A Hierarchic Method of Individual Tree Canopy Segmentation Combing UAV Image and LiDAR
Wang, Rui,Han, Jiangang,Hu, Xiangdong,Zhao, Yuankun,Wang, Qian,Sun, Hao,Xie, Yuning
DOI: https://doi.org/10.1007/s13369-024-09465-w
IF: 2.807
2024-08-26
Arabian Journal for Science and Engineering
Abstract:The individual tree crown segmentation (ITCs) technology based on unmanned aerial vehicle (UAV) remote sensing is an important means for remote forestry parameter investigation. However, due to the characteristics of tree canopy heterogeneity and high canopy density in forest land, ITCs accurate segmentation of forest land still faces challenges. In this paper, a method, which combines UAV point cloud and image to obtain spatial information and texture information of forest land, is proposed so as to achieve accurate segmentation of individual tree crowns. Firstly, edge-energy function is proposed for UAV image segmentation. Subsequently, the approach involves using a improved method of kernel density estimation on the long axis and improved Gaussian mixture model (GMM) to segmenting three-dimensional (3-D) clustering components. To validate the efficacy of the algorithm proposed in this research, we used 5 plots to conduct comparative experiments. The experimental findings demonstrated that our algorithm outperforms the other three methods across various stand densities, yielding a higher f -value. Especially in plot3, the f -value of the method is at least 0.82, which is 0.07, 0.04 and 0.04 higher than that of BMask R-CNN, layer stacking and GMM, respectively. Moreover, Our algorithm still shows superiority in the segmentation of different tree species. The experimental results show that our algorithm can not only provide accurate segmentation results in dense forest areas, but also has generalization.
multidisciplinary sciences