双重注意力机制的电力走廊点云语义分割

doi:10.13474/j.cnki.11-2246.2025.0421

测绘通报 ›› 2025, Vol. 0 ›› Issue (4): 127-133.doi: 10.13474/j.cnki.11-2246.2025.0421

• 学术研究 • 上一篇

双重注意力机制的电力走廊点云语义分割

李建¹, 王健^1,2, 王雷³, 李敏⁴, 杨立克⁵, 赵艺龙⁵

1. 山东科技大学测绘与空间信息学院, 山东青岛 266590;
2. 青岛市北斗导航空间信息技术重点实验室, 山东青岛 266590;
3. 北京建筑大学测绘与城市空间信息学院, 北京 102612;
4. 国网山东省电力公司超高压公司, 山东济南 250118;
5. 洛阳市嵩县山金矿业有限公司运营管理部, 河南洛阳 471400

收稿日期:2024-12-25 发布日期:2025-04-28
通讯作者: 王健。E-mail:wangj@sdust.edu.cn
作者简介:李建(1990—),男,硕士生,主要从事三维激光扫描技术研究。E-mail:ballack1977@163.com
基金资助:
国家重点研发计划(2022YFB3903501);山东省自然科学基金(ZR2023MD027)

Dual attention for power corridor point cloud semantic segmentation

LI Jian¹, WANG Jian^1,2, WANG Lei³, LI Min⁴, YANG Like⁵, ZHAO Yilong⁵

1. College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao 266590, China;
2. Qingdao Key Laboratory of Beidou Navigation and Intelligent Spatial Information Technology Application, Qingdao 266590, China;
3. School of Geomatics and Urban Spatial Informatics, Beijing University of Civil Engineering and Architecture, Beijing 102612, China;
4. State Grid Shandong Electric Extrahigh Voltage Company, Jinan 250118, China;
5. Operation and Management Department of Shan jin Mining Co., Ltd., Songxian County, Luoyang 471400, China

Received:2024-12-25 Published:2025-04-28

摘要/Abstract

摘要： 电力走廊的点云场景具有独特性,采用深度学习方法进行语义分割时面临诸多挑战,场景中存在严重的数据倾斜问题。此外,在处理架空电力线和杆塔时,由于局部半径内点数量不足,模型难以提取足够的局部特征,从而降低了对相似对象的分割精度。为此,本文设计了一种基于双重注意力机制的两阶语义输电线路语义分割方法。首先,在数据精简阶段,利用电力传输设备与背景之间的高程差异,通过非深度学习方法有效去除大量背景点,加速训练过程并显著缓解数据倾斜问题。然后,提出兼顾全局特征与局部特征的双重注意力模型,提升了相似对象的区分度,且提高了点云分割的精度。经测试,本文的数据精简方法可去除63%以上的背景点,解决部分数据倾斜问题;提出的双重注意力网络对地线、导线和绝缘子的分割效果优于其他方法。

关键词: 点云, 电力走廊, 深度学习, 双重注意力机制, 语义分割

Abstract: The point cloud scene of power corridors is unique and faces a number of challenges when using deep learning methods for semantic segmentation.There is a serious data skew problem in the scene, with background objects such as ground and vegetation occupying the majority, which may lead to degradation of network performance. In addition, when dealing with point cloud data of overhead power lines and power towers, the model has difficulty in extracting enough local features due to the insufficient number of points within the local radius, which reduces the segmentation accuracy of similar objects. For this reason, this paper designs a two-order semantic transmission line semantic segmentation method based on the dual attention mechanism. Firstly, in the data streamlining stage, the elevation difference between the power transmission equipment and the background is utilized to have remove a large number of background points, thus reducing the training burden of the subsequent neural network, accelerating the training process, and significantly alleviating the data skew problem. Then, the dual-attention model, which takes into account both global and local features, is proposed to enhance the differentiation of similar objects and improves the accuracy of point cloud segmentation. After testing, the data streamlining method in this paper can remove more than 63% of background points and partially solve the data skewing problem; the proposed dual-attention network outperforms other methods for segmentation of ground wires, conductors and insulators.

Key words: point cloud, power corridor, deep learning, dual attention, semantic segmentation

中图分类号:

P208

李建, 王健, 王雷, 李敏, 杨立克, 赵艺龙. 双重注意力机制的电力走廊点云语义分割[J]. 测绘通报, 2025, 0(4): 127-133.

LI Jian, WANG Jian, WANG Lei, LI Min, YANG Like, ZHAO Yilong. Dual attention for power corridor point cloud semantic segmentation[J]. Bulletin of Surveying and Mapping, 2025, 0(4): 127-133.

