| 基于地面激光雷达点云数据的树种识别方法 |
| |
| 引用本文： | 王 佳,张隆裕,吕春东,牛利伟.基于地面激光雷达点云数据的树种识别方法[J].农业机械学报,2018,49(11):180-188. |
| |
| 作者姓名： | 王 佳 张隆裕 吕春东 牛利伟 |
| |
| 作者单位： | 北京林业大学,北京林业大学,北京林业大学,北京林业大学 |
| |
| 基金项目： | 国家自然科学基金项目（41401650）、北京市自然科学基金项目（8182038）和中央高校基本科研业务费专项资金项目（2015ZCQ-LX-01） |
| |
| 摘 要： | 为了能够更有效地利用地面激光雷达的点云数据识别树种，以北京林业大学为研究区域，利用FARO Photon 120型地面激光雷达在研究区内获取4个树种、共92棵树木的点云数据。依据点云的三维坐标值提取研究区内立木的胸径、枝下高、树高、冠高、最长冠幅、垂直最长方向冠幅6个测树因子，同时提取由测树因子组合而成具有鲁棒性的6个树形特征参数，包括冠长树高比、胸径树高比、冠高树高比、分枝角、冠长最大冠幅之比、最长冠幅与垂直方向冠幅之比。分别使用测树因子和组合特征参数，采用支持向量机、分类回归决策树和随机森林的方法，对树种进行冠幅自动识别。研究结果表明：使用测树因子树木识别方法，识别平均准确率为0.765，平均召回率为0.778，3种识别方法中，分类效果较好的依次为分类回归决策树、随机森林、支持向量机；使用组合特征参数树木识别方法，识别平均准确率为0.891，平均召回率为0.896，分类效果较好的方法是随机森林和支持向量机，其次是分类回归决策树；总体上来看，不论是对于单个树种还是总体的准确率和召回率，组合特征参数法均高于测树因子法，而对于3种不同的分类方法，随机森林相对最好。研究结果表明，结合地面激光雷达获取的点云和不同机器学习分类方法进行树种识别分类可以达到满意的效果，且能节省大量时间和人力。
|
| 关 键 词： | 树种识别 地面激光雷达 点云 测树因子 组合特征参数 |
| 收稿时间： | 2018/7/13 0:00:00 |
Tree Species Identification Methods Based on Point Cloud Data Using Ground-based LiDAR |
| |
| WANG Ji,ZHANG Longyu,LV Chundong and NIU Liwei.Tree Species Identification Methods Based on Point Cloud Data Using Ground-based LiDAR[J].Transactions of the Chinese Society of Agricultural Machinery,2018,49(11):180-188. |
| |
| Authors: | WANG Ji ZHANG Longyu LV Chundong and NIU Liwei |
| |
| Institution: | Beijing Forestry University,Beijing Forestry University,Beijing Forestry University and Beijing Forestry University |
| |
| Abstract: | The traditional tree species identification depends on time consuming and labor intensive efficiency of artificial field survey. In order to more effectively utilize the point cloud data identification tree of ground based LiDAR, taking Beijing Forestry University as the research area, and FARO Photon 120 ground based LiDAR was used to obtain point cloud data of a sample set of 92 trees, four tree species in the study area. According to the three dimensional coordinate values of point cloud, the six tree measuring factors of breast diameter, height of branches, height of tree, height of crown, width of crown, and the longest direction of vertical trees in the study area were extracted, and the extracted tree measuring factors were combined. The robust tree features six parameters, namely crown length tree height ratio, DBH height ratio, crown height tree height ratio, branch angle, crown length ratio, maximum crown width and vertical direction. For the ratio of crown width, the tree species were automatically identified by using the tree measuring factor and the combined feature point parameters to support the tree sample by using the support vector machine, the classification regression decision tree and the random forest. The results showed that for the tree identification method using tree measuring factor, the average accuracy of recognition was 0.765, and the average recall rate was 0.778. Among the three identification methods, the best effect was classification regression decision, followed by random forest, and finally support vector. Using the combined feature parameter tree identification method, the average accuracy of recognition was 0.891, and the average recall rate was 0.896. The best method was random forest and support vector machine, followed by classification regression decision. In general, the combined feature parameter method had higher accuracy and recall rate of single tree species or overall than those of the tree measuring factor method, random forests were relatively the best for three different classification methods. The research result showed that the tree species identification classification combining the point cloud obtained by ground-based LiDAR and different machine learning classification methods could achieve satisfactory results and save a lot of time and manpower. |
| |
| Keywords: | tree species classification ground based LiDAR point cloud tree measuring factor combined feature parameters |
|
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
