高压输电线路由于长期工作在自然环境中，受雷击、鸟害等因素影响易发生故障，严重影响输电系统的稳定运行。目前，无人机作为重要的电力巡检方式得到广泛应用。然而，无人机巡检后产生海量巡检数据依然需要人眼判断，不仅效率低，而且给巡检人员带来巨大负担。因此，面向复杂巡检环境，本文开展基于深度学习的高压输电线路关键部件故障检测方法研究对于提高线路智能化巡检水平具有重要现实意义和理论价值。本文主要研究工作包括：

（1）针对复杂巡检环境中输电线路多类故障目标的多尺度检测问题，提出一种基于Transformer和感受野模块的YOLOv5输电线路多类故障目标检测算法。首先，搭建了YOLOv5网络，在主干网络引入Transformer模块，通过利用其多头注意力结构获取特征图像素点间的相关性和全局信息，增强故障目标在复杂背景中的显著度，进而提升模型检测精度；其次，在颈部网络中引入感受野模块提取故障目标不同尺度特征，同时利用空洞卷积增大网络感受野，增强颈部特征融合能力，提高模型对多尺度故障目标的分类精度；然后，为了提升边框回归精度，引入CIoU函数，进一步提高检测精度；最后，利用某电力巡检部门近3年的巡检数据对所提算法和7种经典检测算法进行对比实验。实验结果表明，相比于7种对比算法，所提出的算法检测精度最高并且具有较好的检测实时性，其平均检测精度可达95.6%，对于1280×720的巡检图像检测速度为125帧/秒。

（2）针对巡检图像中小目标故障检测精度低的问题，提出一种基于注意力和双向加权特征金字塔网络的YOLOv5输电线路小目标故障检测方法。首先，针对复杂背景中小目标故障特征提取困难导致检测精度低的问题，在主干特征提取网络中引入坐标注意力机制，通过捕获特征图的长距离依赖关系，从而更准确地获得感兴趣区域，增强主干网络的特征提取能力；然后，在Neck部分通过使用加权双向特征金字塔网络更好的平衡故障目标不同尺度的特征信息，提高网络对小目标故障的检测精度；最后，为了提升算法定位精度，引入SIoU损失函数，提高模型收敛速度。实验结果表明，所提出的算法能够较好地检测小目标故障，其平均检测精度为92.8%，对于1280×720的巡检图像算法检测速度可达100帧/秒。

High-voltage transmission lines are prone to faults due to long-term work in the natural environment and interference from lightning strikes, bird damage and other factors, which seriously affects the stable operation of the transmission system. Currently, drones are widely used as an important power inspection method. However, the huge amount of inspection data generated by drones after inspection still requires human judgment, which is not only inefficient but also places a huge burden on inspection personnel. Therefore, for the complex inspection environment, this paper carries out research on the fault detection method of key components of high-voltage transmission lines based on deep learning, which has important practical significance and theoretical value for improving the intelligent inspection level of transmission lines. The main research work in this paper includes:

Aiming at the problem of multi-scale detection of multi-fault targets of transmission lines in complex inspection environment, a multi-fault target detection algorithm of YOLOv5 transmission lines based on Transformer and receptive field block is proposed. Firstly, the YOLOv5 network is built, and the Transformer module is introduced into the backbone network. By using its multi-head attention structure, the correlation and global information between feature image pixels are obtained, which enhances the salience of fault targets in complex background, and then improves the model detection precision. Secondly, the receptive field block is introduced into the neck network to extract the different scale features of fault targets, and the receptive field of the network is enlarged by using dilated convolution to enhance the fusion ability of neck features and improve the classification precision of the model for multi-scale fault targets; Then, in order to improve the bounding regression precision, CIoU function is introduced to further improve the detection precision; Finally, the proposed algorithm is compared with seven classical detection algorithms by using the inspection data of a power inspection department in recent three years. The experimental results show that compared with seven comparison algorithms, the proposed algorithm has the highest detection precision and better real-time detection, with an average detection precision of 95.6% and a detection speed of 125 frames per second for 1280×720 inspection images.

(2) Aiming at the problem of low precision of small target fault detection in inspection images, a small target fault detection method for YOLOv5 transmission line based on attention and bidirectional weighted feature pyramid network is proposed. Firstly, in order to solve the problem of low detection precision caused by the difficulty of fault feature extraction of small targets in complex background, a coordinate attention mechanism is introduced into the backbone feature extraction network, and the region of interest is obtained more accurately by capturing the long-distance dependence of feature maps, thus enhancing the feature extraction ability of the backbone network; Then, in the Neck part, a bidirectional weighted feature pyramid network is used to better balance the feature information of different scales of fault targets, so as to improve the detection accuracy of small target faults. Finally, in order to enhance the positioning precision of the algorithm, SIoU loss function is introduced to improve the convergence speed of the model. The experimental results show that the proposed algorithm can detect small target faults well, with an average detection precision of 92.8%, and the detection speed of the algorithm can reach 100 frames per second for 1280×720 inspection images.

