基于PointNet的三维点云部件分割的抗干扰研究  

Anti Interference Research of 3D Point Cloud Component Segmentation Based on PointNet

作　　者：穆莉莉[1] 单卓佳 Mu Lili;Shan Zhuojia(School of Mechanical Engineering,Anhui University of Science and Technology,Huainan,Anhui 232001)

机构地区：[1]安徽理工大学机械工程学院,安徽淮南232001

出　　处：《绥化学院学报》2023年第8期144-147,共4页Journal of Suihua University

基　　金：安徽省重点研究与开发计划项目(202004a07020046)

摘　　要：三维点云部件分割的训练预测模型是复杂多样的,影响分割结果的因素有很多。针对深度神经网路设置不同的参数对分割精度的影响,提出在经典点云网络模型PointNet上选择不同的激活函数与优化器进行抗干扰研究。首先将网络中数据采样数、学习率、批量大小等参数确定,其次修改激活函数与优化器参数并在ShapeNet数据集上进行实验对比。实验结果表明,在PointNet网络模型中选择ReLU激活函数与SGD优化器测试精确度达到88.39%。The training and prediction models of 3D point cloud component segmentation are complex and diverse,and there are many factors that affect the segmentation results.Aiming at the influence of setting different parameters of the depth neural network on the segmentation accuracy,this paper proposes to select different activation functions and optimizers on the classical point cloud network model PointNet for training and testing.Firstly,the data sampling number,learning rate,batch size and other parameters in the network are determined;Secondly,modify the parameters of the activation function and optimizer,and make experimental comparison on the ShapeNet dataset.The experimental results show that choosing ReLU activation function and SGD optimizer in PointNet network model can achieve higher segmentation accuracy.

关 键 词：点云部件分割 PointNet 深度神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]