BP网络改进模型的性能对比研究  

Comparative Study on Improved Models of BP Networks

作　　者：李朝峰[1] 屈颖歌[1] 夏德深[1] 邹海[2] 

机构地区：[1]南京理工大学计算机系603-2教研室,南京210094 [2]中国矿业大学信息学院,徐州221008

出　　处：《计算机工程与应用》2003年第19期120-121,132,共3页Computer Engineering and Applications

基　　金：南京理工大学博士后基金项目资助

摘　　要：文章通过实例对BP网络的几种代表性改进模型进行了性能对比研究。分析了BP网络基于标准梯度下降法和基于数值优化方法的算法改进和训练策略改进获得的代表性模型的优缺点,并结合遥感图像分类实例,对其收敛速度和分类效果进行了对比分析,其结果可为BP网络模型的选择和改进提供一些思路和借鉴。A comparative study on some typical improved models of BP networks is proposed.The advantages and disadvantages of two improved methods,one of improved algorithm based on standard Gradient descent and numerical optimization,and another of improved training strategy,are analyzed.At last taking remote sensing image classification as samples,comparative study on convergent velocity and classified effect is done.The study can be used for the reference in selecting learning algorithm and improved high performance algorithms of BP networks.

关 键 词：BP网络 梯度下降法 数值优化 训练策略 性能对比 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]