遥感卫星图像系统几何校正多级并行算法  

Multi-level Parallel Algorithm of Systematically Geometric Correction for Remote Sensing Satellite Imagery

作　　者：李雨航[1,2] 李景山[1] 赵仕美 

机构地区：[1]中国科学院遥感与数字地球研究所,北京100094 [2]中国科学院大学,北京100049

出　　处：《遥感信息》2016年第3期51-55,共5页Remote Sensing Information

摘　　要：针对遥感卫星图像数据量大、系统几何校正计算复杂的问题,提出了基于SMP机群的系统几何校正多级并行算法。该算法利用MPI+OpenMP并行编程技术,节点间实现进程级粗粒度的并行,节点内实现线程级细粒度的并行。采用基于冗余存储的数据划分方式,保证了各个节点的负载均衡,减少了数据定位的复杂度;利用并行文件系统进行数据分配,避免了节点间的数据搬移,实现了数据并行读写,节点内部的并行,进一步细化了算法的并行粒度。在SMP机群系统上对资源三号卫星正视相机图像进行算法验证。结果表明,该算法充分利用了SMP机群的计算资源,具有良好的并行性能。To solve the problem of huge amount of remote sensing satellite image data and complex calculation in systematically geometric correction,a multi-level parallel algorithm based on SMP cluster was proposed.Adopting the MPI and OpenMP hybrid programming model,the systematically geometric correction algorithm was implemented in coarse-grained parallelism manner between nodes and fine-grained parallelism manner within nodes.The data was divided based on the redundant storage,which ensured the load balancing and reduced the complexity of the data positioning;distributing data on parallel file system avoided data transit between nodes and enabled data reading and writing in parallel;parallel algorithm was implemented in intra-node,which further fined the parallel granularity.Finally we tested this multi-level parallel algorithm with the NAD image of Ziyuan-3(ZY-3)satellite.Results show that this multi-level parallel algorithm makes full use of computing resources of SMP cluster and improves the efficiency of systematically geometric correction greatly.

关 键 词：SMP机群 系统几何校正 多级并行 MPI OPENMP 加速比 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]