参考文献

[1] 杨东晨.基于Transformer注意力机制的电力走廊激光点云分类方法[D].武汉: 武汉大学, 2023.
[2] 王斐然,韩庚,郭昕阳,等.激光雷达数据下架空输电线路点云场景分割及净空入侵检测[J].测绘通报,2024(5): 133-137.
[4] 田乐萌,冉海涛,付斌,等.D2000S无人机系统在输电线路新建工程中的应用[J].测绘通报,2023(S1):141-147.
[5] 张智前,叶周润,欧鑫,等.基于机载LiDAR系统的电力线点云提取方法研究[J].合肥工业大学学报(自然科学版),2023,46(8): 1103-1108.
[6] ELKHRACHY I.Feature extraction of laser scan data based on geometric properties[J].Journal of the Indian Society of Remote Sensing,2017,45(1): 1-10.
[7] GU Yanfeng,CAO Zhimin,DONG Limin.A hierarchical energy minimization method for building roof segmentation from airborne LiDAR data[J].Multimedia Tools and Applications,2017,76(3): 4197-4210.
[8] MIAO Yanlong,LI Shuai,WANG Liuyang,et al.A single plant segmentation method of maize point cloud based on Euclidean clustering and K-means clustering[J].Computers and Electronics in Agriculture,2023,210: 107951.
[9] KANG C L,WANG F,ZONG M M,et al.Research on improved region growing point cloud algorithm[J].ISPRS-International Archives of the Photogrammetry,Remote Sensing and Spatial Information Sciences,2020,4210: 153-157.
[10] ITAKURA K,MIYATANI S,HOSOI F.Estimating tree structural parameters via automatic tree segmentation from LiDAR point cloud data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2021,15: 555-564.
[11] PRAKHYA S M,LIU Bingbing,LIN Weisi,et al.B-SHOT: a binary 3D feature descriptor for fast keypoint matching on 3D point clouds[J].Autonomous Robots,2017,41(7): 1501-1520.
[12] MOHAMED M,MORSY S,EL-SHAZLY A.Improvement of 3D LiDAR point cloud classification of urban road environment based on random forest classifier[J].Geocarto International,2022,37(27): 15604-15626.
[13] HE Shuting,JIANG Xudong,JIANG Wei,et al.Prototype adaption and projection for few-and zero-shot 3D point cloud semantic segmentation[J].IEEE Transactions on Image Processing,2023,32: 3199-3211.
[14] WANG Guanjian,WANG Linong,WU Shaocheng,et al.Semantic segmentation of transmission corridor 3D point clouds based on CA-PointNet++[J].Electronics,2023,12(13): 2829.
[15] 王郑阳,俞昊,孟昊,等.POWERLINE-ALS: 一种用于输电线路场景深度学习语义分割的机载LiDAR点云数据集[J].山东科技大学学报(自然科学版),2023,42(2): 33-43.
[16] ZHANG Keying,CAI Ruirui,WU Xinqiao,et al.iBALR3D: imBalanced-aware long-range 3D semantic segmentation[C]//Proceedings of the 2nd AAAI Workshop on Artificial Intelligence with Biased or Scarce Data.[S.l.]:MDPI,2024: 6.
[17] CHARLES R Q,HAO Su,MO Kaichun,et al.PointNet: deep learning on point sets for 3D classification and segmentation[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition.Honolulu,HI:IEEE,2017: 77-85.
[18] QI C R,YI Li,SU Hao,et al.PointNet++[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems.Long Beach,California:ACM Press,2017: 5105-5114.
[19] 黄郑,顾徐,王红星,等.基于改进PointNet++的输电杆塔点云语义分割模型[J].中国电力,2023,56(3): 77-85.
[20] ZHANG Wuming,QI Jianbo,WAN Peng,et al.An easy-to-use airborne LiDAR data filtering method based on cloth simulation[J].Remote Sensing,2016,8(6): 501.
[21] DOSOVITSKIY A.An image is worth 16x16 words: Transformers for image recognition at scale[EB/OL].2020-10-22[2024-11-25].https://arxiv.org/abs/2010.11929.
[22] WOO S,LEE D,HWANG S,et al.MKConv: Multidimensional feature representation for point cloud analysis[J].Pattern Recognition,2023,143: 109800.
[23] PAUL S,PATTERSON Z,BOUGUILA N.DualMLP: a two-stream fusion model for 3D point cloud classification[J].The Visual Computer,2024,40(8): 5435-5449.
[24] WANG Yue,SUN Yongbin,LIU Ziwei,et al.Dynamic graph CNN for learning on point clouds[J].ACM Transactions on Graphics,2019,38(5): 1-12.
[25] GUO Menghao,LIU Zhengning,MU Taijiang,et al.Beyond self-attention: external attention using two linear layers for visual tasks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(5): 5436-5447.
[26] GUO Menghao,CAI Junxiong,LIU Zhengning,et al.PCT: Point cloud transformer[J].Computational Visual Media,2021,7(2): 187-199.
[27] LU Dening,GAO K,XIE Qian,et al.3DGTN: 3D dual-attention GLocal transformer network for point cloud classification and segmentation[J].IEEE Transactions on Geoscience and Remote Sensing,2024,62: 5704313.
[28] LIU Xinhai,HAN Zhizhong,LIU Yushen,et al.Point2Sequence[C]//Proceedings of the 33th AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence.[S.l.]:ACM Press,2019: 8778-8785.
[29] QIU Shi,ANWAR S,BARNES N.PnP-3D: a plug-and-play for 3D point clouds[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(1): 1312-1319.
[30] MA Xu,QIN Can,YOU Haoxuan,et al.Rethinking network design and local geometry in point cloud: a simple residual MLP framework[EB/OL].[2024-11-25].https://arxiv.org/abs/2202.07123v2.
[31] ZHAO Hengshuang,JIANG Li,JIA Jiaya,et al.Point transformer[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision.Montreal:IEEE,2021:16239-16248.
[32] WU Rusong,BAI Jing,LI Wenjing,et al.DCNet: exploring fine-grained vision classification for 3D point clouds[J].The Visual Computer,2024,40(2): 781-797.

