Potential of hyperspectral LiDAR in individual tree segmentation: A comparative study with multispectral LiDAR

Individual tree extraction is the basis of obtaining structural parameters of individual tree which will helpful for the inventory, management and protection of urban forest. Single wavelength Light Detection and Ranging (LiDAR) has been widely applied for individual tree segmentation to obtain accurate structural parameters of trees, and it is difficult to segment individual tree of clumped trees. Then, multispectral and hyperspectral LiDAR was developed and used in vegetation remote sensing. Therefore, the main purpose of this study is to discuss the performance of hyperspectral LiDAR on individual tree segmentation with different tree density. Due to the airborne hyperspectral LiDAR data is lack at present, then the simulated airborne hyperspectral LiDAR was used in this study. Random forest algorithm was used to extract tree point clouds. Spectral information and spatial information of simulated hyperspectral LiDAR were applied in individual tree segmentation based on spectral clustering algorithm. Experimental results showed that F1 scores of multispectral and simulated hyperspectral LiDAR in tree point clouds extraction were 89 % and 94 %, respectively. The detection rate of individual tree segmentation by using multispectral and hyperspectral point data in low (medium, high) tree density were 83.53 % (82.46 %, 66.93 %) and 87.05 % (94.74 %, 87.40 %), respectively. It can be found that integrating spectral information can better extract tree point cloud in complex environments and can efficiently improve the accuracy of individual tree segmentation. Therefore, hyperspectral LiDAR possessed the potential for the individual tree segmentation in high tree density region.