摘要
针对传统电力监控存在效率低下、防控不全面和安全风险高等问题,设计了一种基于广义交并比的电力梯子自动化监测方法。首先,从大量监控视频中抽帧处理并制作一组电力梯子检测数据集,用于训练和测试算法。然后,在保证检测精度的前提下,使用MobileNetv2的主干网络替代YOLOv3的主干网络,适度地降低计算冗余度。最后,利用广义交并比损失提升检测精度。结果表明,与现有同类网络相比,提出的检测网络在电力场景梯子检测数据集上的检测精度为93.1%,具有更高的检测精度和速度,适合实际工程应用。
Aiming at the problems of low efficiency,incomplete prevention and control,and high safety risks in traditional power monitoring,a power ladder automation monitoring method based on the generalized intersection over union was proposed.Firstly,a set of power ladder detection datasets was extracted and processed from a large number of surveillance videos for training and testing algorithms;then,while ensuring detection accuracy,MobileNetv2’s backbone network was used to replace YOLOv3’s backbone network,moderately reducing computational redundancy;finally,the generalized intersection and union ratio loss was employed to improve detection accuracy.The results show that compared with existing similar networks,the proposed detection network has the detection accuracy of 93.1%on the power scene ladder detection dataset,which has higher detection accuracy and speed and is suitable for practical engineering applications.
作者
朱建宝
桑顺
俞鑫春
马青山
张斌
Zhu Jianbao;Sang Shun;Yu Xinchun;Ma Qingshan;Zhang Bin(Nantong Power Supply Branch,State Grid Jiangsu Electric Power Co.,Ltd.,Nantong Jiangsu 226006,China;Key Laboratory of Control of Power Transmission and Conversion(Ministry of Education),Shanghai Jiao Tong University,Shanghai 200240,China;Nantong Aowei Information System Engineering Co.,Ltd.,Nantong Jiangsu 226007,China)
出处
《电气自动化》
2024年第4期87-89,共3页
Electrical Automation
基金
电力传输功率变换控制教育部重点实验室开放课题(2021AC03)
国网江苏省电力有限公司科技项目(J2020054)。
关键词
目标检测
电力梯子
深度学习
特征提取
特征融合
object detection
power ladders
deep learning
feature extraction
feature fusion
作者简介
朱建宝(1973-),男,江苏人,高级工程师,从事电力系统安全监察与管理工作。
