利用模糊神经网络实现逆向工程中的区域分割

Fuzzy Self-Organizing Feature Map Networks for Segmentation in Reverse Engineering

论文提出了一种改进的模糊自组织特征映射网络(fuzzySOFM),它不仅显著加快了聚类的速度,而且算法简单。该网络采用由数据点的坐标、估算出的法矢量和曲率构成的八维特征向量作为输入,快速地实现了逆向工程中点云数据的区域分割。与现有方法相比,该方法具有以下优点:第一,具有更高的聚类速度,并可以直接处理含噪声数据;第二,聚类的结果与数据输入的顺序无关;第三,能利用数据的隶属度快速提取出特征线数据,从而将基于面的分割和基于线的分割结合起来。实验结果证明了这种方法的有效性。

An improved fuzzy Self-Organizing Feature Map Network(fuzzy SOFM),which is a simple algorithm and improves convergence speed notablely,is given in this paper.Eight dimensional feature vectors(3-dimensional coordinate,3-dimensional normal vector and2-dimensional curvature)are taken as input for this fuzzy SOFM to realize segmentation in reverse engineering.Compared with other methods,this method can dispose of noisy data and converges more quickly.Its segmentation result is independent of the sequence of data input.Furthermore,points on characteristic curve can also be extracted quickly according to the degree of membership and thus region-based approach is combined with border-based approach.The validity of this method is proved by the real scanned point-cloud.