绝缘子通常用于支撑高压输电线路、电缆终端、开关设备等，以保证电力系统的安全运行。现有绝缘子巡检工作通常采用人工方式，存在效率低下、培训成本高、安全隐患大等问题。针对以上问题，本文利用智能化无人机巡线采集绝缘子图像，研究了基于深度学习的绝缘子故障识别方法，主要研究工作如下：

（1）针对绝缘子目标存在遮挡、背景复杂等干扰，导致现有目标检测算法无法充分获取绝缘子信息的问题，提出了一种融合注意力机制的目标检测算法。算法以YOLOv5为基础，引入Swin-Transformer Block模块改进Neck网络，采取加权box融合算法代替传统的非极大值抑制算法筛选预测结果，保留更多被遮挡目标的特征信息，实验结果表明，改进模型在绝缘子检测任务中，与基准网络相比检测精度提高了5.4%。

（2）针对传统无人机多传感器融合成本高昂、算法复杂等问题，提出了一种基于视觉引导的无人机轨迹规划算法。依托目标检测算法结果对绝缘子进行坐标标定，在改进的单目视觉测距算法补充深度信息的基础上，应用双向RRT（Rapidly-exploring Random Tree，快速随机扩展树）算法规划出无人机飞抵工作位置的飞行轨迹。

（3）针对绝缘子故障数据集类别样本不均衡的问题，提出了一种级联ResNet34故障分类模型。首先，采集绝缘子故障数据集并进行了筛选与数据强化，根据绝缘子故障特点建立了分层数据集，提出了先识别绝缘子外形缺失，再细化识别故障外貌特点的技术思想；其次，根据数据集的分布构筑了级联分类网络，训练得到适用于绝缘子故障识别场景的模型；最后，在故障测试数据集上通过实验验证，级联故障分类模型的平均识别精度较基准的ResNet34分类模型提升了4.2%。

为验证上述方法的有效性，搭建了四旋翼无人机巡线仿真平台并进行验证，实现算法统一部署、统一测试，测试表明改进的算法运行稳定，能够实现多旋翼无人机的自主引导和故障识别任务。

Insulators are usually used to support high-voltage transmission lines, cable terminals, switchgear, etc. to ensure the safe operation of the power system. The existing insulator inspection work is usually done manually, which has problems such as low efficiency, high training cost, and high safety risks. To address the above problems, this thesis uses intelligent UAV inspection to collect insulator images and investigates the insulator fault identification method based on deep learning, with the following main research works:

(1) A target detection algorithm incorporating attention mechanism is proposed for the problem that insulator targets have interference such as occlusion and complex background, which causes the existing target detection algorithm to fail to obtain insulator information adequately. The algorithm is based on YOLOv5, introduces the Swin-Transformer Block module to improve the Neck network, adopts the weighted box fusion algorithm instead of the traditional non-maximum suppression algorithm to filter the prediction results, and retains more feature information of the occluded targets. 5.4%.

(2) A vision-guided UAV trajectory planning algorithm based on vision guidance is proposed to address the problems of high cost and complex algorithms of traditional UAV multi-sensor fusion. Based on the results of the target detection algorithm, the insulator is calibrated, and the UAV flight trajectory to the working position is planned by applying the bi-directional RRT (Rapidly-exploring Random Tree) algorithm based on the improved monocular visual ranging algorithm to supplement the depth information.

(3) A cascaded ResNet34 fault classification model is proposed to address the problem of unbalanced class samples in insulator fault dataset. First, the insulator fault dataset was collected and filtered and data enhanced, and a hierarchical dataset was established according to the insulator fault characteristics, and the technical idea of first identifying the insulator shape deficiency and then refining the identification of the fault appearance characteristics was proposed; second, a cascade classification network was constructed according to the distribution of the dataset, and a model applicable to the insulator fault identification scenario was trained; finally, the cascade fault classification was tested on the fault test dataset by Finally, the average recognition accuracy of the cascaded fault classification model is improved by 4.2% compared with the benchmark ResNet34 classification model on the fault test dataset by experimental verification.

In order to verify the effectiveness of the above method, a quadrotor UAV patrol simulation platform was built and validated to realize the unified deployment and testing of the algorithm, and the test showed that the improved algorithm operated stably and was able to realize the autonomous guidance and fault identification tasks of the multi-rotor UAV.

