绝缘子作为机械支撑和电气绝缘部件，是输电线路中不可或缺的组成部分。由于其常年暴露在户外，饱受风雨侵蚀和人为损害，对电力系统的正常运行造成了严重影响。因此，对绝缘子进行自爆检测是十分必要的。随着深度学习理论的日益成熟，将深度学习应用于绝缘子自爆检测成为当今的研究热点。基于此，本课题对绝缘子自爆检测进行研究，主要包括以下三个方面：

（1）针对航拍绝缘子图像对比度低、含有噪声的问题，研究绝缘子图像预处理算法。首先，通过非下采样Contourlet变换（Nonsubsampled contourlet transform，NSCT）将图像分解成低频和高频两部分。其次针对传统Retinex算法中高斯滤波器分配滤波权重时只注重像素间的距离而将滤波图像自身内容忽略的问题，本课题使用引导滤波对Retinex的照度分量估计以获取更多的边缘细节信息，并基于灰狼算法（Grey Wolf Optimizer，GWO）对引导滤波的正则化因子进行自适应提取；对高频子图像采用阈值去噪，对大于阈值的高频系数进行放大处理，对低于阈值的高频系数进行抑制。最后，经NSCT反变换，重构出最终的绝缘子图像。与单尺度Retinex、多尺度Retinex、引导滤波三种算法相比，经所提算法处理后的图像有效提高绝缘子图像对比度和去除噪声。

（2）针对绝缘子样本不足导致自爆检测精度低的问题，提出一种基于混合样本与迁移学习的Faster-RCNN绝缘子自爆检测算法。首先，通过模板匹配和Grabcut算法对图像中绝缘子与背景进行分割，并利用像素融合实现分割后的绝缘子图像与新背景图像的融合，生成绝缘子人工样本。其次，搭建Faster-RCNN检测网络。然后，利用ResNet50代替VGG16作为Faster-RCNN的特征提取网络，提取绝缘子高层语义特征。最后，以Faster-RCNN为基础，通过迁移学习方式对适量人工样本与真实样本组成的混合样本进行训练，并使用真实绝缘子测试集进行测试。在样本不足情况下，将所提算法与Backbone分别为GoogleNet、DenseNet169的Faster-RCNN进行对比，mAP@0.5平均提高2.35%，耗时平均减少0.85s。

（3）绝缘子受目标遮挡、复杂环境的影响，导致传统算法难以准确检测。针对此问题，提出一种基于注意力机制和自适应融合的YOLOv5绝缘子故障检测算法。在YOLOv5主干网络中引入CA注意力模块提高目标区域显著度；在其颈部网络中，用BiFPN结构使多尺度特征有效融合，并构建自适应加权特征融合模块使目标特征更好地表达出来；在其预测部分，改进损失函数提高了网络检测精度。引入深度可分离卷积提高了网络检测速度。所提算法较SSD、Faster-RCNN、YOLOv5三种算法相比，mAP@0.5平均提高4.03%，召回率平均提高2.8%，耗时平均减少1.05s。

As mechanical support and electrical insulation components, insulators are an integral part of power transmission lines. Due to their constant exposure to the outdoors, they are subject to weathering and human damage, which has a serious impact on the normal operation of power systems. Self-explosion detection of insulators is therefore essential. With the increasing maturity of deep learning theory, the application of deep learning to insulator self-detonation detection has become a hot research topic today. Based on this, this topic investigates insulator self-explosion detection, which mainly includes the following three aspects.

(1)The insulator image pre-processing algorithm is studied for the problem of low contrast and noise in aerial insulator images. Firstly, the image is decomposed into two parts, low frequency and high frequency, by NSCT(Nonsubsampled contourlet transform). Secondly, aiming at the problem that Gaussian filters in the traditional Retinex algorithm only pay attention to the distance between pixels and ignore the content of the filtered image itself, this project uses guided filtering to estimate the illuminance component of Retinex to obtain more edge detail information, and adaptively extracts the regularization factor of guided filtering based on the Grey Wolf Optimizer (GWO); for the high-frequency sub image, threshold denoising is used to amplify the high-frequency coefficients which are greater than the threshold and suppress the high-frequency coefficients which are less than the threshold. Finally, the final insulator image is reconstructed by NSCT inverse transformation. By comparing with the Single Scale Retinex , Multi-Scale Retinex and Guided Filtering algorithms, both the Entropy and SF indexes are improved, proving that the images processed by this the proposed algorithm can effectively improve the contrast of insulator images and remove random noise.

