人机协同巡检下绝缘子分类及故障检测方法  

Insulator classification and fault detection method under human-machine collaborative inspection

作　　者：王少飞 吴琼水[1] 田猛[1] 王先培[1] WANG Shaofei;WU Qiongshui;TIAN Meng;WANG Xianpei(Electronic Information School,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学电子信息学院,湖北武汉430072

出　　处：《华中科技大学学报（自然科学版）》2024年第1期91-98,111,共9页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(52177109);湖北省重点研发计划资助项目(2020BAB109)。

摘　　要：针对目前电网人机协同巡检过程中数据收集量、来源面和复杂度增加,导致绝缘子目标小、遮挡严重、背景复杂、故障难以识别等问题,提出一种基于Yolov5-TBT模型的绝缘子分类及故障检测算法.首先构建了多目标(3种类别+2种故障)的人机协同巡检数据集,为模型训练提供充足的绝缘子图像,提高模型的鲁棒性和泛化能力.然后,以Yolov5模型为基础,增加Transformer编码器,减少复杂背景的干扰;使用加权双向特征金字塔网络(BiFPN)学习不同层级输入特征的重要性,进行更有效的特征聚合;增加小目标预测分支,提高对小目标的检测效果.实验结果表明:改进模型效果优于Yolov5模型,对伞裙破损故障的检测精度提高了19.4%,整体性能提高3.6%.A Yolov5-TBT model based insulator classification and fault detection algorithm was proposed to address the issues of small insulator targets,severe occlusion,complex background,and difficulty in identifying faults caused by the increase in data collection,source area,and complexity in the current human-machine collaborative inspection process in the power grid.Firstly,a multi-objective(3 categories+2 faults)human-machine collaborative inspection dataset was constructed to provide sufficient insulator images for model training and improve the robustness and generalization ability of the model.Then,based on the Yolov5 model,the Transformer encoder block was added to reduce the interference of complex background.The weighted Bi-directional Feature Pyramid Network(BiFPN)was used to learn the importance of input features at different levels for more effective feature aggregation.A tiny object prediction branch was added to improve the detection effect of the tiny target.The experimental results showed that the improved model is better than Yolov5,the recognition accuracy of the shed damage is 19.4%higher,and the overall performance is improved by 3.6%.

关 键 词：电网巡检 绝缘子分类 故障检测 Yolov5模型 Transformer编码器 

分 类 号：TM769[电气工程—电力系统及自动化]