基于机器学习的高压线数据库创建方法

张衡; 高艳辉; 陆洋

doi:10.13700/j.bh.1001-5965.2024.0311

基于机器学习的高压线数据库创建方法

doi: 10.13700/j.bh.1001-5965.2024.0311

张衡¹,
高艳辉^2, ,,
陆洋¹

1.
南京航空航天大学直升机动力学全国重点实验室，南京 210016
2.
南京航空航天大学无人机研究院，南京 210016

详细信息

通讯作者:
E-mail：gyh2005@nuaa.edu.cn

中图分类号: V275.1
计量
- 文章访问数: 150
- HTML全文浏览量: 98
- PDF下载量: 16
- 被引次数: 0
出版历程
- 收稿日期: 2024-05-11
- 录用日期: 2024-11-29
- 网络出版日期: 2024-12-24
- 整期出版日期: 2026-07-31

Power line database creating method based on machine learning

ZHANG Heng¹,
GAO Yanhui^{2
, ,},
LU Yang¹

1.
National Key Laboratory of Helicopter Aeromechanics，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，China
2.
Unmanned Aerial Vehicles Research Institute，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，China

More Information

Corresponding author: E-mail：gyh2005@nuaa.edu.cn

摘要

摘要:
直升机执行近地飞行任务时，时常发生撞击高压线事故。直升机近地告警系统(HTAWS)是直升机近地防撞的重要手段，但高压线数据库缺失导致其难以有效保障直升机飞行安全。针对该问题，提出一种基于机器学习利用卫星影像创建高压线数据库的方法。针对YOLOv5对卫星影像中小目标检测不敏感、漏检率较高的问题，设计了一种改进YOLOv5目标检测算法，用于识别卫星影像中的高压线塔，采用开源栅格空间数据转换库(GDAL)模块计算得到高压线塔的经纬度。提出一种基于高压线塔阴影获取塔高的方法。基于图神经网络与高压线塔的群组特征，研究了高压线塔连线预测。仿真结果表明：高压线塔的经纬度识别准确，塔高预测平均准确率为94.65%，高压线塔连线预测的马修斯相关系数(MCC)为0.479，所建立的高压线数据库满足直升机高压线防撞告警需求。
- 直升机 /
- 飞行安全 /
- 高压线 /
- 机器学习 /
- 卫星影像
Abstract:
During close-range flying tasks, helicopters are frequently struck electrical lines. Helicopter terrain awareness and warning system (HTAWS) is an essential instrument for preventing helicopter crashes, however it is challenging to reliably assure helicopter flight safety in the absence of a power line database. This research suggests a machine learning-based approach for building a power line database using satellite imagery. Aiming at the problem that YOLOv5 is insensitive to small target detection and has a high rate of missed detection, an improved YOLOv5 is proposed to identify the pylons in satellite images. Than the geospatial data abstraction library (GDAL) module was then used to calculate the longitude and latitude of the pylons. A pylons height acquisition method based on the shadow of pylons was proposed. The prediction of pylon connections was investigated using group features of pylons and graph neural networks. The simulation results demonstrate that the approach presented in this research can determine the longitude and latitude of pylons with 94.65% accuracy when determining the tower’s height. The Matthews correlation coefficient (MCC) value predicted for the connection of pylons is 0.479, which can establish a power line database that meets the requirements of helicopter power line collision warning.
- helicopter /
- flight safety /
- power line /
- machine learning /
- satellite imagery