(2)Due to insufficient insulator samples,a Faster-RCNN insulator self-explosion detection algorithm based on hybrid samples and migration learning is proposed to address the problem of low accuracy.Firstly, the insulator and background in the image are segmented by template matching and Grabcut algorithm, and pixel fusion is used to achieve the fusion of the segmented insulator image with the new background image to generate insulator artificial samples. Next, the Faster-RCNN detection network is built. Then, ResNet50 is used instead of VGG16 as the feature extraction network of Faster-RCNN to extract the insulator high-level semantic features. Finally, the insulator data obtained by mixing an appropriate amount of artificial samples with real samples is trained by parameter migration based on Faster-RCNN and using the real insulator test set. Under the same training environment the proposed algorithm was compared with Backbone's Faster-RCNN for GoogleNet and DenseNet169, respectively. mAP@0.5 was increased by 2.35% on average, and the time was reduced by 0.85s on average.

(3)Insulators are affected by target occlusion and complex environments, making it difficult for traditional algorithms to detect them accurately. To address this problem, a YOLOv5 insulator fault detection algorithm based on attention mechanism and cross-scale adaptive fusion is proposed. The CA attention module is introduced in the YOLOv5 backbone network to improve the target region saliency; in its neck network, the BiFPN structure is used to enable effective fusion of multi-scale features, and an adaptive weighted feature fusion module is constructed to make the target features better expressed; in its prediction part, the loss function is improved to modify the network detection accuracy. The introduction of depth-separable convolution improves the network detection speed. Compared with SSD, Faster-RCNN and YOLOv5, the proposed algorithm improves the mAP@0.5 by an average of 4.03%, the recall rate increases by 2.8% on average, and the time consumption is reduced by an average of 1.05s.

[1]樊宇琦,丁涛,孙瑜歌,等.国内外促进可再生能源消纳的电力现货市场发展综述与思考[J].中国电机工程学报,2021,41(05):1729-1752.

[2]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.

[3]Tang Z, Abera G, Kishore M, et al. An applicable time reversal analysis of guided waves in al-plate with ultrasonic immersion testing[J]. Journal-Korean Physical Society, 2019, 74(4):340-348.

[4]Colwell J E, Hansen, et al. Investigation of diurnal variability of water vapor in Enceladus'plume by the Cassini ultraviolet imaging spectrograph[J]. Geophysical Research Letters,2017,44(2):672-677.

[5]Zhao S N, SongY, Zhao Y F. Infrared target detection method based on the receptive field and lateral inhibition of human visual system[J].Applied Optics, 2017,56(30):8555-8563.

[6]Chen J, Zhu H, Xia W, et al. Self-mixing birefringent dual-frequency laser Doppler velocimeter[J]. Optics Express, 2017, 25(2):560-572.

[7]王道累,张世恒,袁斌霞,等.基于改进YOLOv5的轻量化玻璃绝缘子自爆缺陷检测研究[J/OL].高电压技术:1-9[2023-03-10].DOI:10.13336/j.1003-6520.hve.20221542.

[8]苟军年,杜愫愫,刘力.基于改进掩膜区域卷积神经网络的输电线路绝缘子自爆检测[J].电工技术学报,2023,38(01):47-59.

[9]苟军年,杜愫愫,王世铎,等.轻量化特征融合的CenterNet输电线路绝缘子自爆检测[J/OL].北京航空航天大学学报:1-13[2023-03-10].DOI:10.13700/j.bh.1001-5965.2022.0602.

[10]贾晓芬,于业齐,郭永存,等.航拍绝缘子自爆缺陷的轻量化检测方法[J].高电压技术,2023,49(1):294-300.

[11]彭向阳,易琳,钱金菊,等.大型无人直升机电力线路巡检系统实用化[J].高电压技术，2020, 46(2):384-396.

[12]Wu Y, Zhao G, Hu J, et al. Overhead transmission line parameter reconstruction for UAV inspection based on tunneling magnetoresistive sensors and inverse models[J]. IEEE Transactions on Power Delivery, 2019, 34(3):819-827.

