LiDAR点云特征变量对天然林林分因子模型反演的影响研究  

Research of the Influence of LiDAR Features on the Model Inversion of Natural Forest Stand Factors

作　　者：王浩伟 王照利 杨佳乐 段梦琦 Wang Haowei;Wang Zhaoli;Yang Jiale;Duan Mengqi(Aerial Photography and Remote Sensing Group Co.Ltd.,Xi'an,Shannxi 710199;Northwest Surveying,Planning and Designing Institute of National Forestry and Grassland Administration,Xi'an,Shannxi 710048)

机构地区：[1]中煤航测遥感集团有限公司,西安710199 [2]国家林业和草原局西北调查规划设计院,西安710048

出　　处：《中国煤炭地质》2022年第8期73-79,共7页Coal Geology of China

基　　金：中国煤炭地质总局科技创新项目(ZMKJ-2020-T04,ZMKJ-2020-J09-02)。

摘　　要：机载激光雷达技术为获取高时空分辨率的空间信息提供了新的解决方法,近年来被应用于森林资源规划设计与调查,通过空间点云数据提取林木特征,并在此基础上推演出相关林分因子。以陕西省安康市岚皋县国营林场中部分天然林为研究对象,在利用机载激光雷达点云数据进行森林资源调查时,通过对不同特征变量在不同林分因子模型反演中重要程度的分析,利用特征权重评估特征变量的贡献值,并以此为依据筛选出优秀的核心特征子集,采用随机森林算法对研究区蓄积量、林分平均高以及株数进行回归预测。实验结果表明该方法具有较优的拟合精度及较小的相对均方误差,且对不同密度下的点云数据均具有较高的稳定性。LiDAR technology provides a new solution for obtaining spatial information with high spatial-temporal resolution.In recent years,it has been gradually applied to forest resource planning,design,and investigation,extracting forest tree features through spatial point cloud data,and deducing related forest stand factors on this basis.This paper takes part of the natural forests in the State-owned Forest Farm in Langao County,Ankang City,Shaanxi Province as the research object.In this experiment,when using LiDAR point cloud data for forest resources investigation,based on the analysis of the importance of different characteristic variables in the model inversion of different forest stand factors,using feature weights to evaluate the contribution value of feature variables.Depending on this to screen out excellent core feature subsets,the random forest algorithm was used to make regression predictions on the forest stock volume,the average height of the stand and the number of trees.The result of the experiment shows that the proposed method has excellent fitting accuracy and smaller relatively mean squared error,and has high stability for point cloud data under different densities.

关 键 词：激光雷达点云 林分因子回归 随机森林 特征权重 核心特征子集 

分 类 号：TN958.98[电子电信—信号与信息处理]