HTML全文

图 1 建立高压线数据库技术流程

Figure 1. Technical flowchart for establishing database of power line

下载: 全尺寸图片幻灯片

图 2 ECA模块流程

Figure 2. Diagram of ECA module

下载: 全尺寸图片幻灯片

图 3 改进YOLOv5网络结构

Figure 3. Improved YOLOv5 network architecture

下载: 全尺寸图片幻灯片

图 4 卫星影像原图与阴影检测结果

Figure 4. Satellite image and shadow detection results

下载: 全尺寸图片幻灯片

图 5 阴影图像后处理结果

Figure 5. Shadow image post-processing results

下载: 全尺寸图片幻灯片

图 6 高压线塔阴影长度提取

Figure 6. Extraction of shadow lengths for pylons

下载: 全尺寸图片幻灯片

图 7 图自编码器模型

Figure 7. Graph autoencoder model

下载: 全尺寸图片幻灯片

图 8 高压线高程数据库可视化结果

Figure 8. Visualization results of power line elevation database

下载: 全尺寸图片幻灯片

图 9 直升机高压线防撞告警方法流程

Figure 9. Flowchart of helicopter power line collision alarm method

下载: 全尺寸图片幻灯片

图 10 可视化检测结果

Figure 10. Visual detection results

下载: 全尺寸图片幻灯片

图 11 卫星影像目标检测结果

Figure 11. Satellite image target detection results

下载: 全尺寸图片幻灯片

图 12 卫星影像阴影检测

Figure 12. Satellite image shadow detection

下载: 全尺寸图片幻灯片

图 13 高压线塔预测连线与实际连线的对比

Figure 13. Comparison between predicted and actual connection of pylons

下载: 全尺寸图片幻灯片

表 1 评价指标对比

Table 1. Comparison of evaluation indicators

算法	P	R	mAP
SSD	0.798	0.773	0.794
YOLOv5x	0.927	0.906	0.921
改进YOLOv5	0.952	0.927	0.942

下载: 导出CSV

表 2 高压线塔位置信息

Table 2. Pylons location information

高压线塔编号	x像素位置/像素	y像素位置/像素	经度/（°）	纬度/（°）
1	487	203	122.28854584	46.06790769
2	584	222	122.28958868	46.06776812
3	410	1202	122.28772831	46.06046811
4	1321	311	122.29749370	46.06710283
5	767	837	122.29155207	46.06319002
6	998	614	122.29403042	46.06484638
7	297	886	122.28651165	46.06282711
8	398	864	122.28759956	46.06298531

下载: 导出CSV

表 3 高压线塔仿真预测高度与实际高度对比

Table 3. Comparison between measured heights and actual heights of pylons

序号	经度/（°）	纬度/（°）	像素长度/像素	阴影长度/m	测量塔高/m	实际塔高/m	准确率/%
1	122.2911764	46.067198	39.0	31.2	52.0	60.0	86.7
2	122.2918388	46.0670749	45.2	36.2	60.3	61.0	98.9
3	122.2867993	46.0623141	44.6	35.7	59.5	57.0	95.7
4	122.2950974	46.0675656	58.6	46.9	78.2	72.5	92.7
5	122.2922180	46.0657677	63.5	50.8	84.7	83.5	98.6
6	122.2901552	46.0644212	60.2	48.2	80.3	76.5	95.3

下载: 导出CSV

表 4 图神经网络对MCC的影响

Table 4. Influence of graph neural networks on MCC

图神经网络	MCC
GCN	0.479
GAT	0.45
GraphSAGE	0.449

下载: 导出CSV

表 5 学习率对MCC的影响

Table 5. Impact of learning rate on MCC

学习率	MCC
0.1	0.372
0.01	0.398
0.007	0.396
0.005	0.393
0.003	0.445
0.001	0.479
0.0008	0.478

下载: 导出CSV

参考文献(21)

