Please wait a minute...
img

官方微信

遥感技术与应用  2021, Vol. 36 Issue (6): 1299-1305    DOI: 10.11873/j.issn.1004-0323.2021.6.1299
LiDAR专栏     
基于样本加权PointNet++的输电通道点云分类研究
陈正宇1(),彭淑雯2,朱号东1,张春涛1,习晓环2()
1.中国能源建设集团江苏省电力设计院有限公司,江苏 南京 211102
2.中国科学院空天信息创新研究院 数字地球重点实验室,北京 100094
LiDAR Point Cloud Classification of Transmission Corridor based on Sample Weighted-PointNet++
Zhengyu Chen1(),Shuwen Peng2,Haodong Zhu1,Chuntao Zhang1,Xiaohuan Xi2()
1.China Energy Engineering Group Jiangsu Power Design Institute Co. ,Ltd. ,Nanjing,211102,China
2.Key Laboratory of Digital Earth,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China
 全文: PDF(2169 KB)   HTML
摘要:

输电通道内地物要素复杂,机载LiDAR获取的电力线、杆塔、植被等地物点云密度差异大、空间分布不规则,实际应用中“所见即所得”的应用需求对点云的高效自动化分类带来挑战。将深度学习中的PointNet++算法用于输电通道机载点云自动分类研究,分析样本加权对不同密度点云数据分类精度的影响,利用两组实验数据验证算法的精度和效率,并与随机森林分类算法进行比较。结果表明:基于样本加权PointNet++的方法在输电通道点云自动化分类方面适用性更强,平均F1值87.14%,且分类精度和效率均优于随机森林方法。

关键词: 机载LiDAR输电通道点云分类PointNet++样本加权    
Abstract:

Due to the irregular spatial distribution, various density of point cloud data and the complexity of power scenarios, the application requirements of "what you see is what you get" in practical applications and higher requirements are put forward for automatic point cloud classification. In this paper, PointNet++ algorithm of deep learning is applied to the classification of airborne LiDAR point cloud in transmission corridor, and the end-to-end automatic point cloud classification is achieved. At the same time, the effect of sample weighting on classification accuracy is analyzed. Two test datasets are used to verify the accuracy and efficiency of the proposed algorithm, and compared with the results from random forest algorithm. The experimental results show that the algorithm based on sample weighted-PointNet++ is suitable for transmission corridor point cloud classification and reaches 87.14% on the macro average F score. Moreover, the classification performance and time-consuming are better than that of random forest.

Key words: Airborne LiDAR    Transmission corridor    Point cloud classification    PointNet++    Sample weighted
收稿日期: 2021-07-24 出版日期: 2022-01-26
ZTFLH:  TM75  
基金资助: 中国能源建设集团科技项目(CEEC2020-KJ05)
通讯作者: 习晓环     E-mail: chenzhengyu@jspdi.com.cn;xixh@aircas.ac.cn
作者简介: 陈正宇(1982-),男,江苏南京人,高级工程师,主要从事多源空间数据在电力工程中的应用研究。 E?mail: chenzhengyu@jspdi.com.cn
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  
陈正宇
彭淑雯
朱号东
张春涛
习晓环

引用本文:

陈正宇,彭淑雯,朱号东,张春涛,习晓环. 基于样本加权PointNet++的输电通道点云分类研究[J]. 遥感技术与应用, 2021, 36(6): 1299-1305.

Zhengyu Chen,Shuwen Peng,Haodong Zhu,Chuntao Zhang,Xiaohuan Xi. LiDAR Point Cloud Classification of Transmission Corridor based on Sample Weighted-PointNet++. Remote Sensing Technology and Application, 2021, 36(6): 1299-1305.

链接本文:

http://www.rsta.ac.cn/CN/10.11873/j.issn.1004-0323.2021.6.1299        http://www.rsta.ac.cn/CN/Y2021/V36/I6/1299