[13]Haiyan C, Yongjie Z, Rui C, etal. Self-Shattering Defect Detection of Glass Insulators Based on Spatial Features[J].Energies,2019, 12(3):543-556

[14]廖圣龙,安居白.输电线路对绝缘子破损航拍图像检测仿真[J].计算机仿真, 2016, 33(04):176-179+291.

[15]REDDY M J B, CHANDRA B K, MOHANTA D K．A DOST based approach for the condition monitoring of 11 kV distribution line insulators[J]. IEEE Transactions on Dielectrics ＆ Electrical Insulation, 2011, 18(2): 588-595．

[16]方挺,韩家明.航拍图像中绝缘子缺陷的检测与定位[J].计算机科学,2016,43(S1):222-225.

[17]Zhao Z, Liu N, Wang L. Localization of multiple insulators by orientation angle detection and binary shape prior knowledge[J]. IEEE Transactions on Dielectrics &Electrical Insulation, 2016, 22(6): 3421-3428.

[18]张晶晶,韩军,赵亚博,等. 形状感知的绝缘子识别与缺陷诊断[J].中国图象图形学报, 2014, 19(8):1194-1201.

[19]Zhai Y, Chen R, Yang Q, et al. Insulator ault detection based on spatial morphological features of aerial images[J]. IEEE Access, 2018, 6: 35316-35326.

[20]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.

[21]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(6): 844-849.

[22]陈芳,姚建刚,李佐胜,等.绝缘子串红外图像中单个绝缘子盘面的提取方法[J].电网技术.2010,34(5):220-224.

[23]赵永强,饶元,董世鹏,等.深度学习目标检测方法综述[J].中国图象图形学报,2020,25(4):629-654.

[24]段仲静,李少波,胡建军,等.深度学习目标检测方法及其主流框架综述[J].激光与光电子学进展,2020,57(12):59-74.

[25]GIRSHICK R, DONAHUE J, DARRELL T, et al. Richfeature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Columbus, OH, USA, 2014: 580-587.

[26]GIRSHICK R. Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV).Santiago, Chile, 2015: 1440-1448.

[27]侯春萍,章衡光,张巍,等.输电线路绝缘子自爆缺陷识别方法[J].电力系统及其自动化学报,2019,31(6):1-6.

[28]刘思言,王博,高昆仑,等.基于R-FCN的航拍巡检图像目标检测方法[J].电力系统自动化,2019,43(13):162-168.

[29]何宁辉,王世杰,刘军福,等.基于深度学习的航拍图像绝缘子缺失检测方法研究[J].电力系统保护与控制,2021,49(12):132-140.

[30]白洁音,赵瑞,谷丰强,等．多目标检测和故障识别图像处理方法[J]．高电压技术，2019, 45(11): 3504-3511．

[31]Sadykova D, Pernebayeva D, Bagheri M, et al. IN-YOLO: Real-time detection of outdoor high voltage insulators using UAV imaging[J]. IEEE Transactions on Power Delivery, 2019, 35(3): 1599-1601.

[32]Hossain A K M M, Jin Y, Soh W S, et al. SSD: A robust RF location fingerprint addressing mobile devices' heterogeneity[J]. IEEE Transactions on Mobile Computing, 2011, 12(1): 65-77.

[33]Cai Y, Luan T, Gao H, et al. YOLOv4-5D: An effective and efficient object detector for autonomous driving[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70(14): 1-13.

[34]Li S, Li Y, Li Y, et al. YOLO-FIRI: Improved YOLOv5 for Infrared Image Object Detection[J]. IEEE Access, 2021, 9: 141861-141875.

[35]王旭红,李浩,樊绍胜,等.基于改进SSD的电力设备红外图像异常自动检测方法[J]. 电工技术学报, 2020, 35(增刊1): 302-310.

[36]律方成,牛雷雷,王胜辉,等.基于优化YOLOv4的主要电气设备智能检测及调参策略[J].电工技术学报,2021,36(22):4837-4848.

[37]王书坤,高林,伏德粟,等.改进的轻量型YOLOv5绝缘子缺陷检测算法研究[J]. 湖北民族大学学报(自然科学版)，2021，39(04): 456-461.

[38]Li Y, Ni M, Lu Y. Insulator defect detection for power grid based on light correction enhancement and YOLOv5 model[J]. Energy Reports, 2022, 8: 807-814.

[39]吴一全,殷骏,戴一冕.基于人工蜂群优化的NSCT域图像模糊集增强方法[J].华南理工大学学报(自然科学版),2015,43(1):59-65.

