Study On GA-RLS-RFM Ortho-rectification Optimization Algorithm Based On FPGA For Satellite Images

One of the prerequisites for making thematic maps is to orthorectify satellite images in advance.However,traditional ortho-rectification methods based on groundbased processing platforms are hard to meet the application scenarios which have high timeliness requirements,such as the rapid response for disaster emergency rescue,and the real-time monitoring for fixed targets,etc.Moreover,the traditional method to collect ground control points(GCPs)for image ortho-rectification usually requires human intervention and is time-consuming,thus the timeliness of ortho-rectification can't be ensured.Particularly,it is very difficult or even impossible to collect GCPs in the areas with less obvious ground characteristics,such as mountains and deserts in harsh environments,as well as in the overseas military sensitive zones.Therefore,in order to quickly ortho-rectify satellite images under little or no GCPs,orthorectification optimized algorithms of satellite images FPGA-oriented are systematically studied.The main research contents are as follows.(1)Based on the previous study results of other researchers,the Viewing Geometry Model(VGM)of linear array push-sweep satellite images is established using the ephemeris and attitude data.In addition,the FPGA hardware architecture of the VGM positioning algorithm of satellite images is designed using the pipeline structure and the design method combining data flow serial computation with modules parallel computation.The designed hardware architecture is able to balance the processing speed,calculation accuracy and the utilization ratio of hardware resources.(2)Generally,it needs to perform the complex multiplication and inverse operation for the large matrixes,when using the Least Square(LS)method to solve Rational Function Model(RFM)parameters.However,these complex operations not only consume large FPGA hardware resources,but also affect the speed of solving the parameters of RFM model.To overcome the disadvantages of the algorithm of solving RFM model parameters using LS,which are not good for FPGA hardware implementation,the algorithm of solving RFM model parameters using Recursive Least Square(RLS)is proposed.The RFM model determined by the proposed algorithm is named RLS-RFM model.In addition,the FPGA hardware architecture of the proposed algorithm is presented.The proposed FPGA hardware architecture adopts a fast-parallel matrix multiplication structure to speed up the parameter solution of RFM model.(3)Since the correlation between parameters of RLS-RFM model would affect the correction accuracy,thus,to quickly obtain the best structure of RLS-RFM model and improve the accuracy of ortho-rectification,the GA-RLS-RFM ortho-rectification optimized algorithm for satellite images FPGA-oriented is proposed and implemented.Particularly,the designed FPGA hardware architecture of GA-RLS-RFM orthorectification algorithm can be used to implement the ortho-rectification without GCPs by RLS-RFM model,when GCPs are not available.(4)The experiment results show that:(i)The VGM model of satellite images has the potential of high positioning accuracy without GCPs.(ii)The RLS algorithm can effectively solve RFM model parameters,and the RLS-RFM model can achieve the same correction accuracy as VGM model.(iii)the Genetic algorithm can effectively reduce the number of RLS-RFM model parameters,and maintain,even improve the correction accuracy.(iv)The calculation precisions of FPGA are consistent with PC.For example,when performing the ortho-rectification of images without GCPs using the RLS-RFM model,for SPOT-6(mountain)image and SPOT-6(airport)image,the maximum deviations between FPGA results and PC results in column direction are0.0782 pixels and 0.1026 pixels,respectively;and in row direction are 0.1302 pixels and 0.1380 pixels,respectively.(v)For the speed of data processing,FPGA has advantages over PC.