【摘要】 针对机载激光雷达点云中基于栅格化的冠层高度模型(CHM)所导致的原始点云数据丢失问题,提出了一种应用高斯模型聚类的单木信息提取方法。采用形态学开运算和高斯平滑方法形成高斯冠层最大模型(GCMM)能减少无关局部最大值对单木分割的影响,利用局部最大值法初步探测树冠顶点,通过最速下降法建立混合高斯模型得到树木位置和冠幅。利用聚类分析划分临近点云归属,进而实现单木参数准确提取,并提取单木最高点为树高。将点云分割方法应用于美国蓝岭地区6块圆形针叶林样地(r=30 m)。结果表明:单木分割F为0.89,正确分割单木树高提取精度95%,冠幅提取精度91%。结合实测数据对提取到的树高和冠幅进行相关性分析,树高R~2=0.92,平均误差为-0.83 m;冠幅R~2=0.84,平均误差为-0.42 m。相比于分水岭算法,高斯模型聚类方法F提高了11.2%,正确分割单木树高及冠幅提取精度提高了5.5%、5.8%,树高R~2提高0.08,平均误差减少0.58 m;冠幅R~2提高0.11,平均误差减少0.63 m。更多还原

【Abstract】 Aiming at the problem of the original point cloud data loss caused by the rasterized canopy height model(CHM) in the airborne LIDAR point cloud, a single tree information extraction method with Gaussian model clustering was proposed. The use of morphological opening operation and Gaussian smoothing method to form Gaussian canopy maximum model(GCMM) can reduce the influence of irrelevant local maximum on single tree segmentation. The local maximum method is used to initially detect the crown apex, and the mixed Gaussian model is established by the steepest descent method for tree location and crown width. Cluster analysis is used to divide the attribution of adjacent point clouds, so as to achieve accurate extraction of single tree parameters, and extract the highest point of single tree as tree height.The point cloud segmentation method was applied to six circular coniferous forest plots(r=30 m) in the Blue Ridge region of the United States. The single tree segmentation F was 0.89, the correct extraction of single tree segmentation had a high extraction accuracy of 95%, and the extraction accuracyof the crown width is 91%. The correlation between the extracted tree height and crown width is analyzed with the measured data. The tree height R~2 is 0.92, and the average error is-0.83 m; the crown width R~2 is 0.84, and the average error is-0.42 m. For the watershed algorithm, the Gaussian model clustering method F is increased by 11.2%, the accuracy of the correct segmentation of single-tree trees and the extraction of crown width are increased by 5.5% and 5.8%, the tree height R~2 is increased by 0.08, and the average error is reduced by 0.58 m. The crown width R~2 is increased by 0.11, and the average error is reduced by 0.63 m.更多还原