基于网络图的地基激光雷达复杂树木点云枝叶分离方法

地基激光雷达树木点云数据的枝叶分离是精确计算地上生物量和叶面积指数的重要前提，也是树木三维建模的重要步骤。然而，山地复杂树木冠幅大且结构复杂，从而造成树叶与枝干之间的相互遮挡，因此很难获取高质量的点云数据，目前对其实现枝叶分离依然存在较大的困难。利用地基激光雷达FARO Focus3DX330获取三维激光点云数据，提出了一种基于网络图的树木点云枝叶分离方法。首先，采用LeWos模型对点云进行初步的枝叶分离，分离出枝干和叶片点云。在此基础上，针对枝干和叶片混合点云通过路径追踪检测算法来精细分离枝干和叶片。随着路径长度从10增加到100，枝干点不断增加，叶片点不断减少，枝叶分离精确度、枝干F分数、叶片F分数、Kappa系数均先增加后减少。综合这4项精度评价指标，选取各个树木最优路径长度执行路径追踪检测算法。通过与LeWos模型、Tlseparation模型和高斯混合模型等主流枝叶分离方法比较发现，该方法精度更优，精确度为91.97%。而且，该方法的枝干F分数和叶片F分数均大于85%，这表明该方法具有很好的平衡性。该方法仅使用路径长度，不考虑几何特征，因此极大地提高了针叶树木的枝叶分离精度。...

A Method for Separating Leaf and Wood Components of Complex Tree Point Cloud Data based on Network Graph with Terrestrial Laser Scanning

The leaf and wood separation of the terrestrial laser scanning tree point cloud data is an important prerequisite for the accurate estimation of above-ground biomass and leaf area index， and it is also an important step for three-dimensional modeling of a tree. However， complex trees in mountain areas have large crowns and complex structures， resulting in mutual occlusion between leaves and branches. Therefore， it is difficult to obtain high-quality point cloud data. At present， it is still difficult for complex trees to separate leaf and wood components. High-resolution point clouds were acquired with Faro Focus^3D X330. This paper proposes a method for leaf and wood separation of tree point cloud based on network graph. First， the LeWos model is used to perform preliminary leaf and wood separation on the point cloud， and separate the wood and leaf point cloud. On this basis， the path retrace detection algorithm is used to finely separate the leaf and wood for the mixed point clouds. As the retrace steps increases from 10 to 100， the wood points continue to increase， the leaf points continue to decrease， the accuracy， wood F-score decreases， leaf F-score and the Kappa coefficient first increases and then decreases. By comparing with LeWos model， Tlseparation model and Gaussian mixture model， it is found that the research method in this paper has the better precision， with an accuracy of 91.97%. Moreover， the wood F-score and leaf F-score of the proposal method are both greater than 85%， which means that the proposal method has a good balance when classifying wood and leaf. The proposal method only uses the retrace steps and does not consider the geometric characteristics， so the accuracy of branch and leaf separation of coniferous trees is greatly improved. At the same time， the method in this paper has a good effect on the separation of wood and leaf components of point clouds with different densities and different species. Therefore， the method in this paper is more robust. Accurate wood and leaf separation of tree point clouds is of great significance to forest resource management and biodiversity research.