摘要
曲面重建是CAGD中的重点研究课题,而神经网络具有很好的非线性逼近能力,文中将二者结合,给出了一种利用神经网络将三维数据点拟合为NURBS曲面的方法。提出的前馈型神经网络包含四个隐层,其中一层的激活函数为B样条基函数。由数学推导可知,该网络可以表达NURBS曲面,通过对控制顶点及其权重的学习,可以用该网络来重建NURBS曲面。权值的调整通过误差反传与梯度下降法实现。实验结果表明,文中提出的方法是可行的。
Surface reconstruction is the focus in CAGD research. Neural network has a good ability of non-linear approximation. In this paper,give a method for the NURBS surface fitting using neural network. The proposed feed- forward neural network contains four hidden layers. The activation function of one layer is the B- spine basic function. From mathematical derivation know that NURKS surface can be expressed by this network. Through the study of the control points and weights, the network can be used to NURBS surface reconstruction. Experiments show that the method is feasible.
出处
《计算机技术与发展》
2009年第9期65-68,共4页
Computer Technology and Development
基金
国家自然科学基金(60873093)
