一种3D点云分割改进型边缘特征提取网络

An Improved Edge Feature Detection Network Based on 3D Point Cloud Segmentation

针对点云分割中边缘特征提取不足导致局部特征信息不完整、整体分割精度下降的问题,提出一种3D点云分割改进型边缘特征提取网络。为提升边缘特征表达能力,对深层边缘特征提取层做出改进,引入基于残差结构改进的多层感知机结构,形成边缘特征提取单元,该单元将原始点云特征与多层感知机提取边缘特征相融合,获取更加丰富完整的边缘信息,提高3D点云分割网络模型精度。在Shapenet数据集上的实验结果表明,提出的3D点云分割改进型边缘特征提取网络优于现存同类方法,相较于LDGCNN链接动态图网络,点云分割准确率提升了1.17%。

An improved edge extraction network based on 3D point cloud segmentation was proposed to solve the problem of incomplete local feature information and decreased overall segmentation accuracy caused by insufficient edge feature extraction in point cloud segmentation.To enhance edge feature expression ability and improve the deep edge feature extract layer,we introduced a multilayer perceptron structure based on the residual structure improvement,and formed edge feature extraction units.The units can integrate the original point cloud features with the edge features extracted by the multi-layer perceptron to obtain edge information with more richness and integrity and improve the precision of 3D point cloud segmentation network.The experimental results on Shapenet data set show that the proposed improved edge extraction network for 3D point cloud segmentation is superior to the existing similar methods.The accuracy of point cloud segmentation is improved by 1.17%,compared with LDGCNN.