优化超体素凹凸性的矿区地物提取算法研究  

Mining Feature Extraction Algorithm for Optimizing Super-Voxel Concavity and Convexity

作　　者：汪骁 周大伟[1] 占喜林 周健 刁鑫鹏[1] 耿智江 WANG Xiao;ZHOU Dawei;ZHAN Xilin;ZHOU Jian;DIAO Xinpeng;GENG Zhijiang(School of Enovironment and Spatial Informatics,China University of Mining and Technology,Xuzhou 221116,China)

机构地区：[1]中国矿业大学环境与测绘学院,江苏徐州221116

出　　处：《金属矿山》2023年第10期162-167,共6页Metal Mine

基　　金：国家自然科学基金项目(编号:52104174);宁夏回族自治区重点项目(编号:2022BEG03065)。

摘　　要：快速准确地获取矿区地物信息对煤矿安全开采工作的规划与部署具有重要的指导意义。不同于传统人工调查受到矿区环境因素的限制,机载Li DAR作为近年来矿区地表监测的新手段,可以高效获取矿区地表点云,利用点云分割算法实现矿区地物的提取。但常规点云分割算法主要适用于分割地形相对简单的区域,在复杂地形的地面点与非地面点聚类分割效果较差。基于此,提出一种优化超体素凹凸性的聚类分割算法用于矿区的复杂地形特征地物分割,主要步骤为:(1)对矿区点云进行超体素过分割;(2)根据模糊聚类算法(FCM)对超体素的边界进行细化;(3)使用增加高程阈值的凹凸聚类算法对超体素进行分割,提取机载Li DAR点云的地面点与地物点。以鄂尔多斯某矿开采工作面为例进行实验,并与改进前超体素分割算法和区域生长算法的分割效果进行比较。结果显示:整体上看,本文算法在保证数据完整的前提下,分割效果仍可以满足矿区在地物分割工程上的需求;分割细节上看,比改进前的超体素分割算法的加权平均精确度提升了13.1%,加权平均召回率提升了2.3%;比区域增长算法的加权平均精确度提升了30%,加权平均召回率提升了6.3%。Quick and accurate acquisition of ground object information in mining areas has important guiding significance for the planning and deployment of coal mine safety mining.Unlike traditional manual surveys that are limited by environmental factors in mining areas,airborne LiDAR,as a new means of surface monitoring in mining areas in recent years,can efficiently obtain surface point clouds in mining areas,and use point cloud segmentation algorithms to realize the extraction of ground features in mining areas.However,the conventional point cloud segmentation algorithm is mainly suitable for segmenting areas with relatively simple terrain,and the clustering and segmentation effect of ground points and non-ground points in complex terrain is poor.Based on this,this paper proposes a clustering segmentation algorithm that optimizes super-voxel concavity and convexity for complex terrain feature segmentation in mining areas.The main steps are:①Super-voxel over-segmentation of mining area point clouds.②The boundaries of the supervoxels are refined according to the fuzzy clustering algorithm(FCM).③Use the bump clustering algorithm with increased elevation threshold to segment the supervoxels,and extract the ground points and object points of the airborne LiDAR point cloud.This paper takes a mining face of a mine in Ordos as an example to conduct experiments,and compares the segmentation effect with the improved super-voxel segmentation algorithm and region growing algorithm.The results are as follows:On the whole,under the premise of ensuring the integrity of the data,the segmentation effect of the algorithm in this paper can still meet the needs of the mining area in the segmentation of ground features;It has increased by 13.1%,and the weighted average recall rate has increased by 2.3%;compared with the weighted average precision of the region growth algorithm,it has increased by 30%,and the weighted average recall rate has increased by 6.3%.

关 键 词：点云分割 地物提取 超体素 凹凸性 模糊聚类 

分 类 号：TD173[矿业工程—矿山地质测量]