基于GPU的SAR方位向信号分解的高效实现方法

A highly efficient GPU-based method for decomposing SAR azimuth signal

摘要:

提出了一种基于图形处理器(GPU)的SAR方位向信号分解的高效实现方法。SAR方位向信号可以通过四参数Chirplet分解方法来分解。此方法的关键难题是计算量过大,计算量主要由2部分组成：构建Chirp原子库，以及SAR方位向信号在过完备库上分解的计算量。与传统的CPU相比，GPU更加适用于密集型和大量数据并行化的计算。提出将算法的核心部分移植到GPU上进行并行计算，充分挖掘其运算潜能。结果表明：该方法与传统的基于CPU的算法相比有两位数以上的效率提升。

Abstract:

A highly efficient Graphics Processing Unit(GPU) based method for decomposing SAR azimuth signal is presented. SAR azimuth signal is decomposed by the four-parameter Chirplet decomposition method. This method needs excessive computation mainly to construct Chirp atomic dictionary and decompose SAR azimuth signal in the over-complete dictionary. Compared with CPU, GPU is well suited for data parallel computations with high arithmetic intensity. A new method is introduced that takes full advantage of GPU’s computation capability by transferring the key part of the algorithm to GPU.The experimental results shows that this method is more than ten times as fast as traditional CPU based algorithms.