Larch Standing Tree And Fallen Wood Segmentation Based On LiDAR Point Cloud Morphology Features

Light detection and Ranging(Li DAR)technology is applied in various fields including forestry quickly by the high quality 3-dimension spatial information and high spectrum resolution respectively.Airborne Laser Scanning(ALS)is capable of obtaining individual-tree scale canopy structures from the upper canopy,providing stand scale forest parameters,and individual-tree scale forest parameters.Terrestrial Laser Scanning(TLS)has the ability to detect three-dimensional scenes within the forest,especially the fine vertical structure and geometry of the underlying canopy.Therefore,the combination of ALS and TLS can completely depict the vertical structure of the forest.We use the ALS canopy morphology to compelete high-density plantation individual tree crown(ITC)segmentation,and use the TLS data morphological characteristics to carry out fine classification and fallen wood segmentation of near-ground point clouds.The work was specifically carried out as follows:An algorithm combining the region growing segmentation and the canopy morphological characteristics is proposed for the ITC segmentation of high density plantation.First,the CHM threshold value is determined based on the CHM.This threshold value is used to control the single tree canopy range segmented by the region growing method.The possible single tree position is determined based on the point cloud distance feature;then the vertical segmentation section is constructed for the segmentation cell of the region growing.After the point cloud was projected into the two-dimensional space,the best profile was screened based on the number of profile.The crown of the best profile was selected by fitting the Gaussian function increasing one by one.The residual correlation coefficient was used to determine the number of single trees contained in each cell.Finally,K-means clustering is used to achieve ITC segmentation of point cloud and the accuracy of tree and tree height verification is detected based on the measured data.Single tree segmentation using canopy morphology can effectively improve segmentation accuracy.The accuracy rate of segmentation of 8 experimental plots is above 82%.The R2 of estimation tree height and measured height is 0.86,and RMSE was 3.2m.Experiments in different forest ages and plant density plots showed that the algorithm can achieve individual tree segmentation of high-density and closed artificial larch in point cloud scale.Terrestrial Laser Scanning(TLS)can effectively describe complex forest scenes and makes up the shortage of ALS under crown detection.The purpose of this study is to classify ground point cloud within the height of 1.3m into ground,vegetation,fallen wood,and standing trunk based on the TLS obtained from fallen wood plots in Daxing'anling.Fallen wood cloud point was segmented and merged.In order to avoid the difference in cloud density and the morphological characteristics introduced by occlusion,the optimal 3D neighborhood of each individual point was calculated through the Shannon entropy constructed by linearity,planarity,and scattering.Shannon entropy can be maximized across the increasing KNN neighbor with an interval of 5 points.The optimal neighborhood size was used to compute the covariance eigenvalues in order to construct 3D and 2D features.Key features were selected following the Recursive Feature Elimination(RFE)criteria and a random forest classification algorithm was carried out to classify the points.Noise removing approach was applied to the fallen wood points classified by self-adjust k NN features and the Random Sample Consensus(RANSAC)segmentation was implemented to segment cylinders.Fallen wood cylinders were selected and merged based on the axis direction less than 12°and the distance less than 0.1m between each other.The overall classification accuracy of self-adjust k NN method in plot A,B,and C was 93.17%,94.52%,and 95.16% respectively Point cloud of plot B and C was classified based on the model we trained using plot A.All fallen woods were the same number with that of the ground measurement and the parameters of fallen wood can be estimated roughly.Compared with the non-self-adjust k NN method,the near-ground point cloud classification accuracy has been improved by the self-adjusted k NN point cloud feature.Classification of plot B and C using the training result of plot A suggested that the selected key features in the complex forest can explain the dependent variable well.RANSAC can effectively segment the cylinder as well as estimate the parameters of the fallen wood.In conclusion,the canopy form of ALS can effectively achieve the separation of single wood canopies,and on this basis can provide single-scale forest parameters.The morphological features of TLS can achieve the fine classification of forest scenes and the division of fallen wood.Therefore,the combination of the two will have great practical value and application prospects in the construction of complete forest three-dimensional scenes,the extraction of high-precision forest parameters,and the investigation of forest resources.