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. 57, Issue 8, 081019(2020)

Object Classification Method for Multi-Source Fusion Point Clouds Based on Point-Net

Xiaosong Shi*, Yinglei Cheng, Doudou Xue, and Xianxiang Qin
Author Affiliations
  • Information and Navigation College, Air Force Engineering University, Xi'an, Shaanxi 710077, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (10)
    Equations (0)
    References (17)
    Cited By (5)
    Tools
    Paper Information
    Topics
    Figures & Tables(10)
    Flow chart of proposed method

    Fig. 1. Flow chart of proposed method

    Download full size
    Process of registration and fusion

    Fig. 2. Process of registration and fusion

    Download full size
    Architecture of Point-Net

    Fig. 3. Architecture of Point-Net

    Download full size
    Architecture of multi-scale Point-Net

    Fig. 4. Architecture of multi-scale Point-Net

    Download full size
    Simulation data. (a) Point cloud data; (b) remote-sensing image data; (c) overlapping image

    Fig. 5. Simulation data. (a) Point cloud data; (b) remote-sensing image data; (c) overlapping image

    Download full size
    Binary images of point clouds and remote-sensing image, results of registration, and fusion point clouds. (a) Binary image of point clouds; (b) binary image of remote-sensing image; (c) registration result; (d) fusion point cloud data

    Fig. 6. Binary images of point clouds and remote-sensing image, results of registration, and fusion point clouds. (a) Binary image of point clouds; (b) binary image of remote-sensing image; (c) registration result; (d) fusion point cloud data

    Download full size
    Training error curve

    Fig. 7. Training error curve

    Download full size
    Classification results. (a)(b) Without RGB point clouds; (c)(d) with RGB point clouds

    Fig. 8. Classification results. (a)(b) Without RGB point clouds; (c)(d) with RGB point clouds

    Download full size

    • Table 1. Classification accuracy of different methods

      View table

      Table 1. Classification accuracy of different methods

      AlgorithmCategoryF1 /%Overall accuracy /%
      Ground79.2
      SVMVegetation81.278.7
      Building78.5
      Artificiality45.1
      Ground89.4
      Point-NetVegetation87.484.7
      Building78.2
      Artificiality24.2
      Multi-scalePoint-Netwithout RGBGround92.7
      Vegetation91.692.5
      Building93.3
      Artificiality57.1
      Multi-scalePoint-Netwith RGBGround98.5
      Vegetation97.696.8
      Building96.1
      Artificiality76.9

    • Table 2. Classification time of different methodss

      View table

      Table 2. Classification time of different methodss

      MethodPoint-NetMulti-scale Point-Net without RGBMulti-scale Point-Net with RGB
      Training time437546495568
      Test time204.9209.8204.9
    Tools

    Get Citation

    Copy Citation Text

    Xiaosong Shi, Yinglei Cheng, Doudou Xue, Xianxiang Qin. Object Classification Method for Multi-Source Fusion Point Clouds Based on Point-Net[J]. Laser & Optoelectronics Progress, 2020, 57(8): 081019

    Download Citation

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

    Category: Image Processing

    Received: Sep. 2, 2019

    Accepted: Sep. 16, 2019

    Published Online: Apr. 3, 2020

    The Author Email: Xiaosong Shi (shixiaosong321@126.com)

    DOI:10.3788/LOP57.081019

    Topics

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