摘要
提出基于自适应学习速率动量梯度下降的BP算法进行物体识别,并以修正的Hu不变矩特征作为BP神经网络的输入,通过训练对网络的权值和阈值进行调整.该算法使BP神经网络在学习速率和稳定性上有了进一步的提高.仿真结果表明该方法对物体的平移、旋转、缩放都具有不变性,从而验证了该方法的有效性.
BP algorithm based on self adapting learning rate with momentum gradient reduction is presented in the paper. The modified Hu invariant moments are used as the input of BP neural network, and weights and threshold values are changed by training. The algorithm upgrades the learning rate and stability of the BP neural network. The simulation results demonstrate that the method is invariant to the translation, rotating and scale of objects. So the efficiency is proved in the paper.
出处
《微电子学与计算机》
CSCD
北大核心
2008年第4期152-155,159,共5页
Microelectronics & Computer
基金
国家自然科学基金项目(60475021)
河南省杰出青年基金项目(0412000400)
关键词
特征提取
不变矩
BP神经网络
物体识别
feature extraction
invariant moments
BP neural network
objects recognition
