基于PointNet优化网络的铁路站台语义分割  

Railway platform semantic segmentation based on PointNet optimized neural network

作　　者：鲁子明 黄世秀[1] 季铮[2] 张思仪 黄翔翔 LU Ziming;HUANG Shixiu;JI Zheng;ZHANG Siyi;HUANG Xiangxiang(College of Civil Engineering,Hefei University of Technology,Hefei 230009,China;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430072,China;School of Computer and Information,Anhui Polytechnic University,Wuhu 241000,China)

机构地区：[1]合肥工业大学土木与水利工程学院,安徽合肥230009 [2]武汉大学遥感信息工程学院,湖北武汉430072 [3]安徽工程大学计算机与信息学院,安徽芜湖241000

出　　处：《现代电子技术》2024年第3期68-72,共5页Modern Electronics Technique

基　　金：安徽省自然科学基金项目(2208085QD106)。

摘　　要：铁路站台点云语义分割是对铁路侵界现象进行检测的关键环节。文中以新型激光扫描测量系统采集的具有三维空间信息的点云数据为基础,在获取初步分割结果的基础上,设计PointNet网络整体结构提取点云数据全局特征,采用多层次金字塔结构对网络进行局部特征提取优化,实现铁路站台点云数据语义分割。研究表明,所提方法对实验点云数据的分割准确率达到84.5%,在铁路工程应用中的点云总体分割精度达到75.34%,在铁路检测中实现了大范围多尺度点云数据的可靠语义分割,满足铁路侵界现象检测分析需求。The semantic segmentation of railway platform point clouds is a key step in detecting railway intrusion.The article is based on point cloud data with three⁃dimensional spatial information collected by a new laser scanning measurement system.On the basis of obtaining preliminary segmentation results,the overall structure of the PointNet network is designed to extract global features of point cloud data.A multi⁃leveled pyramid structure is used to optimize local feature extraction of the network,so as to achieve semantic segmentation of railway platform point cloud data.The research has shown that the proposed method achieves a segmentation accuracy of 84.5%for experimental point cloud data,and the overall segmentation accuracy of point clouds in railway engineering applications reaches 75.34%.It achieves reliable semantic segmentation of large⁃scale multi⁃scale point cloud data in railway detection,which meets the needs of railway intrusion detection and analysis.

关 键 词：点云分割 深度学习 铁路站台 铁路侵界 PointNet 金字塔结构 深度神经网络 语义分割 

分 类 号：TN711-34[电子电信—电路与系统] TP391.4[自动化与计算机技术—计算机应用技术]