摘要
针对卷积神经网络(CNN)算法收敛速度慢的问题,本文采用粒子群算法(PSO),将CNN的训练参数和误差函数分别作为PSO的粒子和适应度函数,对CNN的误差反传阶段进行改进.通过对AR人脸数据库性别识别的实验仿真,验证了改进的算法收敛速度快并且识别精度高.
In view of slow convergence speed of the Convolutional Neural Network （CNN） problems, the training paramete~ and the error function of CNN as particles of PSO and the fitness function were used to im- prove the back propagation stage of CNN based on the Particle Swarm Optimization （PSO） algorithm. By the ex- perimental simulation of face gender recognition based on AR face database, the improved algorithm has fast con- vergence speed and high precision.
作者
裴子龙
邢进生
PEI Zi-long XING Jin-sheng(College of Mathematics and Computer Science, Shanxi Normal University, Linfen 041000, Shanxi, China)
出处
《山西师范大学学报（自然科学版）》
2017年第2期22-26,共5页
Journal of Shanxi Normal University(Natural Science Edition)
基金
山西省自然科学基金(2015011040)
关键词
卷积神经网络
粒子群算法
人脸性别识别
Convolutional Neural Network
Particle Swarm Optimization
facial gender recognition