[1] 李岳, 周路遥, 李特, 邵先军, 初金良, 高磊. 500kV线路瓷绝缘子雷击断串故障分析[J]. 浙江电力, 2020, 39(04):17-21.

[2] 徐刚. 架空输配电线路绝缘子闪络故障分析[J]. 光源与照明, 2022(09):78-80.

[3] 雷海洋, 周玉龙. 输电线路绝缘子故障分析与检测方法综述[J].山东工业技术, 2018(08):167.

[4] 贾培亮. 架空输电线路绝缘子掉串故障分析[J]. 现代工业经济和信息化, 2017, 7(21):87-88.

[5] 谢秋伟, 邓淋升, 宁仕元, 汤浩, 吴小虎. 多旋翼无人机在输电线路质量验收中的应用研究[J]. 应用能源技术, 2022(07):26-29.

[6] Zhao M, Li Y, Hao G, et al. Multi-Sensor Information Fusion Predictive Control Algorithm[J]. International Journal of Multimedia and Ubiquitous Engineering, 2016, 11(3): 169-178.

[7] 杨璐雅, 黄新波, 张烨. 基于边缘检测的瓷质绝缘子裂缝特征检测方法[J]. 广东电力, 2018, 31(07):106-111

[8] 朱文天, 胡立坤, 王帅军. 改进色差法的绝缘子图像分割方法[J]. 广西大学学报(自然科学版), 2019, 44(05):1276-1283.

[9] 陈文浩, 姚利娜, 李丰哲. 无人机电网巡检中的绝缘子缺陷检测与定位[J]. 计算机应用, 2019, 39(S1):210-214.

[10] Tan P, Li X, Xu J, et al. Catenary insulator defect detection based on contour features and gray similarity matching[J]. Journal of Zhejiang University-SCIENCE A, 2020, 21(1): 64-73.

[11] 吕泽卿, 付兴建, 张世卓. 基于改进GA-KSW熵法的无人机拍摄绝缘子故障识别[J]. 机床与液压, 2022, 50(8):188-192.

[12] Girshick R, Donahue J, Darrell T, et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014: 580-587.

[13] Larrauri J, I,Dorrosal G, Gonzalez M. Automatic system for overhead power line inspection using an Unmanned Aerial Vehicle—RELIFO project[C]// In: Unmanned Aircraft Systems， 2013:244-252.

[14] He H, Luo D, Lee W J, et al. A contactless insulator contamination levels detecting method based on infrared images features and RBFNN[J]. IEEE Transactions on Industry Applications, 2018, 55(3):2455-2463.

[15] Wang Y, Wang J, Gao F, et al. Detection and recognition for fault insulator based on deep learning[C]// 2018 11th International congress on image and signal processing, BioMedical Engineering and Informatics (CISP-BMEI). IEEE, 2018:1-6.

[16] 周自强, 赵淳, 范鹏. 基于多尺度特征融合Faster R-CNN的绝缘子自爆缺陷研究[J]. 水电能源科学, 2020, 38(11):187-189+44.

[17] 杨罡, 张娜, 晋涛, 等. 基于YOLO算法的无人机航拍图片绝缘子缺失检测[J]．科技资讯, 2018, 16(24):30~31.

[18] Zhai Y, Wang D, Zhang M, et al. Fault detection of insulator based on saliency and adaptive morphology[J]. Multimedia Tools and Applications, 2016, 76(9):1~14

[19] Hui H A, Xh A, Ys A, et al. An insulator self-blast detection method based on YOLOv4 with aerial images[J]. Energy Reports, 2022, 8:448-454.

[20] Krizhevsky A, Sutskever I, Hinton G E. Imagenet Classification with Deep Convolutional Neural Networks[J]. Communications of the ACM, 2017, 60(6):84-90.

[21] Girshick R. Fast R-CNN[C]// Proceedings of the IEEE International Conference on Computer Vision. 2015:1440-1448.

[22] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6):1137-1149.

[23] HR K M, GKIOXARI G, DOLLAR P, et al. Mask RCNN[C]// 2017 IEEE International Conference on Computer Vision(ICCV), October 22-29, 2017, Venice, Italy. New York： IEEE, 2017:2980-2988.

