基于多层K-means在森林点云中的单木识别算法

doi:10.13466/j.cnki.lyzygl.2022.01.015

摘要/Abstract

摘要：

针对目前激光雷达(Light Detection And Ranging,LIDAR)采集的森林点云数据使用K-means算法进行单木识别时,算法的收敛时间长,以及对于株树密度大的森林场景容易产生过聚类的问题,提出了一种改进的多层K-means单木识别算法。以黑龙江省佳木斯市孟家岗林场落叶松人工林的点云数据为实验对象,通过RANSAC算法和半径离群值算法去除数据中的地面点与非树干且非地面点,最后通过多层K-means算法进行单木的识别。结果表明,改进后的多层K-means算法单木识别的识别率达到了91.01%,误判树木的数量为0,算法收敛时间相较于传统K-means算法缩短了48.13%。可以得出结论,多层K-means算法效率更高,对复杂密集的林分样地进行单木识别有更好的效果,降低了勘测森林结构的成本,这对于森林结构参数的测算、森林资源的保护与总体规划工作有重要意义。

关键词: 森林结构, 点云数据, K-means算法, 单木识别, 激光雷达

Abstract:

When the K-means algorithm is used for single tree recognition for the forest point cloud data collected by the current Lidar (Light Detection And Ranging,LIDAR),the algorithm has a long convergence time and is prone to clustering for forest scenes with high density of trees. An improved multi-layer K-means single tree recognition algorithm was proposed.Taking the point cloud data of the Larix olgensis plantation in Mengjiagang Forest Farm in Jiamusi City,Heilongjiang Province as the experimental object,the RANSAC algorithm and radius outlier denoising algorithm were used to remove ground points and non-trunk and non-ground points in the data.Finally,the single tree recognition was carried out through the multi-layer K-means algorithm.The results showed that the improved multi-layer K-means algorithm had a single tree recognition accuracy rate of 91.01% and the false calculation amount of the trees was 0.Compared with the traditional K-means algorithm,the convergence time of the algorithm was shortened by 48.13%.It can be concluded that the multi-layer K-means algorithm is more efficient,and single tree identification in complex and dense forest plots has better results.The cost of surveying forest structure is reduced,which is of great significance to the calculation of forest structure parameters,the protection of forest resources and the overall planning.

Key words: forest structure, point cloud data, K-means algorithm, single tree recognition, Light Detection And Ranging

中图分类号:

S758

谷志新, 裴方睿. 基于多层K-means在森林点云中的单木识别算法[J]. 林业资源管理, 2022, 0(1): 124-131.

GU Zhixin, PEI Fangrui. Single Tree Recognition Algorithm Based on Multi-Layer K-means in Forest Point Cloud[J]. FOREST RESOURCES WANAGEMENT, 2022, 0(1): 124-131.

