BP算法的多核并行研究及其在枣无损检测的应用  

Parallel research of BP algorithm based on multi-core and its application in nondestructive detection of jujube

作　　者：黄磊[1] 王凡[1] 吴素萍[1] 

机构地区：[1]宁夏大学数学计算机学院,宁夏银川750021

出　　处：《计算机工程与设计》2016年第9期2502-2506,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60963004);国家科技支撑计划课题基金项目(2014BAD05B02)

摘　　要：为提高BP神经网络算法的训练速度,针对BP网络训练过程中层间数据传播,权值更新耗时大的问题进行深入分析。基于单机多核CPU系统环境,运用OpenMP并行技术,设计实现多核架构下基于OpenMP的BP并行算法。将该算法应用于基于高光谱图像技术的长枣糖度无损检测中,实验结果表明,与原串行算法相比,并行算法可以保证算法的相对误差较低,取得较好加速比,该算法在不同平台随着核数的增加,加速比也在增加,验证了其可行性、有效性和可扩展性。To improve the training speed of BP neural network algorithm,in the process of BP network training,the time consumed in the inter layer data transmission and weight updating were analyzed deeply.The BP parallel algorithm based on single machine multi-core CPU system and OpenMP was designed and implemented using OpenMP parallel technology.The parallel algorithm was applied to the nondestructive detection of the sugar content of the long jujube based on hyper spectral image technology.Experimental results show that compared with the original serial algorithm,the parallel algorithm can ensure a lower relative error,and it can achieve better speedup,and the speedups of the parallel algorithm in different platforms increases along with the number of cores increases.The feasibility,efficiency and scalability of the parallel algorithm are demonstrated.

关 键 词：BP神经网络 OpenMP并行 多核 并行优化 无损检测 

分 类 号：TP302.7[自动化与计算机技术—计算机系统结构]