Research On Single Tree Segmentation And DBH Parameter Extraction Algorithm Based On Point Cloud Data

Forest resources are one of the most important natural resources.Accurate and efficient acquisition of the location and parameter information of individual trees in the forest is the most basic task in forestry investigations,and it is also an important prerequisite for carrying out various forestry ecological studies.As an active remote sensing technology,lidar technology has developed rapidly in recent years.It can obtain point cloud data with high precision,high efficiency,and a wide range,and can effectively express forest structure information,which is playing an increasingly important role in forestry investigations.The more important role,the single-tree segmentation and DBH parameter extraction based on point cloud data has become a research hotspot.At present,the research of single-tree segmentation algorithms based on point cloud data is mostly based on specific types of forests,and the accuracy assessment of tree segmentation algorithms has not yet formed a unified standard,which makes it difficult to assess the advantages and disadvantages of the algorithm.The research of the tree DBH extraction algorithm is mostly focused on whether the accuracy of the algorithm meets the requirements,and there is little research on the accuracy comparison.The user cannot choose an effective DBH extraction algorithm when faced with different shapes of trunk point cloud data.Focusing on the above practical problems,this paper uses ground-based lidar to collect point cloud data of coniferous forest,broad-leaved forest,coniferous broad-leaved forest,and virgin forest,and conducts sufficient and detailed experiments through related processing procedures and algorithms.First of all:The point cloud data of coniferous forest,broad-leaved forest,coniferous broad-leaved forest and primitive forest were processed by using watershed segmentation method,Hough transform circle detection segmentation method and regional growth segmentation method respectively.The algorithm handles the discovery of point cloud data for different woodland types:The three algorithms have obvious differences in segmentation accuracy,but they also show certain rules.In coniferous forests,the regional growth segmentation algorithm and watershed segmentation algorithm have better accuracy.The recall rates are 70% and 54%,and the accuracy rates are 88.1% and 87%.For broad-leaved forests,the Hough transform circle detection algorithm can accurately segment trees,The recall rate and correct rate can reach 83.6% and 86.8%.Compared with the first two woodland types,the accuracy of the three algorithms for coniferous and broad-leaved mixed forests is average.Although the recall rate of the regional growth algorithm is optimally 61.2% among the three algorithms,the accuracy rate is only 48%;For primitive forests,the segmentation accuracy of these three algorithms is generally very low,and their applicability is poor.The recall rate of the regional growth algorithm is up to 53%,and the accuracy rate is only 35.6%.For the DBH extraction algorithm,this paper uses least squares method,convex hull algorithm and Hough transform circle detection algorithm to extract the DBH of 100 trees in the plot.Through comparison and analysis,it can be seen that the least square method is better than the other two algorithms in accuracy,most of the absolute errors are between 0-5mm,and for irregularly shaped trunks,the accuracy of extracting the diameter of the breast diameter is at a disadvantage.Although the convex hull algorithm is less accurate than the least square method,the stability of the algorithm is higher,and it has better applicability for the extraction of the diameter of the deformed trunk.The Hough transform circle detection algorithm has the lowest accuracy in extracting the diameter of the chest.And the stability is also poor.In this research work,three algorithms are used for the segmentation of the single tree based on the point cloud data and the extraction of the breast diameter of the single tree.In broad-leaved forests,the Hough transform circle detection algorithm can accurately segment a single tree.The accuracy of the three algorithms for coniferous and broad-leaved mixed forests is average,and the accuracy of the three algorithms for primitive forests is generally very low.The results of DBH extraction show that the least square method has the best accuracy,but the stability is poor.The convex hull algorithm has lower accuracy,but its stability is better.The Hough transform circle detection algorithm has the worst accuracy and poor stability.