基于三维点云的岩体结构面识别与快速聚类分析
Identification and Fast Clustering Analysis of Rock Discontinuity Surfaces Based on 3D Point Cloud
孔夏丽 1夏永华 2鄢敏 1太浩宇 1李晨 1朱琪3
作者信息
- 1. 昆明理工大学国土资源工程学院,云南昆明 650093
- 2. 昆明理工大学城市学院,云南昆明 650051
- 3. 中国电建集团昆明勘测设计研究院有限公司,云南昆明 650051
- 折叠
摘要
岩体结构面识别与聚类分析是研究岩体结构特征和评估岩体稳定性的基础.为对岩体结构面进行快速、有效聚类,提出一种基于三维点云的岩体结构面识别与快速聚类方法.首先,通过FACET进行点云分割和平面拟合,提取岩体结构面.其次,通过不同岩体结构面之间的相似性距离,计算局部密度和控制距离,并绘制决策图,自动寻找聚类中心和聚类数量.最后,根据边界密度,将结构面划分为核心结构面和离群结构面,剔除异常值.该方法避免了人为主观因素的干扰,提高了聚类分析的准确性.通过对立方体、六面体进行聚类分析,聚类数量与预期相一致,且每簇平均产状与点云结构面拟合结果相近,倾向最大误差分别为0.47°、1.78°,倾角最大误差分别为2.98°、2.57°.同时,聚类性能与K-means、K-means++和DBSCAN聚类算法相比,有了一定程度的提高,最大可达0.834.将其运用于四川省会东县老君峰南侧高陡悬崖岩体结构面分析,无须指定聚类中心和簇数,聚类结果与实测产状、RocScience dips结果相近,精度满足要求,性能较好.
Abstract
The identification and clustering analysis of rock discontinuities are the basis for studying the structural characteris-tics of rock masses and assessing the stability of rock masses.In order to perform fast and effective clustering of rock body dis-continuities,a 3D point cloud-based rock body discontinuity identification and fast clustering method is proposed.Firstly,the point cloud segmentation and plane fitting are performed by FACET to extract the rock body discontinuity surface.Secondly,the local density and control distance are calculated by the similarity distance between the rock discontinuity faces,and the deci-sion map is drawn to find the clustering center and the number of clusters automatically.Finally,according to the boundary density,the rock discontinuities are divided into core discontinuities and outlier discontinuities,and the outliers are eliminated.This method avoids the interference of human subjective factors and improves the accuracy of clustering analysis.Through the clustering analysis of cubic and hexahedral,the number of clusters is consistent with the expectation,and the average yield of each cluster is similar to the fitting results of point cloud discontinuity surface,with the maximum error of dip direction 0.47° and 1.78°,and the maximum error of dip angle 2.98° and 2.57°,respectively.At the same time,the clustering performance is improved to a certain extent compared with K-means,K-means++and DBSCAN clustering algorithms,up to 0.834.Field application to the discontinuous surface of a high,steep cliff in Huidong County,Sichuan Province,demonstrates its effective-ness without predefined clustering centers or numbers,yielding results comparable to measured data and RocScience dips,thereby satisfying accuracy requirements and exhibiting robust performance.
关键词
岩体结构面/三维点云/聚类分析/密度峰值/快速搜索/聚类中心Key words
rock discontinuity/three-dimensional point cloud/cluster analysis/peak density/quick search/clustering center引用本文复制引用
出版年
2024
NETL
NSTL
万方数据