螺丝、销钉等小目标元件对输电线路起着固定、支撑、保护等作用。由于其长期暴露在复杂环境中，受恶劣自然条件、机械张力及材料老化等因素影响，易导致其出现松动、缺失等问题，对输电线路安全稳定运行造成严重影响。因此，面向复杂巡检环境，本文探索一种融合先验知识和注意力机制的输电线路小目标故障检测方法。

论文主要工作包括：

（1）针对航拍图像中小目标故障特征表达能力不足且受复杂背景影响严重，导致其难以准确检测的问题，提出一种动态特征优化下基于注意力机制的输电线路小目标故障检测算法。首先，设计动态特征优化算法以提高原始图像质量，增强小目标故障图像的清晰度；然后，在YOLOv5主干网络中嵌入高效通道注意力模型（Efficient Channel Attention，ECA），通过共享学习参数实现通道之间信息交互来增强故障目标在复杂背景中的显著度；其次，引入深度可分离卷积代替原网络卷积操作，减少网络参数量和运算量；最后，通过在CSPDarkNet中引入Mish激活函数加快网络收敛速度，进一步提升算法检测精度。通过与5种经典目标检测算法对比，实验结果表明，所提出算法具有明显优势，平均精度（mean Average Precision，mAP）可达92.2%。

（2）针对航拍图像中小目标故障像素占比小及故障样本不足造成检测网络精度低的问题，提出一种融合先验知识的输电线路小目标故障检测算法。首先，为解决输电线路小目标故障样本匮乏导致网络检测精度低的问题，在实验室环境下构造输电线路小目标故障数据集对样本进行扩充；其次，构建基于视觉显著性的先验知识迁移模型，从实验室数据集中学习小目标故障特征，并将其迁移至主干网络中；接着，在主干网络中嵌入Transformer模块，从深层特征图中挖掘动静态上下文信息，从而提高网络对故障目标特征提取能力；然后，利用Alpha-CIOU（Alpha-Complete Intersection overUnion）损失函数进一步提高定位准确性并降低网络漏检率；最后，将其与5种经典检测算法进行对比。实验结果表明，所提出算法mAP值最高，可达94.5%，同时具有良好的实时性，检测速度可达40fps，能够实现小样本条件下小目标故障的准确检测。

Small target components such as screws and pins play a role in fixing, supporting and protecting transmission lines. Due to their long-term exposure to complex environments, they are susceptible to loosening and missing due to factors such as harsh natural conditions, mechanical tension and material ageing, which can have a serious impact on the safe and stable operation of transmission lines. Therefore, facing the complex inspection environment, this paper explores a transmission line small target fault detection method that incorporates a priori knowledge and attention mechanism.

The main work of the paper includes:

(1) To address the problem that small target fault features in aerial images are not sufficiently expressed and seriously affected by complex backgrounds, which makes them difficult to be detected accurately, an attention-based algorithm for small target fault detection on transmission lines with dynamic feature optimization is proposed. Firstly, the dynamic feature optimization algorithm is designed to improve the quality of the original image and enhance the clarity of the small target fault image. Secondly, the Efficient Channel Attention (ECA) model is embedded in the YOLOv5 backbone network to enhance the saliency of fault targets in complex backgrounds by sharing learning parameters to achieve information interaction between channels. Then, the depth-separable convolution is introduced instead of the original network convolution operation to reduce the number of network parameters and amount of operations. Finally, by introducing the Mish activation function in CSPDarkNet, the network convergence speed is accelerated to further improve the detection accuracy of the algorithm. By comparing with five classical target detection algorithms, the experimental results show that the proposed algorithm has obvious advantages and the mean Average Precision (mAP) can reach 92.2%.

(2) A transmission line small target fault detection algorithm incorporating a priori knowledge is proposed to address the problems of low detection network accuracy due to the small percentage of small target fault pixels in aerial images and insufficient fault samples. Firstly, to solve the problem of low detection accuracy due to the lack of transmission line small target fault samples, a transmission line small target fault dataset is constructed in the laboratory environment to expand the samples. Secondly, an a priori knowledge migration model based on visual saliency is constructed to learn small target fault features from the laboratory dataset and migrate them to the backbone network. Then, the Transformer module is embedded in the backbone network to mine dynamic and static contextual information from the deep feature map, thus improving the network's ability to extract features from fault targets. Furthermore, the Alpha-CIOU（Alpha-Complete Intersection overUnion）loss function is used to further improve the localization accuracy and reduce the network's miss detection rate. Finally, it is compared with five classical detection algorithms. The experimental results show that the proposed algorithm has the highest mAP value of 94.5% and also has good real-time performance with detection speed up to 40fps, which can achieve accurate detection of small target faults under small sample conditions.

[1]隋宇, 宁平凡, 牛萍娟, 等. 面向架空输电线路的挂载无人机电力巡检技术研究综述[J]. 电网技术, 2021, 45(09): 3636-3648.

