基于GPU的加速船舶轨迹相似性度量与聚类

GPU-based ship trajectory similarity measurement and clustering

摘要: 针对使用中央处理器(Central Processing Unit, CPU)硬件实现密度聚类、相似性度量等算法提取船舶习惯航迹的过程中存在复杂度高、计算时间长等方面的不足，提出使用图形处理器(Graphics Processing Unit, GPU)高性能计算及GPU优化算法以提升船舶轨迹相似性度量与聚类的速度性能，大幅缩短船舶轨迹特征提取过程中的时间开销。利用长江南槽交汇水域船舶自动识别系统(Automatic Identification System, AIS)动态船舶轨迹信息进行方法验证，通过对比传统基于CPU的方法验证了所提出的基于GPU的船舶轨迹相似性度量及聚类算法存在较优的速度性能，为快速提取研究水域中的船舶特征提供新的理论依据。

Abstract: GPU（Graphics Processing Unit）, instead of CPU（Central Processing Unit）, is used to accomplish ship trajectory similarity measurement and clustering. For CPU, to carry out similarity measurement and clustering needs complex programming and long computing time. On contrary, GPU, featuring high computing capacity, with special optimization algorithm can do the job much faster. The design is verified through processing the AIS（Automatic Identification System） data from the converging area of the Yangtze River South Channel. The improvement in processing speed is proved.