[1]	Iseler L, De Maio J. An analysis of U.S. civil rotorcraft accidents by cost and injury (1990-1996): NASA/TM-2002-2096/5[R]. Moffett Field: NASA Ames Research Center, 2002: 17-26.
[2]	Chandrasekaran R, Payan A P, Collins K B, et al. Helicopter wire strike protection and prevention devices: review, challenges, and recommendations[J]. Aerospace Science and Technology, 2020, 98: 105665.
[3]	Chan N. Cable-cutting device: US 4215833A[P]. 1980-08-05.
[4]	Harris C M, Achour G, Payan A P, et al. Use of machine learning to create a database of wires for helicopter wire strike prevention[C]//Proceedings of the AIAA Scitech 2021 Forum. Reston: AIAA, 2021: 0527.
[5]	Achour N, Peck E, Johnson R, et al. Rotorcraft wire strike prevention and protection: wire database creation, sensor package design, wire cutter modeling, data fusion, and helicopter operators’ inputs: DOT/FAA/AM-23/12[R]. Washington, D.C.: U.S. Federal Aviation Administration, 2023.
[6]	Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 779-788.
[7]	Wang Q L, Wu B G, Zhu P F, et al. ECA-net: efficient channel attention for deep convolutional neural networks[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 11531-11539.
[8]	Ju M R, Luo H B, Wang Z B, et al. The application of improved YOLO V3 in multi-scale target detection[J]. Applied Sciences, 2019, 9(18): 1-14.
[9]	刘同华, 景文博, 姜淑华. 一种基于遥感的快速提取建筑物高度的方法[J]. 科技创新导报, 2016, 13(31): 87-88. Liu T H, Jing W B, Jiang S H. A method for quickly extracting building height based on remote sensing[J]. Science and Technology Innovation Herald, 2016, 13(31): 87-88(in Chinese).
[10]	赵显富, 胡晓雯, 赵轩. 一种基于彩色模型的遥感影像阴影检测方法[J]. 测绘通报, 2014(5): 20-22. Zhao X F, Hu X W, Zhao X. A method of remote sensing images shadow detection based on color models[J]. Bulletin of Surveying and Mapping, 2014(5): 20-22(in Chinese).
[11]	李嘉良, 张超, 齐红甲. 遥感影像提取建筑物高度的方法[J]. 河北联合大学学报(自然科学版), 2013, 35(2): 121-125. Li J L, Zhang C, Qi H J. Analysis of remote sensing images to extract building height[J]. Journal of Hebei United University (Natural Sclence Edition), 2013, 35(2): 121-125(in Chinese).
[12]	李晶慈. 基于图自动编码器的链路预测方法研究[D]. 桂林: 广西师范大学, 2023. Li J C. Research on link prediction methods based on graph autoencoder[D]. Guilin: Guangxi Normal University, 2023(in Chinese).
[13]	Niepert M, Ahmed M, Kutzkov K. Learning convolutional neural networks for graphs[C]//Proceedings of the 33rd International Conference on International Conference on Machine Learning-Volume 48. New York: ACM, 2016: 2014-2023.
[14]	Thekumparampil K K, Wang C, Oh S, et al. Attention-based graph neural network for semi-supervised learning[EB/OL]. (2018-03-10)[2024-05-06]. https://arxiv.org/abs/1803.03735.
[15]	Hamilton W L, Ying R, Leskovec J. Inductive representation learning on large graphs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 1025-1035.
[16]	吴旭峰, 陆洋, 刘玉虎. 基于多传感器信息融合的直升机城市内障碍物主动探测方法[J]. 飞行力学, 2023, 41(2): 64-70. Wu X F, Lu Y, Liu Y H. Active detection method for helicopter urban flight based on multi-sensor information fusion[J]. Flight Dynamics, 2023, 41(2): 64-70(in Chinese).
[17]	周成中, 陆洋. 直升机前视地形告警算法研究[J]. 飞行力学, 2016, 34(3): 30-33. Zhou C Z, Lu Y. Research on forward looking terrain alerting algorithm for helicopter[J]. Flight Dynamics, 2016, 34(3): 30-33(in Chinese).
[18]	黄山笑. 直升机预测式防撞地告警算法研究[D]. 南京: 南京航空航天大学, 2021. Huang S X. Research on predictive ground collision avoidance warning algorithm for helicopter[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2021(in Chinese).
[19]	汪宋, 费树岷. SSD(Single Shot MultiBox Detector)目标检测算法的研究与改进[J]. 工业控制计算机, 2019, 32(4): 103-105. Wang S, Fei S M. Research and improvement of SSD (Single Shot MultiBox Detector) target detection algorithm[J]. Industrial Control Computer, 2019, 32(4): 103-105(in Chinese).
[20]	迟盛元, 白岩, 孟祥民. 基于YOLOv5的PCB缺陷检测算法研究[J]. 科学技术创新, 2024(1): 106-109. Chi S Y, Bai Y, Meng X M. Research on the PCB defect detection algorithm based on YOLOv5[J]. Scientific and Technological Innovation, 2024(1): 106-109(in Chinese).
[21]	李乐依. 基于遥感卫星光学影像的输电通道巡检技术研究[D]. 北京: 华北电力大学, 2022. Li L Y. Research on power transmission channel inspection technology based on remote sensing satellite optical image[D]. Beijing: North China Electric Power University, 2022(in Chinese).