[2]赵杰伦, 张兴忠, 董红月. 基于尺度不变特征金字塔的输电线路缺陷检测[J]. 计算机工程与应用, 2022, 58(08): 289-296.

[3]程海燕, 韩璞, 王迪, 等. 一种电网巡检航拍图像中绝缘子定位方法[J]. 系统仿真学报, 2017, 29(06): 1327-1336.

[4] 顾晓东, 唐丹宏, 黄晓华. 基于深度学习的电网巡检图像缺陷检测与识别[J]. 电力系统保护与控制, 2021, 49(05): 91-97.

[5]缪希仁, 林志成, 江灏, 等. 基于深度卷积神经网络的输电线路防鸟刺部件识别与故障检测[J]. 电网技术, 2021, 45(01): 126-133.

[6]崔江波, 侯兴松. 基于注意力机制的YOLOv4输电线路故障检测算法[J]. 国外电子测量技术, 2021, 40(07): 24-29.

[7]刘兰兰, 万旭东, 汪志刚, 等. 基于超分辨率重建与多尺度特征融合的输电线路缺陷检测方法[J]. 电子测量与仪器学报, 2023, 37(01):130-139.

[8]肖志云, 王海强. 图像双分割与小波域多特征融合的高压输电线路典型小目标故障识别[J]. 电网技术, 2021, 45(11): 4461-4470.

[9]Montambault S, Beaudry J, Toussaint K, et al. On the application of VTOL UAVs to the inspection of power utility assets[C]// 2010 1st international conference on applied robotics for the power industry(CAIPR). IEEE, 2010: 1-7.

[10]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.

[11]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.

[12]Wei J, Zhou J, Du H, et al. Flying velocity constraint control for quad-rotor system based on finite-Time control technique[C]// 2020 39th Chinese Control Conference (CCC). Shenyang, China: IEEE, 2020: 311-316.

[13]刘志颖, 缪希仁, 陈静, 等. 电力架空线路巡检可见光图像智能处理研究综述[J]. 电网技术, 2020, 44(03): 1057-1069.

[14]龙乐云, 周腊吾, 刘淑琴, 等. 改进YOLOv5算法下的输电线路外破隐患目标检测研究[J]. 电子测量与仪器学报, 2022, 36(11): 245-253.

[15]王凌, 张冰, 陈锡爱. 基于计算机视觉的钢轨扣件螺母缺失检测系统[J]. 计算机工程与设计, 2011, 32(12): 4147-4150.

[16]付晶, 邵瑰玮, 吴亮, 等. 利用层次模型进行训练学习的线路设备缺陷检测方法[J]. 高电压技术, 2017, 43(01): 266-275.

[17]金立军, 姚春羽, 闫书佳, 等. 基于航拍图像的输电线路异物识别[J]. 同济大学学报(自然科学版), 2013, 41(02): 277-281.

[18]Zhao Z, Liu N, Wang L. Localization of multiple insulators by orientation angle detection and binary shape prior knowledge[J]. IEEE Transactions on Dielectrics and Electrical Insulation, 2015, 22(6): 3421-3428.

[19]Wu Q, An J, Lin B. A texture segmentation algorithm based on PCA and global minimization active contour model for aerial insulator images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2012, 5(5): 1509-1518.

[20]金立军, 胡娟, 闫书佳. 基于图像的高压输电线间隔棒故障诊断方法[J]. 高电压技术, 2013, 39(05): 1040-1045.

[21]黄新波, 张晓霞, 李立浧, 等. 采用图像处理技术的输电线路导线弧垂测量[J]. 高电压技术, 2011, 37(08): 1961-1966.

[22]李朝阳, 阎广建, 肖志强, 等. 高分辨率航空影像中高压电力线的自动提取[J]. 中国图象图形学报, 2007, 12(6): 1041-1047.

[23]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.

[24]刘传洋, 吴一全. 基于深度学习的输电线路视觉检测方法研究进展[J/OL].中国电机工程学报:1-24.

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

[26]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.

[27]Dai J, Li Y, He K, et al. R-FCN: Object detection via region-based fully convolutional networks[C]// Proceeding of the 30th International Conference on Neural Information Processing System(NIPS), Barecelona, Spain: MIT Press, 2016: 379-387.

[28]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.

[29]Zhao J, Zhang K, Wang Z, et al. Limited sliding network: Fine grained target detection on electrical infrastructure for power transmission line surveillance[J]. International Journal of Circuit Theory and Applications, 2021, 49(4): 1212-1224.

[30]赵丽娟,柳长安,张正,等.输电线路中销钉缺陷的自适应检测技术研究[J]. 华中科技大学学报(自然科学版), 2023, 51(02): 109-115+160.

[31]翟永杰, 杨旭, 赵振兵, 等. 融合共现推理的Faster R-CNN输电线路金具检测[J]. 智能系统学报, 2021, 16(02): 237-246.

[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]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.

