基于Kohonen神经网络的B样条曲面重构

探讨了三维散乱数据点的自由曲面自组织重构方法。建立了基于自组织特征映射神经网络的矩形网格重构模型及其训练算法。所建模型利用神经元对曲面散乱数据点的学习和训练来模拟曲面上点与点之间的内在关系,节点连接权向量集作为对散乱数据点集的工程近似化并重构曲面样本点的内在拓扑关系。通过该方法不仅能够对无规则散乱数据点进行逼近,并且通过该方法得到的曲面也可以作为后继曲面重构的初始曲面。仿真实验表明,所建神经网络模型可实现三维密集无规则数据点的曲面自组织重构集自压缩于一体。

B-spline surface reconstruction based on Kohonen neural network

The approach to the freeform surface self-organizing reconstruction for the dense 3D scattered data was discussed. Based on the self-organizing feature map neural network, a rectangle mesh reconstruction approach and the training algorithm were developed. The inherent topologic relations between the scattered points on the surface were learned by the self organizing feature map neural network. The weight vectors of the neurons on the output layer of the neural network were used to approximate the scattered data points. By this approach, not only to approximate the scattered data points and the surface which is reconstructed by this method can be as base surface for further process, but also the experiment indicates that by this approach, the reconstruction of the surface and the reduce of the dense scattered data points are combined into the same process. The computer simulation result shows that this method is effective.