摘要
针对传统BP神经网络容易陷入局部极小、收敛速度慢和确定隐含层的神经元个数比较困难等缺点,从结构和算法两方面对BP神经网络进行改进。改进后的网络具有较快的收敛速度和较短的运行时间,加强了BP神经网络的学习能力和自适应能力,并将其应用于物体的分类识别,取得了良好的效果。仿真结果表明了此改进方法的可行性和有效性。
Since the Back Propagation （BP） neural network into local minimum, low speed of convergence and difficult in has such shortcomings as being prone to fall determining the numbers of neural cell for the hidden layers, the BP neural network was improved from the structure and the algorithm in this paper. The improved BP neural network has faster convergence speed and shorter running time. And the learning ability and adaptability of the BP neural object recognition and obtained a network was strengthened accordingl favorable result. The feasibility and y. The improved method was applied to validity of it was proved by a series of simulation experiments.
出处
《电光与控制》
北大核心
2012年第4期68-71,共4页
Electronics Optics & Control
基金
河南省基础与前沿技术研究计划项目(102300410113)
河南省重点科技攻关项目(092102210293)
关键词
物体识别
BP算法
神经网络
改进
object recognition
BP algorithm
neural network
improvement
