基于空间感知和特征增强的三维点云分类与分割研究
Research on classification and segmentation of 3d point cloud based on spatial awareness and feature enhancement
投稿时间:2022-11-09  修订日期:2023-02-14
DOI:
中文关键词:  三维点云  空间感知  特征增强  分类与分割
英文关键词:3d point cloud  spatial awareness  feature enhancement  classification and segmentation
基金项目:天津市研究生科研创新项目
作者单位邮编
方银 天津理工大学电气工程与自动化学院 300384
张惊雷* 天津理工大学工程训练中心 300384
文彪 天津理工大学电气工程与自动化学院 
摘要点击次数: 185
全文下载次数: 0
中文摘要:
      针对直接处理点云数据的深度神经网络PointNet 无法充分学习点云形状信息的问题,提出一种融合空间感知模块和特征增强模块的三维点云分类与分割方法SAFE-PointNet 。首先,设计了空间感知模块,使特征提取网络在特征升维时融合了包含空间结构的权重信息,增强了特征在空间上的表现力。其次,设计了特征增强模块,通过把增强后的几何信息和附加信息拆分并分别进行编码,达到充分利用点云附加信息的目的。实验结果表明,在ModelNet40和S3DIS数据集上,SAFE-PointNet 与其他10种经典网络相比具有更高的分类和分割精度。
英文摘要:
      Aiming at the problem that the deep neural network PointNet which directly processes point cloud data cannot fully learn the shape information of point cloud, a 3D point cloud classification and segmentation method SAFE-PointNet is proposed, which combines spatial awareness module and feature enhancement module. First, the spatial awareness module is designed to make the feature extraction network integrate the weight information of spatial structure when the feature dimension is raised, enhancing the expression of the feature in space. Secondly, the feature enhancement module is designed to make full use of the additional information of the point cloud by splitting and encoding the enhanced geometric information and additional information respectively. The experimental results show that on ModelNet40 and S3DIS datasets, SAFE-PointNet achieves higher classification and segmentation accuracy compared with other 10 classical networks.
    下载PDF阅读器
关闭

版权所有:《光电子·激光》编辑部  津ICP备12008651号-1
主管单位:天津市教育委员会 主办单位:天津理工大学 地址:中国天津市西青区宾水西道391号
技术支持:北京勤云科技发展有限公司