图1  PointNet++点云分类网络架构[10]
训练集测试集1测试集2
地面28 862 24362.42%726 39120.19%2 332 92039.36%
植被16 669 54936.05%2 814 94878.24%3 542 65459.76%
杆塔135 4270.29%6 2960.18%10 9190.18%
电力线228 6310.49%11 4140.32%28 5530.48%
建筑341 2120.74%38 6191.07%12 6890.21%
总和46 237 062100%3 597 668100%5 927 735100%
表1  数据集各类别数目统计
地面植被杆塔电力线建筑pre/%
测试集1测试集2测试集1测试集2测试集1测试集2测试集1测试集2测试集1测试集2测试集1测试集2
地面508 4631 975 150148 603584 49217710030497477.3377.14
植被212 595354 3282 663 9172 954 056241338006837492.6089.27
杆塔09904855005 58310 3064864860085.1883.91
电力线0640129527410 92828 0670097.3798.81
建筑5 3332 3881 9433 605000038 24711 34184.0265.43
rec/%70.085.6694.6383.3988.6894.3995.7498.3099.0489.38--
F1/%73.4880.7293.6186.2386.8988.8496.5598.5590.9175.55--
表2  样本加权PointNet++分类性能评价
图2  Test1样本加权前后结果对比
图3  两种算法在两个测试集的分类结果
图4  两种算法分类结果对比
算法测试集评估指标
P/%R/%F/%训练耗时 /h分类耗时 /s
样本加权PointNet++187.3089.6288.291.5109
282.9190.0285.98164
随机森林178.0887.1675.229.65 769
274.5385.9876.7611 282
表3  样本加权PointNet++和随机森林算法的效果对比
1 Du Yuefei, Liu Zhengjun, Feng Tianwen. Application of PPP technology and Airborne LiDAR in power line inspection[J]. Remote Sensing Technology and Application, 2019, 34(2): 263-268.
1 杜跃飞, 刘正军, 冯天文. PPP技术与机载激光雷达在电力线巡检中的应用研究[J]. 遥感技术与应用, 2019, 34(2): 263-268.
2 Guo Tao, Shen Ping, Shi Lei, et al. Study on power tower extraction and fast positioning from Airborne LiDAR data[J]. Remote Sensing Technology and Application, 2018, 33(3): 530-535.
2 虢韬, 沈平, 时磊, 等. 机载LiDAR快速定位高压电塔方法研究[J]. 遥感技术与应用, 2018, 33(3): 530-535.
3 Li W., Luo Z., Xiao Z., et al. A GCN-based method for extracting power lines and pylons from Airborne LiDAR data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022,60:1-14. DOI: 10.1109/TGRS.2021.3076107.
doi: 10.1109/TGRS.2021.3076107
4 Tan J, Zhao H, Yang R, et al. An entropy-weighting method for efficient power-line feature evaluation and extraction from LiDAR point clouds[J]. Remote Sensing,2021,13(17):3446. DOI: 10.3390/rs13173446.
doi: 10.3390/rs13173446
5 Wang Pinghua, Xi Xiaohuan, Wang Cheng, et al. Study on power line fast extraction based on Airborne LiDAR data[J]. Science of Surveying and Mapping, 2017, 42(2): 154-158.
5 王平华, 习晓环, 王成, 等. 机载激光雷达数据中电力线的快速提取[J]. 测绘科学, 2017, 42(2): 154-158.
6 Chen S, Wang C, Dai H, et al. Power pylon reconstruction based on abstract template structures using Airborne LiDAR data[J]. Remote Sensing,2019,11(13):1579. DOI:10.3390/rs11131579.
doi: 10.3390/rs11131579
7 Zhang Jixian, Lin Xiangguo, Liang Xinlian. Research progress and prospect of point cloud information extraction[J]. Acta Geodaetica et Cartographica Sinica, 2017, 46(10):262-271.
7 张继贤, 林祥国, 梁欣廉. 点云信息提取研究进展和展望[J]. 测绘学报, 2017, 46(10): 262-271.
8 Su H, Maji S, Kalogerakis E, et al. Multi-view convolutional neural networks for 3D shape recognition[C]∥ International Conference on Computer Vision,Santiago,Chile,13-16.
8 Dec., 2015: 945-953.
9 Qi C, Su H, Matthias N, et al. Volumetric and multi-view CNNS for object classification on 3D data[C]∥ IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 27-30.
9 June, 2016: 5648-5656.
10 Qi C, Li Y, Su H, et al. Pointnet++: Deep hierarchical feature learning on point sets in a metric space[C]∥ Neural Information Processing Systems, Long Beach, USA, 3-9.
10 Dec., 2017: 5099-5108.
11 Li Y, Bu R, Sun M, et al. PointCNN: convolution on x-transformed points[C]∥Neural Information Processing Systems, Montreal, Canada, 3-8
11 Dec., 2018: 820-830.
12 Hackel T, Savinov N, Ladicky L,et al. Semantic3D.net: a new large-scale point cloud classification benchmark[J]. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, 2017, IV-1/W1: 91-98.
13 Huang Y, Du Y, Shi W. Fast and accurate power line corridor survey using spatial line clustering of point cloud[J]. Remote Sensing,2021,13(8):1571. DOI:10.3390/rs13081571.
doi: 10.3390/rs13081571
14 Kamousi P, Lazard S, Maheshwari A, et al. Analysis of farthest point sampling for approximating geodesics in a graph[J]. Computational Geometry,2016,57:1-7. DOI:10.1016/j.comge 0.2016.05.005.
doi: 10.1016/j.comge 0.2016.05.005
15 Kim H B, Sohn G. Point-based classification of power line corridor scene using random forests[J]. Photogrammetric Engineering & Remote Sensing, 2013, 79(9):821-833.
16 Peng S, Xi X,Wang C,et al. Systematic comparison of power corridor classification methods from ALS point clouds[J]. Remote Sensing,2019,11(17):1961. DOI:10.3390/rs11171961.
doi: 10.3390/rs11171961
[1] 常潇月,荆林海,林沂. 基于机载LiDAR数据和Landsat 8影像的芬兰北部泰加林—苔原过渡带提取[J]. 遥感技术与应用, 2021, 36(6): 1259-1271.
[2] 杜伟,刘洋,杨国柱,王和平,李致东,李俊磊,习晓环. 输电通道机载激光点云多级配准研究[J]. 遥感技术与应用, 2021, 36(6): 1294-1298.
[3] 赵勋,岳彩荣,李春干,谷雷,张国飞. 基于机载LiDAR点云数据森林郁闭度估测[J]. 遥感技术与应用, 2020, 35(5): 1136-1145.
[4] 董保根,马洪超,车森,解龙根,何乔. LiDAR点云支持下地物精细分类的实现方法[J]. 遥感技术与应用, 2016, 31(1): 165-169.