摘要
高压配电网所处环境复杂,多架设在地势严峻、环境复杂的区域,极易附着异物。异物附着在架空线路上不仅存在短路的风险,还会影响供电效率,甚至危及配电网附近居民的安全,若不及时发现并清理将会对其平稳运营造成极大危害。针对上述问题,提出了一种基于YOLOv3改进的视觉显著性分析异物远程定位方法。该方法基于YOLOv3目标检测网络改进,首先通过分析无人机检查过程中拍摄到的视频及图片,并基于人眼感知特性计算视觉显著图像,划分图像中高压配电网的候选区域;然后通过基于颜色、形状等特性对异物进行判别并进行定位;最后使用RepVGG对模型进行优化,并通过增加网络深度和尺度来提升异物检测的精确度。实验结果表明,相比传统异物检测方法,本文提出的模型准确度提升了11%,显著地提升了异物远程定位的准确度和效率。
The environment in which high-voltage distribution networks are located is complex,and they are often installed in areas with severe terrain and complex environments,making it easy for foreign objects to attach.Foreign matters attached to overhead lines not only have the risk of short circuits,but also affect the efficiency of power sup⁃ply,and even endanger the safety of residents near the distribution network.If they are not found and cleared in time,it will cause great harm to its smooth operation.To solve this problem,this paper proposes an improved visual saliency analysis method for foreign object remote locations based on YOLOv3.This method is based on the improve⁃ment of YOLOv3 target detection network.Firstly,by analyzing the videos and pictures taken during the inspection of UAV,and calculating the visually significant image based on the human eye perception characteristics,the candidate areas of the high-voltage distribution network in the image were divided.Then the foreign objects were identified and located based on the characteristics of color,shape and so on.Finally,RepVGG was used to optimize the model,and the accuracy of foreign object detection was improved by increasing the network depth and scale.The results of the ex⁃periment show that the accuracy of the proposed model is improved by 11%compared with the traditional method,significantly improving the accuracy and efficiency of remote localization of foreign objects.
作者
程航
韩旭
林青照
周云婷
CHENG Hang;HAN Xu;LIN Qing-zhao;ZHOU Yun-ting(State Grid Fuzhou Power Supply Company,Fuzhou Fujian 350001,China;Nanchang University,Nanchang Jiangxi 330031,China)
出处
《计算机仿真》
北大核心
2023年第12期139-144,共6页
Computer Simulation
基金
国网福建省电力有限公司科技项目(5213102000FB)。
关键词
异物远程定位
高压配电网
计算机视觉
Foreign body remote location
High voltage distribution network
Computer vision
