摘要
为克服传统BP算法收敛速度慢、易陷入局部极小等缺陷,本文利用遗传算法的全局寻优能力对神经网络的初始权和阈值进行优化,并将其运用到摄像机BP神经网络标定.采用遗传算法构建的神经网络,在不增加网络结构复杂度的情况下,大大提高了样本训练的精度和成功率,保证了网络的泛化能力.实验结果表明,该算法具有较高的标定精度,而且可行.
In order to overcome traditional BP algorithm's defects such as slow convergence speed,ease into the local minimum,we use genetic algorithm to optimize the initialized weights and threshold of neural network because of its global optimization ability,and apply it to the camera calibration of BP neural network.Neural network was constructed by the genetic algorithm under the circumstances without any increasing complexity in network structure,which greatly improves the precision and success rate of sample training,and ensures the network's generalization ability.Experimental results show that the algorithm has a high calibration precision and the feasibility can be readily verified.
出处
《西安建筑科技大学学报(自然科学版)》
CSCD
北大核心
2011年第4期604-608,共5页
Journal of Xi'an University of Architecture & Technology(Natural Science Edition)
基金
国家十二五科技支撑计划重点项目(2010BAE00372-2)
陕西省自然科学基金项目(2007E218)
陕西省科技攻关项目(2011K10-18)
陕西省教育厅自然科学专项资助项目(09JK559)