【摘要】 混合像元分解是遥感技术向定量化、精细化发展的重要技术,是关系到地物精细分类的重要操作环节,而线性光谱模型确实是目前解决混合像元问题的有效策略。针对高光谱遥感影像数据量大,混合像元分解计算耗时长的问题,提出了一种基于CUDA的高光谱遥感端元投影向量法实现方法。在分析高光谱图像端元投影向量法串行算法的基础上,建立了在CUDA架构下以像元点为基准产生相应的进程数,每个进程负责一单位像元点的计算方式。实验结果表明,将该方法应用于实际的高光谱遥感影像的混合像元中,可极大地提高传统中央处理器(CPU)的运算效率。更多还原

【Abstract】 Mixed pixel decomposition is an important technology for remote sensing to be more quantitative and more detailed;it′s also an important operation links for meticulous classification.At present,linear spectral mixture model has been considered as a effective way to extract useful information from the mixed pixel.To overcome the drawback of time consuming of decomposition of mixed pixels caused by the large volume of data in hyperspectral image,a new high speed approach based on CUDA is presented.This article discusses elementary principle of spectral vector projection by sequential algorithm method,according to the number of pixel;create process in the CUDA framework,each process is in charge of a unit point calculation.Experimental result shows the effectiveness can be obviously improved with this method.更多还原