图/表 15

图1

表1

图2

图3

图4

图5

图6

表2

表3

图7

图8

图9

表4

图10

表5

参考文献 23

[1]	Paniagua-Ramirez A, Krupinska O, Jagdeo V, et al. Carbon storage estimation in a secondary tropical forest at CIEE Sustainability Center,Monteverde,Costa Rica[J/OL]. Scientific Report, 2021,11:23464.(2021-12-06)[2021-12-10]. https://doi.org/10.1038/s41598-021-03004-5.
[2]	李云红, 邵英男, 陈瑶, 等. 云冷杉红松林和蒙古栎红松林对土壤活性有机碳含量的影响[J]. 森林工程, 2021,37(3):60-66.
[3]	王长青, 邢万里, 邢艳秋, 等. 基于双尺度体元覆盖密度的TLS点云数据单木识别算法[J]. 森林工程, 2018,34(5):63-71.
[4]	宫大鹏, 李炳怡, 刘晓东. 草原火烧严重度燃烧指数的适用性比较分析[J]. 生态学报, 2018,38(7):8.
[5]	饶月明, 王川, 黄华国. 联合多源遥感数据监测四川木里县森林火灾[J]. 遥感学报, 2020,24(5):559-570.
[6]	徐婷. 基于LiDAR和OLI多光谱数据的森林生物量估算方法及应用[D]. 南京:南京林业大学, 2015.
[7]	郭慧, 宋文龙, 宋佳音, 等. 彩色森林冠层半球图像的分割方法[J]. 西部林业科学, 2015,159(4):146-149.
[8]	刘浩然, 范伟伟, 徐永胜, 等. 基于多源数据协同作业的森林信息提取研究进展[J]. 世界林业研究, 2020,33(1):33-37.
[9]	李杨先, 张慧春, 杨旸. 一种基于图像处理技术的植物形态表型参数获取方法[J]. 林业工程学报, 2020,5(6):128-136.
[10]	Montaghi A, Corona P, Dalponte M, et al. Airborne laser scanning of forest resources:An overview of research in Italy as a commentary case study[J]. International Journal of Applied Earth Observation and Geoinformation, 2013,23(1):288-300. doi: 10.1016/j.jag.2012.10.002
[11]	花伟成, 田佳榕, 孙心雨, 等. 基于TLS数据的杨树削度方程建立及材积估算[J]. 南京林业大学学报:自然科学版, 2021,45(4):41-48.
[12]	卢军, 刘宪钊, 孟维亮, 李红军. 基于地面激光点云数据的单木三维重建方法[J]. 南京林业大学学报:自然科学版, 2021,45(6):193-199.
[13]	Williams J, Schönlieb C B, Swinfield T, et al. 3D Segmentation of Trees Through a Flexible Multiclass Graph Cut Algorithm[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020,58(2):754-776. doi: 10.1109/TGRS.36
[14]	St-Onge B A, Achaichia N. Measuring forest canopy height using a combination of lidar and aerial photography data[J]. International Archives of Photogrammetry and Remote Sensing, 2001,XXXIV(3/W4):131-137.
[15]	刘鲁霞, 庞勇, 李增元. 基于地基激光雷达的亚热带森林单木胸径与树高提取[J]. 林业科学, 2016,52(2):26-37.
[16]	霍朗宁, 张晓丽. 基于机载LiDAR点云多层聚类的单木信息提取及其精度评价[J]. 林业科学, 2021,57(1):85-94.
[17]	杜意鸿, 尹田, 周雪梅, 等. 倾斜摄影测量技术提取油松单木信息[J]. 北京林业大学学报, 2021,43(4):77-86.
[18]	Li Wenkai, Guo Qinghua, Jakubowski M K, et al. A new method for segmenting individual trees from the lidar point cloud[J]. Photogrammetric Engineering and Remote Sensing, 2012,78(1):75-84. doi: 10.14358/PERS.78.1.75
[19]	孙拱, 辛颖, 闫灿. 基于改进的K-means算法估测单木树高[J]. 森林工程, 2019,35(6):31-35.
[20]	Ghahremani M, Williams K, Corke F, et al. Direct and accurate feature extraction from 3D point clouds of plants using RANSAC[J]. Computers and Electronics in Agriculture, 2021,187(13):106240. doi: 10.1016/j.compag.2021.106240
[21]	丁少鹏, 刘如飞, 蔡永宁, 等. 一种顾及地形的点云自适应坡度滤波方法[J]. 遥感信息, 2019,164(4):108-113.
[22]	徐姗姗. 图匹配算法激光扫描点云树干分割[J]. 中国图象图形学报, 2021,26(5):1095-1104.
[23]	李增元, 刘清旺, 庞勇. 激光雷达森林参数反演研究进展[J]. 遥感学报, 2016,20(5):1138-1150.

参数名称	数值
绝对精度	3cm
相对精度	5cm
扫描距离	100m
水平视场角	360°
垂直视场角	180°(-90°~90°)

层数/层	聚类后包含的点/个
2	5362
3	7512
4	7730
5	8925
6	8620
7	7759
8	6623

层数/层	识别率/%	层数/层	识别率/%
2	86.37	6	91.01
3	89.99	7	91.37
4	91.01	8	89.68
5	91.01

单木识别方法	单木总数/株	识别率/%
多层K-means	649	91.01
单层K-means	649	86.37