[1] 缪希仁, 刘志颖, 鄢齐晨. 无人机输电线路智能巡检技术综述[J]. 福州大学学报:自然科学版, 2020, 48(2): 198-209.

[2] 郝帅, 杨磊, 马旭等. 基于注意力机制与跨尺度特征融合的YOLOv5输电线路故障检测[J]. 中国电机工程学报, 2023, 43(06): 2319-2331.

[3] 赵振兵, 蒋志钢, 李延旭等. 输电线路部件视觉缺陷检测综述[J]. 中国图象图形学报, 2021, 26(11): 2545-2560.

[4] Foudeh H A, Luk P C K, Whidborne J F. An advanced unmanned aerial vehicle (UAV) approach via learning-based control for overhead power line monitoring: A comprehensive review[J]. IEEE Access, 2021, 9: 130410-130433.

[5] 邵瑰玮, 刘壮, 付晶等. 架空输电线路无人机巡检技术研究进展[J]. 高电压技术, 2020, 46(1): 14-22.

[6] Li X F, Liu H Y, Liu G H, et al. Transmission line pin defect detection based on deep learning [EB/OL]. Power System Technology, [2020-11-09].

[7] Yang L, Fan J, Liu Y, et al. A review on state-of-the-art power line inspection techniques[J]. IEEE Transactions on Instrumentation and Measurement, 2020, 69(12): 9350-9365.

[8] Zhen J, Qiang N, Liu L, et al. Discussion about recent applications of power big data in smart grid[C]//2019 IEEE PES Asia-Pacific Power and Energy Engineering Conference (APPEEC). IEEE, 2019: 1-5.

[9] Meng F, Xu B, Zhang T, et al. Application of AI in image recognition technology for power line inspection [EB/OL]. Energy Systems, 2021-02-04:

[10] Liu X, Miao X, Jiang H, et al. Data analysis in visual power line inspection: An in-depth review of deep learning for component detection and fault diagnosis[J]. Annual Reviews in Control, 2020, 50: 253-277.

[11] 和敬涵, 罗国敏, 程梦晓等. 新一代人工智能在电力系统故障分析及定位中的研究综述[J]. 中国电机工程学报, 2020, 40(17): 5506-5515.

[12] Fan L, Li J, Pan Y, et al. Research and application of smart grid early warning decision platform based on big data analysis[C]//2019 4th International Conference on Intelligent Green Building and Smart Grid (IGBSG). IEEE, 2019: 645-648.

[13] Zhai Y, Wang D, Zhang M, et al. Fault detection of insulator based on saliency and adaptive morphology[J]. Multimedia Tools and Applications, 2017, 76: 12051-12064.

[14] 戴玉静,吕东辉,郭松鸽.基于颜色和纹理特征的输电线路锈蚀区域检测[J].工业控制计算机,2018,31(09):39-40+43.

[15] 万迪明,张杰,郭祥富.一种基于视觉显著性分析的输电线路异物检测方法[J].电视技术,2018,42(01):106-110.

[16] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. nature, 2015, 521(7553): 436-444.

[17] 赵振兵, 张薇, 翟永杰等. 电力视觉技术的概念、研究现状与展望[J]. 电力科学与工程, 2020, 36(01): 1-8.

[18] 谢庆, 张煊宇, 王春鑫等. 新一代人工智能技术在输变电设备状态评估中的应用现状及展望[J]. 高压电器, 2022, 58(11): 1-16.

[19] 牛洪超, 胡晓兵, 罗耀俊. 基于M-YOLO的自动驾驶下目标识别算法[J]. 计算机工程与设计, 2022, 43(08): 2213-2220.

[20] Liu X, Miao X, Jiang H, et al. Data analysis in visual power line inspection: An in-depth review of deep learning for component detection and fault diagnosis[J]. Annual Reviews in Control, 2020, 50: 253-277.

[21] 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 CVPR). Columbus, OH, USA: IEEE, 2014: 580-587.

[22] He K, Zhang X, Ren S, et al. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,37(9):1904-1916.

[23] Girshick R. Fast R-CNN[C]//Proceedings of IEEE International Conference on Computer Vision. Washington D. C. USA: IEEE Press, 2015: 1440-1448．

[24] 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, 2017, 39(6): 1137-1149.

[25] Dai J, Li Y, He K, et al. R-fcn: Object detection via region-based fully convolutional networks[J]. Advances in neural information processing systems, 2016, 29.

[26] He K, Gkioxari G, Dollár P, et al. Mask r-cnn[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2961-2969.

[27] Lan M, Zhang Y, Zhang L, et al. Defect detection from UAV images based on region-based CNNs[C]//2018 IEEE International Conference on Data Mining Workshops (ICDMW). IEEE, 2018: 385-390.

[28] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv:1409.1556, 2014.

[29] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 770-778.

[30] 顾超越, 李喆, 史晋涛, 等. 基于改进Faster-RCNN的无人机巡检架空线路销钉缺陷检测[J]. 高电压技术, 2020, 46(09): 3089-3096.

[31] 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.