[24] Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-time Object Detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016:779-788.

[25] Liu W, Anguelov D, Erhan D, et al. SSD: Single Shot Multibox Detector[C]// European Conference on Computer Vision. Springer, Cham, 2016:21-37.

[26] Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017:7263-7271.

[27] Redmon J, Farhadi A. YOLOv3: An Incremental Improvement[J]. arXiv preprint arXiv: 1804. 02767, 2018.

[28] Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4: Optimal Speed and Accuracy of Object Detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[29] Shen Z, Shi H, Feris R, Cao L, Yan S, Liu D, Wang X, Xue X, and Huang T. S. (2017b). Learning object detectors from scratch with gated recurrent feature pyramids. arXiv preprint arXiv:1712.00886.

[30] Law H, Deng J. Cornernet: Detecting Objects as Paired Keypoints[C]// Proceedings of the European Conference on Computer Vision (ECCV). 2018:734-750.

[31] Ge Z, Liu S, Wang F, et al. YOLOX: Exceeding YOLO Series in 2021[J]. arXiv preprint arXiv:2107.08430, 2021.

[32] 沈昕格, 金海, 郭亮. 基于反步法的四旋翼无人机自适应控制研究[J]. 电子科技, 2022, 35(03):32-37.

[33] Morrison D, Corke P, Leitner J. Multi-View Picking: Next-best-view Reaching for Improved Grasping in Clutter[C]// International Conference on Robotics and Automation. 0.

[34] 刘昌辉, 帅考, 杨维荣. 嵌入式视觉的测距系统设计[J]. 武汉工程大学学报, 2015, 37(4):65-68.

[35] 曹毓, 冯莹, 赵立双, 等. 相机姿态安装误差对单目视觉定位精度的影响[J]. 传感器与微系统, 2012, 31(12):23-26.

[36] A Gariépy, Ruel J C, Chaib-Draa B, et al. GQ-STN: Optimizing One-Shot Grasp Detection based on Robustness Classifier[J]. IEEE, 2019.

[37] Yu J, Weng K, Liang G, et al. A vision-based robotic grasping system using deep learning for 3D object recognition and pose estimation[C]// IEEE International Conference on Robotics & Biomimetics. IEEE, 2014.

[38] 赵文一. 无人机视觉辅助自主降落系统研究[D]. 哈尔滨工业大学, 2018.

[39] 单一. 基于视觉导航的四旋翼无人机自主着降控制研究[D]. 南京航空航天大学, 2018.

[40] 夏令志, 程登峰, 秦金飞, 李森林, 季坤. 一起220kV架空输电线路双伞型瓷绝缘子炸裂故障分析[J]. 电气技术, 2017(09):72-74+80.

[41] 王志惠, 路蔚旭, 何喜梅, 陈文强, 刘高飞, 张烁. 750kV GIS设备支撑绝缘子故障分析[J]. 青海电力, 2016, 35(03):56-60.

[42] 胡洲, 王志胜, 甄子洋, 等. 多传感器系统的延长周期全信息融合滤波[J]. 传感器与微系统, 2010, 29(05):37-40.

[43] 管叙军, 王新龙. 视觉导航技术发展综述[J]. 航空兵器, 2018(5):3-8.

[44] 杨忠炯, 邹绍昕, 周立强. 基于视觉引导的采样系统设计[J]. 制造业自动化, 2023, 45(01):212-215.

[45] 高静辉. 输电线路绝缘子故障分析与检测方法综述[J]. 山东工业技术, 2016(15):165-166.

[46] 熊志豪, 邓涛, 郑宇惟, 周鑫, 吴玉泽. 基于状态观测器的四旋翼无人机轨迹跟踪控制[J]. 电光与控制, 2022,29(06):50-56.

[47] 王俊平, 徐刚. 电力多旋翼无人机巡检控制系统的设计与实现[J]. 机械设计与制造, 2022(10):75-80.

[48] 王志胜, 甄子洋, 等. 基于联合滤波器的高精度卫星定姿系统设计[J]. 中国空间科学技术, 2018(3).

[49] 万树平. 基于信噪比的多传感器数据融合方法[J]. 传感技术学报, 2018(01):182-185.

