局部结构增强的三维点云分类  

3D Point Cloud Classification with Local Structure Enhancement

作　　者：杨涛[1] 周嘉持 韩景红[1] 任贺春[1] 李瑞瑞 陈子逸 YANG Tao;ZHOU Jia-chi;HAN Jing-hong;REN He-chun;LI Rui-rui;CHEN Zi-yi(China Science and Technology Museum,Beijing 100101,China;Information Engineering College,Capital Normal University,Beijing 100048,China;Information College,Beijing University of Chemical Technology,Beijing 100029,China)

机构地区：[1]中国科学技术馆,北京100101 [2]首都师范大学信息工程学院,北京100048 [3]北京化工大学信息学院,北京100029

出　　处：《数学的实践与认识》2025年第3期179-186,共8页Mathematics in Practice and Theory

摘　　要：3D点云分类任务面临诸多挑战,其中点云数据的无序性和分布不均问题尤为突出.现有的方法大多缺乏对局部拓扑结构的深入理解和建模,导致整体分类准确率不高针对这一问题,本文提出了一种创新且易于实现的分类方法.该方法通过设计局部拓扑编码增强模块,嵌入局部几何信息,从而显著提升了分类精度.该模块可作为插件灵活整合到现有的3D点云分类网络中.实验部分,将该模块应用于PointNet和PointNExT网络,并在ModelNet40、ScanObjectNN和ShapeNetPart数据集上进行测试.结果显示,该方法有效提升了分类准确率,表现出较强的适应性,为3D点云分类提供了一种高效且实用的解决方案.This study focuses on the task of 3D point cloud classification and proposes an innovative yet easy-to-implement classification method aimed at addressing the unordered nature and uneven distribution of point cloud data.The designed local topology encoding enhancement module embeds local geometric information,significantly improving classification accuracy.It can be flexibly integrated into state-of-the-art 3D point cloud classification networks as a plug-and-play component.We applied this module to PointNet and Point-NExT networks and conducted tests on the ModelNet40,ScanObjectNN,and ShapeNetPart datasets.The results demonstrate that the proposed method effectively enhances classification accuracy,showcasing strong adaptability and providing an efficient and practical solution for 3D point cloud classification.

关 键 词：残差网络 多层感知机 局部拓扑 点云 3D分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]