Research Of GPU Arrays Parallel Algorithm For Remote Sensing Image Geometric Correction

Information technology and sensor technology’s rapid development,results in a sharp increase in the amount of remote sensing image data.At the same time,the accuracy requirements and real-time requirements of practical applications are also constantly improving.Traditional remote sensing image parallel processing based on multi-core CPU or CPU cluster has better acceleration effect on large scale cluster,but it has higher cost,higher power consumption and is not suitable for satellite application environment with limited CPUs.GPU(Graphic Processingg Unit),due to its multi-core parallel features,brings powerful data processing capabilities.It also has the advantages of low power consumption and low cost.Therefore,the use of GPU array to accelerate remote sensing image processing is of great significance for the construction of high performance,low power consumption and low cost real-time computing system on satellites.In this paper,cross-platform features of OpenCL(Open Computing Language)are used to realize GPU calls on different platforms,so as to make full use of all computing resources.At the same time,OpenMP and MPI technologies,which are used by traditional multi-core cpus or CPU clusters in parallel,are introduced to realize the call of GPU array,forming different levels of parallel computing architecture: Multi-core parallel architecture based on single-node single-gpu,multi-thread + multi-core parallel architecture based on single-node multi-gpu system,multi-process + multi-core parallel architecture based on multi-node multi-gpu system,multi-process + multi-thread + multi-core parallel architecture based on multi-node multi-gpu system.The effects of different task partitioning models on computing performance are analyzed for collaborative computing of different GPU arrays,and the optimization strategy is in-depth studied according to the GPU architecture.Aiming at the geometric correction process which is the most complex and intensive to calculate in remote sensing image preprocessing,this paper makes an in-depth study of the workflow of geometric correction algorithm.Through parallel analysis of different coarse and fine granularity,parallel acceleration schemes are designed according to different resolution of remote sensing images,so as to map the parallel geometric correction algorithm to parallel computing architecture of corresponding levels:For the geometric correction of remote sensing images with low resolution,the fine-grained OpenCL implementation under different computing devices on different platforms is verified.The memory model of OpenCL is used to study the optimization strategy,and the geometric correction effect and acceleration ratio of different interpolation methods are verified.For the middle and high resolution satellite remote sensing images,on the basis of direct data partitioning,redundancy data partitioning and local output partitioning based on the applicability of single GPU processing and GPU array parallelism are respectively improved to achieve coarse-grained partitioning,and then fine-grained partitioning is used to block images.Finally,in the hardware environment of GPU arrays on different platforms,the parallel optimization of geometric correction algorithm is designed.The experimental results show that the parallel scheme and optimization algorithm can greatly improve the processing efficiency of geometric correction of remote sensing images.