一种基于图割的机载LiDAR单木识别方法

A graph cut-based approach for individual tree detection using airborne LiDAR data

针对现有机载激光雷达数据单木分割算法在密集林区中探测精度较低的问题,结合林木冠层空间结构分层的特点,提出一种从机载点云数据直接分离单木的方法。首先,对原始点云数据进行去噪、滤波、高程归一化;然后基于冠层高度模型计算局部最大值以确定冠层表面的明显树顶,以此作为单木位置的先验知识,继而采用归一化割(normalized cut, Ncut)方法实现冠层的初始分割;最后,以全局最大值代替局部最大值,并将冠层形状、冠层最小点数作为约束条件,再次利用Ncut方法完成对漏检单木的探测,进而实现单木的精确探测。实验结果表明,针对密集林区的单木分割,本方法有效地减少了漏识单木,整体精度达90%以上,将有助于单木三维结构定量描述及参数反演。

The accuracy of individual tree segmentation using airborne LiDAR(light detection and ranging) data is generally low for dense forests. A new method, which considers the vertical stratification of forest canopy,is proposed in this study to detect individual tree with high accuracy using airborne LiDAR data. First, several data preprocessing steps are conducted, including noise removal, point cloud filtering, and elevation normalization. Secondly, the initial canopy segmentation is achieved by the normalized cut(Ncut) segmentation with a prior knowledge of individual tree position derived from the local maximum of the canopy height model(CHM). Finally, the Ncut method is used to reduce the leakage rate of individual tree detection by setting the global maximum of CHM as the individual tree position and considering the shape and the minimum point number of the canopy. The results show that the proposed method effectively improves the accuracy of tree detection,which contributes to the quantitative description and parameter inversion in individual tree scale.