摘要
K均值算法是一种常用的聚类分析方法,广泛应用于图像处理和机器学习等领域。但该算法具有较高的计算复杂度,导致了算法具有较大的局限性。为了提高算法的运行效率,本文在深入分析算法基本原理的基础上,利用CUDA架构提供的强大计算能力对该算法进行了并行化改进。实验结果表明,算法在取不同的聚类数时均取得了较高的加速比。
K - means algorithm is a commonly used method of clustering analysis, which is widely used in the fields such as image processing and machine learning. Because of this algorithm has high computational complexity, this algorithm has great limitations in the fields of high real - time demand. In order to improve the efficiency of this algorithm, we deeply analysis the basic principle of this algorithm, and using powerful computation ability provided by CUDA to realize the algorithm, for improved parallel. The experimental results show that the parallel algorithm when using different clustering number has achieved a higher speedup.
出处
《测绘与空间地理信息》
2015年第9期42-44,共3页
Geomatics & Spatial Information Technology
基金
国家863项目(2013AA7032031D)资助
关键词
K均值算法
聚类分析
图形处理器
计算统一设备架构
K - means algorithm
clustering analysis
graphic processing unit
compute unified device architecture
