统一设备计算架构下的栅格河网提取并行算法  

Parallel algorithm of raster river network extraction based on CUDA

作　　者：王玉着[1] 刘修国[1] 张唯[1] 

机构地区：[1]中国地质大学(武汉)信息工程学院,武汉430074

出　　处：《计算机应用》2015年第4期960-963,967,共5页journal of Computer Applications

基　　金：国家科技支撑计划项目(2011BAH06B04);国家自然科学基金资助项目(41001225)

摘　　要：针对大规模高分辨率数字地形数据提取栅格河网效率低下的问题,提出了基于统一设备计算架构(CUDA)利用淹没模型提取栅格河网的并行算法。使用图形处理器(GPU)将汇流累积量计算分解为独立的多任务并行处理,通过数据异步传输减少数据交换时间,进而加速河网提取的运算。实验结果表明,该算法运行效率明显优于串行河网提取算法,在NVIDIA Geforce GTX660上对数据量为600 MB(网格大小为9784×8507)数字高程模型(DEM)数据提取河网加速比达到62。Concerning the low efficiency of calculating flow accumulation on high resolution digital terrain data, a parallel algorithm was put forward based on the Compute Unified Device Architecture（ CUDA） and flooding model. Based on the technology of Graphic Processing Unit（ GPU）, two strategies were designed to improve the speed of the extraction. Firstly, the calculation of flow accumulation was divided into a plurality of independent tasks for parallel processing. Secondly, the time of data exchange was reduced through the asynchronous data transmission. The experimental results show that the efficiency of the parallel algorithm is superior of the serial algorithm. The acceleration of river network extraction reached 62 times in NVIDIA Geforce GTX660 for 600 MB DEM data with 9 784 × 8 507 grid size.

关 键 词：统一设备计算架构 数字高程模型 栅格河网 D8算法 并行计算 

分 类 号：TP333.9[自动化与计算机技术—计算机系统结构]