多场景机载LiDAR点云单木提取

Individual tree extraction in multiple scenarios with airborne LiDAR point cloud data

机载激光雷达(LiDAR)测量技术点云数据采集效率高、覆盖范围大,在树木三维空间采集方面具有显著的优势。本文在城市绿化林地、农村果树林地、山区经济林地这三种典型场景中各选取了一块实验区,采用多级分类的思路,通过粗差点检测、点云滤波、高度阈值分割、平面生长等处理步骤,对整个区域的点云数据进行语义标号,相关的语义类别包括:粗差、地面、植被和建筑物,再使用谱聚类算法对植被类点云进行单木提取。实验结果表明,城市绿化林地树木分布规整,树冠彼此相对独立,检测率最高(83.6%);而农村果树林地存在植被点与房屋点更难区分,低矮树木漏识别多于城市场景,检测率略低于城市场景(81.2%);山区经济林地树木密度最高,树冠相互粘连、遮盖,检测率最低(76.8%)。

Airborne light detection and ranging(LiDAR)technology features high acquisition efficiency and large coverage of point cloud data,which has significant advantages in the three-dimensional spatial acquisition of trees.In this paper,experimental areas were selected from three typical scenarios of urban green woodland,rural fruit-bearing forests,and economic forests in mountainous areas,and the principle of multi-level classification was adopted.The point cloud data of the entire area was semantically labeled through the processing steps of gross error point detection,point cloud filtering,height threshold segmentation,plane growth,etc.,and the related semantic categories included gross error,ground,vegetation,and buildings.Then,the spectral clustering algorithm was used to extract individual trees from vegetation point clouds.The experimental results show that the distribution of trees in urban green woodland is regular,and the canopy is relatively independent of each other.The detection rate is the highest(83.6%).However,it is more difficult to distinguish between vegetation points and building points in rural fruit-bearing forests,and the identification rate of shrub-like trees is lower than that in the urban scenario.In addition,the detection rate is slightly lower than that in the urban scenario(81.2%).The density of trees in economic forests in mountainous areas is the highest,and the canopies are close and shaded by each other,with the lowest detection rate(76.8%).