[1]	符强, 钟振, 纪元法, 任风华. 基于动态场景的实时语义SLAM算法[J]. 测绘通报, 2025, 0(4): 27-33.
[2]	朱松, 陈安泰, 华远盛, 姜文宇, 元鹏鹏, 朱家松. 面向漂流胶囊机器人图像的管道缺陷检测[J]. 测绘通报, 2025, 0(4): 34-38.
[3]	马于博, 张爱国, 王浩宇, 刘帅琪, 靳镜宇, 沈占锋, 李均力. 基于U-ConvHDNet模型的戈壁砾幕层提取[J]. 测绘通报, 2025, 0(4): 68-74.
[4]	郝佳, 姚国英, 周剑, 王斯远, 肖进胜. 基于图像和点云融合的三维小目标检测方法[J]. 测绘通报, 2025, 0(3): 33-38.
[5]	张富杰, 王留召, 钟若飞, 许梦兵, 靳欢欢. 基于车载激光点云的路灯提取方法[J]. 测绘通报, 2025, 0(3): 46-51.
[6]	谢巴图, 胡佳睿, 潘俊. 基于双路全局信息优化网络的遥感影像海陆分割算法[J]. 测绘通报, 2025, 0(3): 52-58,86.
[7]	赵兴旺, 赵妍, 刘超, 刘春阳. 基于改进的DeepLabV3+网络的Sentinel-1影像水体提取[J]. 测绘通报, 2025, 0(3): 66-70.
[8]	马海志. 基于深度学习的隧道超高分辨率图像病害检测框架[J]. 测绘通报, 2025, 0(3): 105-110.
[9]	张璇, 杨本勇, 文武, 邓维熙, 周何帆. 改进的DeepLabV3+模型用于四川县域大小春作物识别[J]. 测绘通报, 2025, 0(3): 133-137,167.
[10]	苟长龙, 庞敏, 杨扬. 改进的U-Net卷积网络在遥感影像地物分类中的应用[J]. 测绘通报, 2025, 0(3): 150-155.
[11]	宋培焱, 叶勤, 曾亮, 罗俊奇, 张硕, 尹长江. 面向密度明显差异点云的室内场景三维Mesh模型POCO重建方法[J]. 测绘通报, 2025, 0(2): 7-12,47.
[12]	乌萌, 熊超. 利用激光点云测距影像的城市环境动态目标检测优化方法[J]. 测绘通报, 2025, 0(2): 35-40.
[13]	石佳俊, 臧玉府, 肖雄武, 张莹滢. 非刚性匹配的激光点云与多光谱影像深度融合[J]. 测绘通报, 2025, 0(2): 41-47.
[14]	马耀昌, 王进, 董秀军, 叶飞, 冯国正. 面向水旱灾害的激光点云综合精度评价[J]. 测绘通报, 2025, 0(2): 175-179,184.
[15]	王召泽, 韩若愚, 赵智伟, 路奇琛. 手持三维激光扫描仪在城市更新立面测量中的应用[J]. 测绘通报, 2025, 0(1): 133-137.

双重注意力机制的电力走廊点云语义分割

Dual attention for power corridor point cloud semantic segmentation

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价