[32] 张烨, 高玉菡, 黄新波等. 基于Fine Mask RCNN的110～220kV输电铁塔涉鸟故障识别与评估[J]. 电网技术, 2022, 46(06): 2132-2144.

[33] Chen W, Li Y, Li C. A visual detection method for foreign objects in power lines based on mask R-CNN[J]. International Journal of Ambient Computing and Intelligence (IJACI), 2020, 11(1): 34-47.

[34] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]//Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, Springer International Publishing, 2016: 21-37.

[35] Fu C Y, Liu W, Ranga A, et al. Dssd: Deconvolutional single shot detector [EB/OL]. arXiv:1701.06659, 2017-01-23:

[36] 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, IEEE, 2016: 779-788.

[37] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.

[38] Redmon J, Farhadi A. Yolov3: An incremental improvement [EB/OL]. arXiv:1804.02767v1, 2018-04-08:

[39] Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection [EB/OL]. arXiv:2004.10934v1, 2020-04-23:

[40] Wu T H, Wang T W, Liu Y Q. Real-time vehicle and distance detection based on improved yolo v5 network[C]//2021 3rd World Symposium on Artificial Intelligence (WSAI). IEEE, 2021: 24-28.

[41] Ge Z, Liu S, Wang F, et al. Yolox: Exceeding yolo series in 2021[J]. arXiv preprint arXiv:2107.08430, 2021.

[42] Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J]. arXiv preprint arXiv:2207.02696, 2022.

[43] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.

[44] 李瑞生, 张彦龙, 翟登辉, 等. 基于改进SSD的输电线路销钉缺陷检测[J].高电压技术, 2021, 47(11): 3795-3802.

[45] 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.

[46] 唐翔翔, 沈薇, 朱明, 等. 基于改进YOLOv4的输电线路异物检测算法[J]. 安徽大学学报(自然科学版), 2021, 45(05): 58-63.

[47] 黄悦华, 陈照源, 陈庆等. 基于边缘计算和改进YOLOv5s算法的输电线路故障实时检测方法[J]. 电力建设, 2023, 44(01): 91-99.

[48] Han G, Li T, Li Q, et al. Improved Algorithm for Insulator and Its Defect Detection Based on YOLOX. Sensors. 2022; 22(16):6186.

[49] 宋智伟, 黄新波, 纪超等. 基于Flexible YOLOv7的输电线路绝缘子缺陷检测和故障预警方法[EB/OL]. 高电压技术, 2023-04-15:

[50] 王健, 王凯, 刘刚等. 基于生成对抗网络和RetinaNet的销钉缺陷识别[J]. 华南理工大学学报(自然科学版), 2020, 48(02): 1-8.

[51] Liu J, Jia R, Li W, et al. High precision detection algorithm based on improved RetinaNet for defect recognition of transmission lines[J]. Energy Reports, 2020, 6: 2430-2440.

[52] Krizhevsky A, Sutskever I, Hinton G. ImageNet classification with deep convolutional neural networks[C]. Advances in Neural Information Processing Systems (ANIPS), Curran Associates Inc, USA, 2012:1097-1105.

[53] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 770-778.

[54] Zou Z, Chen K, Shi Z, et al. Object detection in 20 years: A survey[J]. Proceedings of the IEEE, 2023. 111(3): 257-276

[55] Uijlings J R R, Van De Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International journal of computer vision, 2013, 104: 154-171.

[56] Liu. S, Qi L, Qin H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 8759-8768.

[57] Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection [EB/OL]. https://doi.org/10.48550/arXiv.1612.03144, 2017-04-19:

[58] 祝星馗, 蒋球伟. 基于CNN与Transformer的无人机图像目标检测研究[J]. 武汉理工大学学报(信息与管理工程版), 2022, 44(02): 323-331.

[59] 郑伟, 赵金芳, 张奕婧等. 基于感受野扩增和注意力机制的U-Net脑肿瘤MR图像分割[J].河北大学学报(自然科学版), 2022, 42(05): 542-551.

[60] Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: Faster and better learning for bounding box regression [EB/OL]. https://doi.org/10.1609/aaai.v34i07.6999, 2020-04-03

[61] Powers D M W. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation [EB/OL]. arXiv:2010.16061v1, 2020-10-11:

[62] Hou Q, Zhou D, Feng J. Coordinate attention for efficient mobile network design[EB/OL] https://doi.org/10.48550/arXiv.2103.02907, 2021-03-04:

[63] Tan M, Pang R, Le Q V. Efficientdet: Scalable and efficient object detection[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 10781-10790.

[64] Gevorgyan Z. SIoU loss: More powerful learning for bounding box regression[EB/OL]. https://doi.org/10.48550/arXiv.2205.12740, 2022-05-25:

[65] Hu J, Shen L, Sun G. Squeeze-and-excitation networks[J]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.

[66] Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module[J]. arXiv: 1807.06521v2, 2018.

[67] Park J, Woo S, Lee J Y, et al. BAM: Bottleneck attention module[J]. arXiv preprint arXiv: 1807.06514v2, 2018.