BP网络初始权值的选取方法  

A method of choosing BP network's initial weights

作　　者：乔双[1] 董智红[1] 

机构地区：[1]东北师范大学物理系,吉林长春130024

出　　处：《东北师大学报（自然科学版）》2004年第3期25-30,共6页Journal of Northeast Normal University(Natural Science Edition)

基　　金：国家高技术研究发展计划项目(2002AA632080)

摘　　要：　研究了运用BP网络对经过旋转任意角度的二值图像的训练和识别,提出了具有一定借鉴性的网络初始权值的选取方法,并在实验中对各种不同的旋转图像进行了训练和识别,所训练出的BP网络正确识别率高达90%以上.Using BP network to train and recognize 2-value images which is circumrotated in arbitrary angle.Basing on this,the paper analyses and proposes a method of establishing BP network's initial weights.The problem is often encountered in target recognition based on BP network.In experiments,all kinds of circumrotated 2-value images are trained and recognized. The results indicate that using the trained BP network can make the correct rate exceed 90%.

关 键 词：BP算法 初始权值 神经网络 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]