Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 60, Issue 14, 1415003(2023)

Point Cloud Classification Method Based on Graph Convolution and Multilayer Feature Fusion

Sheng Tian* and Anyang Long
Author Affiliations
  • School of Civil and Transportation, South China University of Technology, Guangzhou 510641, Guangdong, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (12)
    Equations (15)
    References (23)
    Cited By (1)
    Tools
    Paper Information
    Topics
    Figures & Tables(12)
    Point cloud classification network structure

    Fig. 1. Point cloud classification network structure

    Download full size
    Graph structure feature extraction module

    Fig. 2. Graph structure feature extraction module

    Download full size
    Construction process of graph structure

    Fig. 3. Construction process of graph structure

    Download full size
    Structure of MFF module

    Fig. 4. Structure of MFF module

    Download full size
    Visualization of partial point cloud model classification results

    Fig. 5. Visualization of partial point cloud model classification results

    Download full size
    Influence of K value on classification accuracy

    Fig. 6. Influence of K value on classification accuracy

    Download full size
    Influence of sampling points on classification accuracy

    Fig. 7. Influence of sampling points on classification accuracy

    Download full size
    Classification performance under different noise types. (a) Gaussian noise; (b) random noise

    Fig. 8. Classification performance under different noise types. (a) Gaussian noise; (b) random noise

    Download full size
    Influence of point loss and noise on classification accuracy

    Fig. 9. Influence of point loss and noise on classification accuracy

    Download full size

    • Table 1. Comparison of classification results on ModelNet40 dataset

      View table

      Table 1. Comparison of classification results on ModelNet40 dataset

      MethodInputOA /%mAcc /%
      MVCNN［7］Multiview90.1
      VoxNet［9］Voxel85.9
      PointNet［11］Points89.286.2
      PointNet++［12］Points+Normal91.9
      SO-Net［13］Points+Normal90.987.3
      DGCNN［18］Points91.288.8
      LDGCNN［19］Points92.990.3
      Proposed methodPoints93.290.4

    • Table 2. Influence of neighborhood selection method on classification accuracy

      View table

      Table 2. Influence of neighborhood selection method on classification accuracy

      MethodOAmAcc
      ball query（r=0.1）92.589.9
      ball query（r=0.2）92.990.6
      ball query（r=0.3）92.690.1
      KNN（K=20）93.290.4

    • Table 3. Comparison of ablation experiment results

      View table

      Table 3. Comparison of ablation experiment results

      ModelOAmAcc
      A：original edge function + maxpooling91.489.0
      B：improved edge function + maxpooling91.789.2
      C：improved edge function + attentive pooling92.389.7
      D：improved edge function + attentive pooling + MFF93.290.4
    Tools

    Get Citation

    Copy Citation Text

    Sheng Tian, Anyang Long. Point Cloud Classification Method Based on Graph Convolution and Multilayer Feature Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1415003

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Machine Vision

    Received: Jun. 28, 2022

    Accepted: Aug. 29, 2022

    Published Online: Jul. 25, 2023

    The Author Email: Tian Sheng (shitianl@scut.edu.cn)

    DOI:10.3788/LOP221933

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology