基于进化策略的BP神经网络匹配算法研究  

BP neural network for matching algorithm based on evolutionary strategy

作　　者：李俊山[1] 韩先锋[1] 李建军[1] 李堃[2] 

机构地区：[1]第二炮兵工程学院计算机与指挥自动化系,陕西西安710025 [2]西安电子科技大学智能信息处理研究所,陕西西安710071

出　　处：《红外与激光工程》2006年第z4期1-6,共6页Infrared and Laser Engineering

基　　金：预研资助项目(103020305)

摘　　要：研究了利用 BP 神经网络理论进行景象匹配的方法,将灰度不变矩和边缘不变矩同时作为匹配特征应用于 BP 神经网络景象匹配中,并对经典的进化策略进行改进,提出并实现了一种高效的基于进化策略的BP 神经网络景象匹配算法。实验结果证明了算法寻优的全局性和精确性。The theory and the algorithm of BP neural network are applied in the research of scene matching. Both gray invariant moments and edge invariant moments are introduce into the matching algorithm based on BP neural network.The traditional evolution strategy is improved,and a new efficiency neural network are proposed for scene matching based on evolutionary strategy.Resolve the problem of slow convergence and relapsing into local extremism of the BP neural network.The paper and the experiment results analyses its capability and proves it.

关 键 词：进化策略 神经网络 景象匹配 特征设计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]