基于GPU的遥感图像前期处理算法研究与应用

Research and application of GPU-based preprocessing algorithms for remote sensing image

针对传统的遥感图像前期处理算法在面对海量地面数据时计算时间很难满足需求的问题,基于RPC模型的遥感成像几何校正算法的并行加速和基于SIFT特征提取的图像匹配技术的并行加速研究。针对几何校正的主要步骤及其速度瓶颈问题,提出了可采用的并行加速方法,同时结合SIFT的特点提出了并行优化加速的方案。采用基于数据划分的并行方法对遥感图像的几何校正和SIFT特征提取算法进行加速。最后利用CUDA环境,在CPU+GPU异构系统下,设计试验对两个算法优化并行提速,试验结果表明,提出的加速方案和优化算法能大幅提高遥感图像的前期处理效率。

Since the computing time of the traditional preprocessing algorithm that is faced with massive ground data is difficult to satisfy the requirement of practical application,the parallel acceleration algorithms of remote sensing image geometric correction based on RPC model and image matching technique based on SIFT feature extraction are studied. For the key steps and the speed bottleneck of the geometric correction,the adoptable parallel acceleration method is presented,and the parallel optimization acceleration scheme is proposed in combination with the characteristics of SIFT. The parallel algorithm based on data partitioning is adopted to accelerate the geometric correction algorithm and SIFT feature extraction algorithm for remote sensing image. In the CUDA environment,the two parallel optimization algorithms were speeded up by design experiment in CPU＋GPU heterogeneous system. The test results show that the proposed acceleration scheme and optimization algorithm can greatly improve the preprocessing efficiency of the remote sensing image.