[34]赵振兵, 熊静, 李冰, 等. 基于改进Cascade R-CNN的典型金具及其部分缺陷检测方法[J]. 高电压技术, 2022, 48(03): 1060-1067.

[35]Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]// European Conference on Computer Vision(ECCV). Amsterdam, The Netherlands: Springer, 2016: 21-37.

[36]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(ICCV). Venice, Italy: IEEE, 2017: 2980-2988.

[37]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(CVPR). Las Vegas, USA: IEEE, 2016: 779-788.

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

[39]Redmon J, Farhadi A. Yolov3: An incremental improvement[EB/OL]. . https://arxiv.org/abs/1804.02767.

[40]Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection[EB/OL]. https: //arxiv.org/abs/2004.10934.

[41]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.

[42]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(08): 2979-2987.

[43]刘国文, 张彩霞, 李斌, 等. 基于改进RetinaNet模型的接触网鸟巢检测[J]. 数据采集与处理, 2020, 35(03): 563-571.

[44]Liu Z, Wu G, He W, et al. Key target and defect detection of high-voltage power transmission lines with deep learning[J]. International Journal of Electrical Power & Energy Systems, 2022, 142: 108277.

[45]Liu M, Li Z, Li Y, et al. A fast and accurate method of power line intelligent inspection based on edge computing[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-12.

[46]赵振兵, 王帆帆, 刘良帅, 等. 基于注意力特征融合YOLOv5模型的无人机输电线路航拍图像金具检测方法[J]. 电测与仪表, 2023, 60(03): 145-152.

[47]张永翔, 吴功平, 刘中云, 等. 基于YOLOv3网络的输电线路防震锤和线夹检测迁移学习[J]. 计算机应用, 2020, 40(S2): 188-194.

[48]黄力, 万旭东, 王凌云, 等. 基于图像增广与迁移学习的输电线路金具多目标实时检测方法[J]. 电子测量技术, 2022, 45(20): 135-142.

[49]郝帅, 张旭, 马旭, 等.基于CBAM-YOLOv5的煤矿输送带异物检测[J]. 煤炭学报, 2022, 47(11): 4147-4156.

[50]Albawi S, Mohammed T A, Al-Zawi S. Understanding of a convolutional neural network[C]//2017 international conference on engineering and technology (ICET). Antalya, Türkiye, IEEE, 2017: 1-6.

[51]Gulcehre C, Moczulski M, Denil M, et al. Noisy activation functions[C]//International conference on machine learning(ICML). New York, USA, 2016: 3059-3068.

[52]Nair V, Hinton G E. Rectified linear units improve restricted boltzmann machines[C]//Proceedings of the 27th international conference on machine learning (ICML). Haifa, Israel, 2010: 807-814.

[53]Amari S I. Natural gradient works efficiently in learning[J]. Neural computation, 1998, 10(2): 251-276.

[54]张麒麟, 林清平, 肖蕾. 改进YOLOv5的航拍图像识别算法[J]. 长江信息通信, 2021, 34(3): 73-76.

[55]王菲, 王球, 任佳依, 等. 变电站电气设备检测与三维建模系统[J]. 电测与仪表, 2021, 58(3): 160-167.

[56]Rezatofighi H, Tsoi N, Gwak J, et al. Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 658-666.

[57]Sandler M, Howard A, Zhu M L, et al. MobileNetV2: Inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR), Salt Lake City, USA: IEEE, 2018: 4510-4520.

[58]Misra D. Mish: A self regularized non-monotonic activation function[J]. arXiv preprint arXiv: 1908. 08681, 2019.

[59]Fan H Y, Ma J S, Fan H H, et al. Iterative quadtree decomposition based automatic selection of the seed point for ultrasound breast tumor images[J]. Multimedia Tools and Applications, 2017, 76(3): 3505-3517.

[60]Mun J, Jang Y, Nam Y, et al. Edge-enhancing bi-histogram equalisation using guided image filter[J]. Journal of visual communication & image representation, 2019, 58: 688-700.

[61]Ma J Y, Ma Y, Li C. Infrared and visible image fusion methods and applications: A survey[J]. Information Fusion, 2019, 45: 153-178.

[62]Khishe M, Mosavi M R. Chimp optimization algorithm[J]. Expert systems with applications, 2020, 149: 113338.

[63]Hu J, Shen L, Sun G. Squeeze and excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, USA: IEEE, 2018: 7132-7141.

[64]Wang Q, Wu B, Zhu P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C]. Proceedings of 2020 IEEE Conference on computer Vision and Pattern Recognition(CVPR), Virtual, online, United states. IEEE, 2020: 11531-11539.

[65]龚任杰, 郑智辉, 丛龙剑, 等. 小样本条件下异源图像迁移学习的红外目标检测与识别[J]. 西北工业大学学报, 2021, 39(S1): 84-88.

[66]Li Y, Yao T, Pan Y, et al. Contextual transformer networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(02)1489-1500.