Individual-Tree Segmentation and Extraction based on LiDAR Point Cloud Data

Abstract

To extract forest parameters and individual tree information accurately and efficiently from plantations, this study focuses on a plantation of Pinus tabulaeformis in Chongli District in China. Utilizing LiDAR point cloud data and ground-measured data from 30 plots, we examined the sensitivity of individual tree segmentation to key parameters by varying the grid values of the point cloud distance discriminant clustering algorithm and adjusting the canopy height resolution (CHR) of the watershed algorithm. The objective was to identify the optimal parameters for both algorithms in terms of tree height extraction precision. In the task of individual tree extraction, the point cloud distance discriminant clustering algorithm outperformed the watershed algorithm. This was evidenced by significantly higher recall, precision, and F1-score. However, in terms of tree height precision, as measured by the coefficient of determination and root mean square error (RMSE), the watershed algorithm proved superior. Specifically, the watershed algorithm achieved a coefficient of determination of 0.88 and an RMSE of 0.93 meters, indicating greater precision in estimating tree parameters. Nonetheless, the optimal parameter settings for the watershed algorithm need to be adjusted based on stand density. Thus, through this study, we found that for individual-tree extraction from LiDAR point cloud data, the initial setting of different grid values and resolutions has a significant impact on segmentation precision. It is essential to design tailored approaches for processing point cloud data under varying environmental conditions to achieve optimal results and precision.