【摘要】 现有的遥感数据时空融合算法复杂,计算时间长,获取海量时序的高时空分辨率遥感影像非常困难。因此,通过分析GPU并行运算模式与遥感数据时空融合算法的实现步骤,合理地设计了一种基于CPU-GPU异构混合编程的遥感数据时空融合并行处理算法流程,将融合算法中的数据密集型计算部分由CPU移植到GPU中执行。遥感数据时空融合算法种类繁多,不同算法的可并行程度与算法复杂度有着很大的差异,选取3种不同类型的遥感数据时空融合算法STDFA、STARFM、CDSTARFM进行GPU并行设计,并使用CUDA架构实现。实验结果表明,基于CPU-GPU异构混合编程技术可大幅度缩减遥感数据时空融合时间,提升计算效率,最高加速比可达到195.6,从而可为海量时空遥感数据的深度应用提供技术支撑。更多还原

【Abstract】 The existing temporal and spatial fusion algorithms for remote sensing data are complex and time-consuming.Therefore,this paper develops a parallel processing algorithm for temporal and spatial fusion algorithm of remote sensing data based on CPU-GPU heterogeneous hybrid programming by analyzing the implementation steps of GPU parallel computing mode and temporal and spatial fusion algorithm of remote sensing data,which the data-intensive computing part of the fusion algorithm is ported from the CPU to the GPU for execution.There are many kinds of temporal and spatial fusion algorithms for remote sensing data,which the degree of parallelism and complexity of different algorithms have huge differences.This paper selects three different types of remote sensing data temporal and spatial fusion algorithms,including STDFA,STARFM and CDSTARFM,for GPU parallel design,and we use CUDA architecture to achieve them.The experimental results show that heterogeneous hybrid programming based on CPU-GPU can greatly reduce the time of temporal and spatial fusion algorithms for remote sensing data and improve the computational efficiency,which the highest acceleration ratio can reach 195.6.Therefore,it can strongly support the practical application of massive temporal and spatial remote sensing data.更多还原