基于GPU加速的遥感影像金字塔创建算法及其在土地遥感影像管理中的应用

摘要：为提高单个计算节点创建影像金字塔的速度，本研究首先将GPU并行技术用于加速影像重采样算法.影像重采样算法是影像金字塔创建算法的核心步骤，由于金字塔创建过程中数据量会不断发生变化，而数据量的大小直接影响GPU重采样算法效率.提出了一种基于阈值的金字塔遥感影像创建算法，算法将GPU并行与CPU串行遥感影像重采样算法结合，在创建影像金字塔时，依据阈值动态选择不同的重采样算法，并将本算法应用到土地遥感影像金字塔管理中.实验采用大小为10371×7945的24位遥感影像进行测试，结果表明：①基于GPU的并行重采样算法的速度最快，是基于CPU串行重采样算法的10倍；②采用本文算法创建金字塔速度是ArcGIS9.3创建金字塔速度的3倍以上.

Parallel image resample algorithm based on GPU for land remote sensing data management

In order to accelerate the image pyramid creation algorithm,first we migrate the image resample algorithm to CUDA（Compute Unified Device Architecture）,which is the core step of image pyramid creation algorithm.Then the image pyramid creation based on threshold is presented to combine GPU and CPU image resample algorithm,considering that the data size could affect the mage resample algorithm＇s efficiency.Finally the algorithm is applied to the management of land remote sensing data.By using 24-bit 10 371×7 945 image for testing,the experiment result shows that： ①the GPU resample algorithm＇s speedup is 10 times faster than CPU resample algorithm＇s,②on using the same type of resample algorithm,our parallel creating pyramid method＇s speedup is 3 times faster than ArcGIS 93＇s.