| Tree structure parameters are important contents of forest resources investigation,and also the basis of dynamic monitoring and management of forest resources.How to quickly,accurately and non-destructively obtain the tree structure parameters and trunk volume are urgent problems which need to be solved by traditional survey methods.The Terrestrial Laser Scanning technology(TLS)can make up for the deficiency of traditional manual investigation by its high precision three-dimensional information representation capability.How to efficiently and completely extract the trunk points from the TLS tree point cloud,and meanwhile retain the properties of the trunk,needs to be further studied.It is very useful for the extraction and subsequent application of corresponding forest parameters.TLS can obtain the point cloud data of sample tree through non-contact and non-destructive way,which provide a new method for the research of taper equation.This paper will take Baima teaching and research base of Nanjing Forestry University and Jiangsu Huanghai seashore National Forest Park(formerly known as Dongtai Forest Farm)as research areas.Based on the TLS data of poplar plantation,the extraction of the trunk points,estimation of tree parameters and construction of taper function are studied.The main results and conclusions are as follows:(1)When using the spatial distribution characteristics(flatness and normal vector)of single tree point cloud data to extract the tree trunk points,the threshold setting has a great influence on the recall rates and precision rates.The results show that when the flatness greater than 0.80 and the angle between the normal vector and the XY plane less than 30 °,the extraction of poplar single tree trunk points is better,which can extract about 80% of the real trunk points,and the noise points are less than 10%.(2)A method of extracting tree trunk points based on point cloud feature by using semi-supervised SVM is proposed.The point cloud feature matrix is obtained by calculation,which is composed of flatness,normal vector,structure tensor,normal vector distribution and spherical neighborhood point normal vector.Combined with machine learning semi-supervised SVM method,the tree trunk points in sample plot are extracted.This method can extract almost 95% of the real trunk points and noise points less than 2%.Compared with the method based on spatial distribution characteristics,this method extracts the tree trunk points better.This method does not need to set the parameter threshold manually,and retains the properties of the trunk very well,which provides foundation for single tree 3D modeling and tree parameter extraction.(3)Tree diameter parameter is obtained by extracted trunk points.The results show that the least square method is better than the convex hull algorithm when extracting DBH.The fitting results of DBH are very close to the measured values,with r=0.9974,RMSE= 0.75 cm,which can meet the actual production needs.The rest stem diameters are extracted by least square method.The correlation coefficient r between the extracted and measured values of the rest stem diameters is 0.9482,with RMSE= 0.42 cm.Therefore,it is feasible to extract stem diameters from TLS data.(4)The extraction method of tree height and height to crown base are improved,and meanwhile the crown width is obtained by extracted trunk points.Correlation coefficient r between the measured value and the extracted value of tree height is up to 0.9593.For the little sample trees,the difference between the extracted tree height and the measured value are small.Therefore,this method can be used to extract the tree height of small broadleaf tree.The accuracy will drop when broadleaf tree is tall and big,because the point cloud at the top of the tree will be covered by the branches and leaves.Correlation coefficient r between the measured value and the extracted value of height to crown base is up to 0.9952,which means that this method has high extraction accuracy.At the same time,TLS data are used to extract the crown width,and the r between the measured crown width and the extracted value is less than 0.90,and the RMSE value is greater than 1.10 m.The accuracy of crown width extraction is general,and further study is needed.(5)Based on the tree trunk points,the taper function of poplar plantation is fitted.For Dongtai forest farm,the fitting accuracy of Schumacher and Hall,Ormerod and Zeng Weisheng model is better,with R2 all greater than 0.9482.The verification accuracy is also excellent,with R2 greater than 0.990.These three models have strong applicability,which fits the growth law of poplar.For Baima teaching and research base,the fitting accuracy and verification result of Schumacher and Hall model are great.The volumes of standing tree are estimated by the optimal taper function,and the relative difference is small when compared with the estimation results of the binary volume equation,and there is no significant difference between them.The research shows that the proposed tree trunk points extracting method can retain most of the real trunk points,and there are almost no noise points.The obtained tree trunk points can improve the extraction accuracy of tree parameters,also provide data basis for drawing tree trunk curve,taper function establishment,tree trunk volume estimation and 3D modeling.The feasibility and reliability of non-contact TLS application in forest resource investigation are verified. |
PDF Full Text Request