车辆图像检索方法的Julia云实现  

Realization of Vehicle Image Retrieval Method in Julia Cloud

作　　者：黄博[1,2] 王莉[1] 张常有[2] 王婷[2] 

机构地区：[1]太原理工大学计算机科学技术学院,太原030024 [2]中国科学院软件研究所并行软件与计算科学实验室,北京100190

出　　处：《小型微型计算机系统》2016年第1期57-60,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61379048)资助;中国科学院联盟院地合作项目资助

摘　　要：为降低开发并行程序的难度,使用高性能计算机提高车辆图像的检索效率,为普通应用程序开发人员构建了Julia云编程环境.该编程环境以Julia为编程语言,开发车辆图像检索专用API接口;该API接口根据计算机底层硬件体系结构的特点提供相应的实现方式,保证运行效率;为多种实现方式设计统一调用的自适应策略,保证用户透明调用.本文使用Julia语言实现了车辆图像检索程序,程序代码量少,容易修改调整.实验结果表明,本文Julia实现的并行方案获得最高40倍加速,且Julia实现比C语言并行实现仅1.93%性能损失.To reduce the difficulties to edit parallel programs, use high performance computers is a popular way. The paper builds a Jul- ia cloud programming environment. The cloud programming environment uses Julia as programming language. Special API interface a- bout vehicle retrieval has been developed. The API interface has different realization based on characteristic of computer underlying ar- chitecture, can be easily called by users. The realizations can also guarantee operating efficiency. Call underlying method depends on auto-adapting strategy. The process is transparent to users. Different realizations unify the same interface. This paper develops a set of vehicle retrieval library. Based on this library, a vehicle retrieval program is realized in Julia, display the sample code and compare the efficiency of the deferent realizations. The experimental results show that Julia program is concise and the parallel-program based on CUDA-lib gets a speed up of 40x than the serial program work on the same image set as a whole and has only 1.93% performance cost comparison with parallel implementation of C language.

关 键 词：车辆检索 Julia云 编程环境 自适应策略 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]