基于CUDA的压缩感知重构算法并行化研究  

Research on the parallelization of compressed sensing reconstruction algorithm with CUDA

作　　者：董蕾[1] 黄方[1,2] 卜栓栓[1] 冯杰[1] 周纪[1] 

机构地区：[1]电子科技大学资源与环境学院,成都611731 [2]电子科技大学大数据研究中心遥感所,成都611731

出　　处：《信息技术》2016年第4期32-36,40,共6页Information Technology

基　　金：四川省应急测绘与防灾减灾工程技术研究中心开放研究基金(K2014B003);国家自然基金(41001221;41201388;41101380;51277167);CAST创新基金;国家博士后基金(011M501400);中国科学院数字地球重点实验室资助(2011 LDE 016);国家留学基金委资助

摘　　要：压缩感知重构算法存在计算量大、运行时间过长的问题,无法满足人们对算法处理实时/准实时性要求。最近几年,GPU计算能力得到很大的提升,已成为提高算法处理速度最有效的方式之一。根据GPU的硬件特性,文中提出了基于CUDA的压缩感知重构算法的并行设计。实验结果表明:在NVIDIA K20Xm平台上运行,并行算法取得的加速比可达到100X。The processing of the serial compressed sensing reconstruction algorithm is computation intensive and lasts for long time,which cannot usually meet the real-time or near real-time requirements in some specific applications. In recent years,the computing based on GPU platform is a hearted research topic and becoming the most effective and efficient way to overcome the mentioned problems in real-time processing due to GPU computing capability has been greatly improved. Under this circumstance,this paper designed and implemented a parallel compressed sensing reconstruction algorithm utilizing CUDA programming model based on the newly GPU hardware. From the experiments on processing some remote sensing images with different sizes,it shows that the parallel algorithm can obtain 100 times of speedup on GPU platform compared with the sequential algorithm running on CPU platform.

关 键 词：压缩感知 统一计算架构 并行计算 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]