Journal of Computer Applications ›› 2020, Vol. 40 ›› Issue (6): 1607-1612.DOI: 10.11772/j.issn.1001-9081.2019101879

• Artificial intelligence • Previous Articles     Next Articles

3D point cloud classification and segmentation network based on Spider convolution

WANG Benjie1, NONG Liping2,3, ZHANG Wenhui1, LIN Jiming1, WANG Junyi1   

  1. 1. School of Information and Communication, Guilin University of Electronic Technology, Guilin Guangxi 541004, China
    2. School of Telecommunication Engineering, Xidian University, Xi’an Shaanxi 710071, China
    3. College of Physical Science and Technology, Guangxi Normal University, Guilin Guangxi 541004, China
  • Received:2019-11-04 Revised:2019-12-22 Online:2020-06-10 Published:2020-06-18
  • Contact: ZHANG Wenhui, born in 1970, M. S., associate professor. Her research interests include computer graphics, computer animation.
  • About author:WANG Benjie, born in 1993, M. S. candidate. His research interests include point cloud recognition, deep learning.NONG Liping, born in 1985, Ph. D. candidate, lecturer. Her research interests include graph signal processing, geometric deep learning.ZHANG Wenhui, born in 1970, M. S., associate professor. Her research interests include computer graphics, computer animation.LIN Jiming, born in 1970, Ph. D., professor. His research interests include wireless communication, mobile communication.WANG Junyi, born in 1977, Ph. D., research fellow. His research interests include graph signal processing, deep learning, wireless network resource management.
  • Supported by:

    National Natural Science Foundation of China (61966007), the Development Foundation of Key Laboratory of Cognitive Radio and Information Processing of Ministry of Education (CRKL180201), the Project of Guangxi Cooperative Innovation Center of Cloud Computing and Big Data (1716), the Leader Foundation of Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing (GXKL06180107, CRKL180106).

基于Spider卷积的三维点云分类与分割网络

王本杰1, 农丽萍2,3, 张文辉1, 林基明1, 王俊义1   

  1. 1.桂林电子科技大学 信息与通信学院,广西 桂林 541004
    2.西安电子科技大学 通信工程学院,西安 710071
    3.广西师范大学 物理科学与技术学院,广西 桂林 541004
  • 通讯作者: 张文辉(1970—)
  • 作者简介:王本杰(1993—),男,河南南阳人,硕士研究生,主要研究方向:点云识别、深度学习。农丽萍(1985—),女,广西天等人,讲师,博士研究生,主要研究方向:图信号处理、几何深度学习。张文辉(1970—),女,湖南益阳人,副教授,硕士,主要研究方向:计算机图形学、计算机动画。林基明(1970—),男,四川三台人,教授,博士,主要研究方向:无线通信、移动通信。王俊义(1977—),男,河北邢台人,研究员,博士,主要研究方向:图信号处理、深度学习、无线网络资源管控。
  • 基金资助:

    国家自然科学基金资助项目(61966007);认知无线电与信息处理教育部重点实验室开发基金资助项目(CRKL180201);广西云计算与大数据协同创新中心项目(1716);广西无线宽带通信与信号处理重点实验室主任基金资助项目(GXKL06180107,CRKL180106)。

Abstract:

The traditional Convolutional Neural Network (CNN) cannot directly process point cloud data, and the point cloud data must be converted into a multi-view or voxelized grid, which leads to a complicated process and low point cloud recognition accuracy. Aiming at the problem, a new point cloud classification and segmentation network called Linked-Spider CNN was proposed. Firstly, the deep features of point cloud were extracted by adding more Spider convolution layers based on Spider CNN. Secondly, by introducing the idea of residual network, short links were added to every Spider convolution layer to form residual blocks. Thirdly, the output features of each layer of residual blocks were spliced and fused to form the point cloud features. Finally, the point cloud features were classified by three-layer fully connected layers or segmented by multiple convolution layers. The proposed network was compared with other networks such as PointNet, PointNet++ and Spider CNN on ModelNet40 and ShapeNet Parts datasets. The experimental results show that the proposed network can improve the classification accuracy and segmentation effect of point clouds, and it has faster convergence speed and stronger robustness.

Key words: Convolutional Neural Network (CNN), Spider convolution, point cloud classification and segmentation, residual block, robustness

摘要:

针对传统的卷积神经网络(CNN)不能直接处理点云数据,需先将点云数据转换为多视图或者体素化网格,导致过程复杂且点云识别精度低的问题,提出一种新型的点云分类与分割网络Linked-Spider CNN。首先,在Spider CNN基础上通过增加Spider卷积层数以获取点云深层次特征;其次,引入残差网络的思想在每层Spider卷积增加短连接构成残差块;然后,将每层残差块的输出特征进行拼接融合形成点云特征;最后,使用三层全连接层对点云特征进行分类或者利用多层卷积层对点云特征进行分割。在ModelNet40和ShapeNet Parts数据集上将所提网络与PointNet、PointNet++和Spider CNN等网络进行对比实验,实验结果表明,所提网络可以提高点云的分类精度和分割效果,说明该网络具有更快的收敛速度和更强的鲁棒性。

关键词: 卷积神经网络, Spider卷积, 点云分类与分割, 残差块, 鲁棒性

CLC Number: