算法改进的自组织神经网络曲面重构

Mesh surface reconstruction based on improved kohonen neural network

摘要:

为提高神经网络法三角网格曲面重构的效率，提出自组织神经网络算法与模糊聚类算法相结合的 改进算法．应用改进算法对大规模散乱点云曲面及花瓶实例进行了网络训练及三角网格重建，在初次网络训 练收敛后，加入模糊聚类计算模块，由模糊聚类算法中隶属度计算来确定输入样本是否可用．与自组织神经 网络算法训练特性进行了比较，结果表明：改进后算法避免了以往算法的重复循环，减少了计算量，加快了网 络训练收敛速度和三角网格曲面重构的速度，仿真重构结果表明：改进后的自组织神经网络算法可实现不同 疏密程度的三角网格曲面重建，并可在保持原数据特征的前提下实现数据精简，较通常算法收敛速度快

Abstract:

To improve the efficiency of triangle mesh surface reconstruction in neural network, an improved Kohonen neural network is put forward, which combines Kohonen neural network and faintness clustering algo- rithm, and by which large scale scattered point clouds triangle mesh surface and vase surface reconstruction have been done. Characteristics comparison is carried out between the improved algorithm and general one, and the results show that the improved algorithm avoids repeat circulation in general algorithm, reduces calcu- lation time, improves the efficiency and rate of the triangle mesh surface reconstruction. Simulation reconstruc- tion result indicates that the improved arithmetic can realize sparse and dense triangle mesh surface reconstruc- tion and data condensation under preconditions with primary data characteristics. The improved arithmetic has fast network convergence speed