摘要
K-Means算法是对遥感图像在没有先验知识情况下进行无监督分类的重要算法之一,在遥感影像的分析中得到了广泛的应用。针对K-Means算法复杂,处理过程中计算时间长的缺点,人们试图寻求快速的并行处理方式。在这种并行化的探索过程中,由于K-Means算法独特的流程结构,使其并行化处理方式难以顺利进行。本文在分析K-Means算法特点的基础上,对其并行化方式进行了深入的研究。针对K-Means算法并行化在处理速度和分类精度方面存在的问题,提出了一种基于分块逼近的算法并行模型,可兼顾并行效率和分类精度之间的综合要求,实现某种精度可控的并行处理。最后,根据实验结果讨论并提出了迭代算法并行化的有效途径。
Remote sensing image-oriented K-Means algorithm is one of important unsupervised clustering algorithms using no former knowledge and has been widely used in remote sensing image analysis. It is useful but hard to parallelize K-Means algorithm because of its algorithm complexity and unique process procedure. This paper makes a study on parallel processing of KMeans algorithm and a new strategy is proposed to improve processing speed and accuracy. This strategy is based on the blockapproaching algorithm-parallel model, and gets a good tradeoff between efficiency and accuracy. Based on the experiments, proposal for further research on iterative algorithms is provided.
出处
《遥感信息》
CSCD
2008年第1期27-30,115,共5页
Remote Sensing Information
基金
国家科技部"十五"科技攻关专项"高性能对地观测微小卫星技术与应用研究"中"小卫星对地观测数据地面预处理系统"的专题(2002BA104A03)
关键词
遥感
无监督分类
K—Means算法
并行算法
数据并行
remote sensing
unsupervised classifieation
K-Means algorithm
algorithm parallel
data-parallel