[50] 张慧莹. 输电线路绝缘子识别与故障状态检测技术研究[D]. 西安工程大学,2018

[51] 杨庆江, 牛姿懿, 冯新宇, 等. 绝缘子陶瓷裂缝识别的 Sobel 改进算法[J]. 黑龙江科技大学学报, 2019, 29(02):190-195.

[52] 袁田, 应斯, 向孟宇, 等. 玻璃绝缘子裂缝的激光检测方法[J]. 电网技术, 2020, 44(08):3156-3163.

[53] 张晓文. 基于卷积神经网络绝缘子外部损伤识别[D]. 内蒙古农业大学, 2020.

[54] 商俊平, 李储欣, 陈亮. 基于视觉的绝缘子定位与自爆缺陷检测[J]. 电子测量与仪器学报, 2017, 31(06):844-849.

[55] Dan Z, Hong H, Qian R, et al. An insulator defect detection algorithm based on computer vision[C]// 2017 IEEE International Conference on Information and Automation (ICIA). IEEE, 2017.

[56] Zhai Y, Chen R, Yang Q, et al. Insulator Fault Detection Based on Spatial Morphological Features of Aerial Images[J]. IEEE Access, 2018, PP:1-1.

[57] 刘轩驿. 无人机图像中绝缘子的识别和故障诊断[D]. 华北电力大学,2018.

[58] 荀子扬. 基于无人机视觉的绝缘子缺陷识别检测研究[D].东北农业大学,2018.

[59] 陈庆, 闫斌, 叶润, 周小佳. 航拍绝缘子卷积神经网络检测及自爆识别研究[J]. 电子测量与仪器学报, 2017, 31(06):942-953.

[60] Feng G, Jiao W, Kong Z, et al. Recognition of insulator explosion based on deep learning[C]// 2017 14th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP). 2017.

[61] 林志成, 缪希仁, 江灏, 陈静. 多模型融合的输电线路绝缘子自爆故障检测[J]. 福州大学学报(自然科学版), 2020, 48(02):217-223.

[62] 冯万兴, 范鹏, 姚翔宇, 谷山强, 周自强, 周盛. 基于深度学习的输电线路绝缘子缺陷识别[J]. 水电能源科学, 2021, 39(01):176-178+50.

[63] 刘锡朋, 杨海川, 石义官, 王尧, 李作武. 基于视觉引导的弹部件高效自动排序技术[J]. 兵工自动化, 2022, 41(11):24-26+31.

[64] 刘建春, 高永康, 陈勇忠. 双目视觉引导的异构件定位抓取系统[J]. 机械设计与制造, 2022(11):285-288+295.

[65] Yang Cheng, Kong Lei, Zhang Zhichao, Tao Ye, Chen Xiaoyu. Exploring the Visual Guidance of Motor Imagery in Sustainable Brain–Computer Interfaces[J]. Sustainability, 2022, 14(21).

[66] Cai Jiaying, Liu Kai. Application of Visual Guidance Design of Tourist City Combined with Deep Learning Computer Visual Symbols[J]. Journal of Function Spaces, 2022, 2022.

[67] Silva R A, Neme J H Z, Margraf E, et al. Analisis of a distance measurement system based on monocular camera[C]// XXXVIII Ibero Latin American Congress on Computational Methods in Engineering. Florianópolis, Santa Catarina, Brazil: CMEF, 2017:312-326.

[68] 杨雪姣.自动驾驶场景下的实时单目测距研究[D].重庆理工大学, 2020.

[69] 葛新锋, 孙书情, 秦涛, 李瑞华. 视觉引导的机器人条烟分拣系统设计与实现[J]. 包装工程, 2022, 43(07):238-243.

[70] 王连庆, 钱莉. 基于视觉引导的工业机器人无序抓取系统设计[J]. 制造业自动化, 2022, 44(03):86-89+196.

[71] 谢秋伟, 邓淋升, 宁仕元, 汤浩, 吴小虎. 多旋翼无人机在输电线路质量验收中的应用研究[J]. 应用能源技术, 2022(07):26-29.

[72] 高伟, 周宸, 郭谋发. 基于改进YOLOv4及SR-GAN的绝缘子缺陷辨识研究[J]. 电机与控制学报, 2021, 25(11):93-104.