基于无人机影像的树顶点和树高提取及其影响因素分析

Tree Vertex and Height Extraction Based on UAV Imagery and Analysis on its Influencing Factors

对基于无人机影像生成的树冠高度模型(Canopy Height Model,CHM),采用局部最大值算法进行树顶点和树高提取的可行性进行了探讨。此外,还探讨了分辨率、窗口大小对于树顶点提取的影响。以密集的针阔混交林为样地,利用SfM(Structure from Motion)算法结合无人机影像对研究区进行三维重建,得到点云、数字表面模型(Digital Surface Model,DSM)、数字高程模型(Digital Elevation Model,DEM)等一系列三维数据并生成CHM。然后,对不同分辨率的CHM使用不同的平滑窗口大小、移动窗口大小组合进行树顶点的提取并对结果进行精度评价。当CHM分辨率为0.4m,平滑窗口大小为3×3像元,移动窗口大小为3×3像元时,树顶点的提取精度最高,F测度为77.08%。将基于该组合提取正确的37个树顶点对应的提取树高与实地测量得到的树高对比,R 2为0.966 9,RMSE为1.411 4m,rRMSE=10.69%。研究结果表明:利用无人机影像可以较好地提取复杂树林的树顶点和树高;基于局部最大值算法提取树顶点,需要根据实际情况确定CHM的分辨率、平滑窗口大小和移动窗口大小,以获得最佳提取结果。

In this paper,the feasibility of using local maximum algorithm to extract tree vertices and tree height based on the canopy height model(CHM)generated by UAV imagery was discussed.In addition,the effect of resolution and window size on tree vertex extraction was also discussed.In this study,a dense conifer-broadleaf forest is used as sample plot.The SfM(Structure from Motion)algorithm was combined with the UAV images to reconstruct the research area in three dimensions,and a series of three-dimensional data such as point cloud,digital surface model(DSM),digital elevation model(DEM)and CHM were generated.Then,for different resolutions of CHM,different smoothing window size and moving window size combinations were used to extract the tree vertices and the accuracies of results were evaluated.When the CHM resolution is 0.4m,the size of smoothing window is 3×3 pixels,and the size of moving window is 3×3 pixels,the extraction accuracy of tree vertex is the highest,and the F-Measure is 77.08%.The extracted tree heights corresponding to the 37 tree vertices extracted based on the combination was compared with the tree heights measured in the field,and the R 2,RMSE,rRMSE is 0.9669,1.4114m,10.69%respectively.The results showed that UAV imagery can be used to extract tree vertices and tree heights of complex forests.Extracting the tree vertices based on the local maximum algorithm needs to determine the resolution of the CHM,the smoothing window size and the moving window size according to the actual situation to obtain the best extraction result.