摘要
针对配网勘灾中人工勘灾效率低下和机巡勘灾需后端分析导致灾情信息反馈不及时的问题,立足于前端实时智能检测模式,提出了基于改进YOLO-ResNet混合神经网络的配网杆塔倾倒实时检测模型。首先,改进传统YOLO-V3的损失函数,利用广义交并比(generalized intersection over union,GIoU)计算目标检测框损失,有效提升杆塔主体检测的准确性。其次,采用ResNet-50定位杆塔端点和中心线,提出一种杆塔姿态判断方法以快速计算杆塔倾斜角度。最后,研发了一种便携式设备并部署了所提模型,以实地采集的数据对模型和设备进行测试,结果表明该设备对杆塔姿态判断的整体准确率达93.48%,设备平均功耗9 W,可用于前端实时智能分析、汇总杆塔受灾情况,验证了模型和设备的有效性。
Aiming at the low efficiency of manual disaster exploration,and the delay information feedback of back-end analysis using unmanned aerial vehicle in distribution network disaster exploration,an instantaneous detection model is proposed for the collapse of distribution network poles based on improved YOLO-ResNet hybrid neural network.Firstly,generalized intersection over union(GIoU)is introduced to improve the traditional YOLO-V3 algorithm to effectively enhance the accuracy of detecting the main body of poles.Then,ResNet-50 algorithm is used to locate endpoints and center line of poles,and a pole attitude judgment method is proposed to quickly calculate the pole tilt angle.Finally,a portable device for the model mentioned is developed,and the model and device are tested with field collected data.The results show that the overall accuracy of pole pose judgment based on the proposed model is 93.48%,the average power consumption of the portable device is 9 W,which is capable to intelligently analyze and summary the damaging status of the poles on the front end,and the effectiveness of the model and device is verified.
作者
张宝星
莫一夫
潘岐深
谢锐彪
ZHANG Baoxing;MO Yifu;PAN Qishen;XIE Ruibiao(Guangdong Power Grid Co.,Ltd.,Guangzhou 510623,China;School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)
出处
《南方电网技术》
CSCD
北大核心
2022年第8期133-141,共9页
Southern Power System Technology
基金
广东电网有限责任公司科技项目(GDKJXM20184286)。
关键词
杆塔检测
姿态判断
配网勘灾
便携式设备
广义交并比
pole detection
pose judgment
distribution network disaster exploration
portable device
generalized intersection over union