[40]吴一全,史骏鹏.基于多尺度Retinex的非下采样Contourlet域图像增强[J].光学学报,2015,35(3):87-96.

[41]龙文,蔡绍洪,焦建军,等.一种改进的灰狼优化算法[J].电子学报,2019,47(01):169-175.

[42]谭宇璇,樊绍胜.基于图像增强与深度学习的变电设备红外热像识别方法[J].中国电机工程学报,2021,41(23):7990-7998.

[43]秦钟,杨建国,王海默,等.基于Retinex理论的低照度下输电线路图像增强方法及应用[J].电力系统保护与控制,2021,49(3):150-157.

[44]袁文海,刘彪,王喆,赛涛,宋昆峰,夏鑫,汪沨,钟理鹏.基于引导滤波算法的绝缘子憎水性图像去噪研究[J].电力科学与技术学报,2021,36(03):135-140.

[45]Martinez-Sanchez A, Kochovski Z, Laugks U, et al. Template-free detection and classification of membrane-bound complexes in cryo-electron tomograms[J]. Nature methods, 2020, 17(2): 209-216.

[46]Subaran T L, Semiawan T, Syakrani N. Mask R-CNN and GrabCut Algorithm for an Image-based Calorie Estimation System[J]. Journal of Information Systems Engineering and Business Intelligence, 2022, 8(1): 1-10.

[47]GIRSHICK R. Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV).Santiago, Chile, 2015: 1440-1448.

[48]Ren Shaoqing,He Kaiming,Girshick Ross,et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.[J]. IEEE transactions on pattern analysis and machine intelligence,2017,39(6):1137-1149

[49]郝帅,马瑞泽,赵新生,等.基于超分辨深度残差网络的玻璃绝缘子自爆故障检测算法[J].高电压技术,2022,48(05):1817-1825.

[50]Aytar Y. Transfer learning for object category detection [D], University of Oxford, UK, 2014

[51]张倩,李明,王雪松,等. 一种面向多源领域的实例迁移学习[J].自动化学报, 2014, 40(6): 1176-1183.

[52]Tajbakhsh N, Shin J Y, Gurudu S R, et al. Convolutional neural networks for medical image analysis: full training or fifine tuning?.[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1299-1312

[53]Segev N, Harel M, Mannor S, et al. Learn on source, refifine on target: a model transfer learning framework with random forests.[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(9): 1811-1824

[54]Shi H, Yang N, Tang H, et al. ASGD: Stochastic Gradient Descent with Adaptive Batch Size for Every Parameter[J]. Mathematics, 2022, 10(6): 863-877.

[55]Kingma D P, Ba J. Adam: A method for stochastic optimization[J]. arXiv preprint arXiv:1412.6980, 2014.

[56]郝帅,张旭,马瑞泽,等.基于改进GoogLeNet的小电流接地系统故障选线方法[J].电网技术,2022,46(1):361-368.

[57]袁智勇,白浩,邵向潮,等.基于一维多尺度轻量级DenseNet的配电网高阻接地故障检测方法[J].南方电网技术,2022,16(6):33-43.

[58]赵文清,程幸福,赵振兵,翟永杰.注意力机制和Faster RCNN相结合的绝缘子识别[J].智能系统学报,2020,15(1):92-98.

[59]郝帅,马瑞泽,赵新生,等.基于卷积块注意模型的YOLOv3输电线路故障检测方法[J].电网技术,2021,45(8):2979-2987.

[60]Hou Q, Zhou D, Feng J. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 13713-13722.

[61]Zhang R, Zhu F, Liu J, et al. Depth-wise separable convolutions and multi-level pooling for an efficient spatial CNN-based steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2019, 50(15): 1138-1150.

[62]Woo S, Hwang S, Jang H D, et al. Gated bidirectional feature pyramid network for accurate one-shot detection[J]. machine vision and applications, 2019, 30(4): 543-555.

[63]侯志强,刘晓义,余旺盛，等.使用GIoU改进非极大值抑制的目标检测算法[J].电子学报,2021,49(4):696-705.

[64]He J, Erfani S, Ma X, et al. Alpha-IoU: A Family of Power Intersection over Union Losses for Bounding Box Regression[J]. Advances in Neural Information Processing Systems, 2021, 34: 20230-20242