensing data,loss of practical value.Therefore,it is necessary to study how to use all kinds of new high-performance computing methods to meet the real-time requirements of some remote sensing applications.At present,GPU(Graphics Processing Unit)high-performance computing architecture can be limited in the environment to provide high-performance computing power,GPU hardware architecture on the design of efficient operation of the algorithm for the above problem provides a viable solution,and has gradually become international To solve the large data calculation and real-time processing of the mainstream program.In the actual application scenario,the orthophoto image of the target area must be obtained through the orthorectification process,but the orthorectification process is intensive and takes a long time.Aiming at the acceleration problem of orthorectification,this paper designs a parallel algorithm based on CUDA(Compute Unified Device Architecture)architecture,which deals with all aspects of orthophoto correction on GPU,and combines the characteristics of GPU and parallel computing.Accelerated optimization,rapid processing of remote sensing image orthorectification.The specific contents of this paper include the following three aspects:(1)RPC model refinementOptical satellite imaging products generally provide RPC files as image orientation parameter files,but due to the possibility of satellite positioning is not allowed,resulting in RPC model error.If you do not refine the correction,correct the results of the image there will be a large positioning error.In this paper,the RPC model is refined by the control point covering the image area,and the control point is obtained by matching the original image with the corresponding DOM control point.(2)parallel algorithm improvement and optimizationThe parallel improvement of the orthogonally corrected serial algorithm under the CUDA programming framework is an effective means to realize the fast orthorectification.This paper first analyzes each aspect of the orthorectification and its improvement method,and then makes the parallel part of the parallel improvement Finally,a parallel algorithm is proposed to optimize the parallel algorithm.Finally,a highly efficient fast orthorectification algorithm is established.(3)rapid orthorectification correction results evaluationIn this paper,the results of fast orthorectification are evaluated from the aspects of accuracy and optimization efficiency.Several sets of contrast experiments are designed,and the feasibility and practical value of fast orthoretic correction algorithm are analyzed.