摘要
提出了一种基于多层前馈神经网络的二维不变性目标识别方法。利用傅里叶描述器提取具有旋转、平移及尺度不变性的目标形状特征。由于所识别的工业工具具有一个自由度, 它们的形状有一定的动态变化范围, 导致同一目标的形状特征矢量的不唯一性。文中采用含有两个隐层的多层前馈网络学习及识别这些特征矢量。在实验中, 对四类机械工具进行测试, 并将所提出方法与最近邻分类器进行比较。结果表明, 具有反向传播( B P)学习算法的多层前馈网络对噪音和形状特征变化具有鲁棒性, 且它还能判断未训练样本。
A neural netw ork based approach is proposed for invariant recognition of tw o dim ensional objects. The industrial tools to be recognized have one degree of freedom , thedynam ic range of their shapes leads the feature vector not uniquely defined even for a singleobject. The Fourier descriptors of objects boundary are taken as the features being invariantto translation, rotation and scale changes. A m ultilayer feedforw ard neuralnet w ith tw o hid den layer classifiers is utilized. The experim ental studies involving four sorts of m echanicaltools are carried out. The perform ance is com pared to a nearest neighbor rule. It is show nthat the approach is robust to not only noisy but also varying feature vector, and the neuralnetw ork can recognize som e untrained objects.
出处
《光学学报》
EI
CAS
CSCD
北大核心
1999年第8期1074-1078,共5页
Acta Optica Sinica
关键词
神经网络
目标识别
傅里叶描述器
形状特征
neural netw orks, target recognition, the Fourier descriptors, shape fea tures.