【摘要】 针对高光谱图像数据量大,光谱匹配计算耗时长的缺点,提出一种CUDA架构下的高光谱图像光谱匹配快速实现方法。首先,在分析高光谱图像特点的基础上,介绍高光谱图像光谱匹配的一般方法;其次,介绍CUDA架构及其特点,依据CUDA架构的特点,将通常的高光谱图像存储格式进行重新组织,并将参考光谱存储在共享内存中,以达到快速访问的目的;最后,将光谱匹配算法中的欧氏距离匹配法、光谱夹角匹配法、光谱相关系数匹配法在CUDA架构下进行实现。将该方法应用于实际的高光谱图像光谱匹配中,可极大地提高传统CPU计算的运算效率。更多还原

【Abstract】 To overcome the drawback of time consuming of spectral matching caused by the large volume of data in hyperspectral image,a new high speed approach based on CUDA is presented.Firstly,the traditional spectral matching approaches are introduced based on the analysis of the characteristic of hyperspectral image.Secondly,the framework and the characteristics of CUDA are briefly discussed.According to the characteristics of CUDA,the general storage format of hyperspectral image is reorganized,and the reference spectrum is stored in shared memory to accelerate the data accessing.Finally,the spectral matching approaches of Euclid distance,spectral angle map,and spectral correlation are implemented in CUDA.This proposed approach can highly improve the efficiency of CPU calculation.更多还原