随着航拍技术的快速发展，航拍图像在绝缘子缺陷检测中扮演着重要的角色。然而， 航拍图像中常常存在复杂背景（如建筑、树木、街道和杆塔等），这给绝缘子缺陷的检 测带来了巨大的挑战。由于背景复杂很难在图像中突出绝缘体缺陷，提高了检测能力复杂程度，而及时发现绝缘缺陷对电力系统的安全运行至关重要，所以应对复杂背景进行绝缘子缺陷检测技术的研究迫切需要。本文旨在探索并构建一种创新的方法，通过级联 网络的结构，充分利用图像特征和上下文信息，以提高绝缘子缺陷检测的准确性和鲁棒 性，为电力系统的安全运行提供了可靠的保证。本文的研究工作总结如下：

（1）在绝缘子目标定位阶段，针对绝缘子多尺度特性和边界信息不明显问题，设计了一种基于双层语义编码和超像素边缘增强的 Ela-SeNet 分割定位算法。采用 Racbam-MobileNetV3 作为算法中的特征提取网络，减少网络参数；通过双层编码器来增强绝缘 子多尺度目标特征信息，同时，融合超像素的方式对边界模糊化进行处理，使目标边缘信息明显。基于 Pytorch 框架对该网络模型进行训练和前向预测，为了体现该网络在像素分割准确率和平均交并比指标的优越性，通过消融实验和不同分割网络的对比实验进行了验证。

（2）在绝缘子缺陷检测过程中，针对不同特征层冲突信息直接融合以及难易样本和正负样本不平衡的问题，采用了一种基于 CF-YOLOv7 的缺陷检测算法。其中，带特征细化结构的跨尺度特征金字塔（Cross Scale Feature Pyramid Network，CSFPN）来抑制不同特征层的冲突信息特征，提高网络的多尺度表达能力；结合 Focal-Loss 的权重调整机制和 SIOU-Loss 的结构相似性的 Focal-SIOU-Loss 回归损失函数，提高回归过程的准确性和稳定性；同时，采用 REP-GhostConv 和 Convext 轻量化模块减少模型参数量，提高检测速度。通过各模块的实验验证以及与原始 YOLOv7 检测算法的对比实验，CF-YOLOv7 缺陷检测网络在损失结果和平均精度方面均取得较好的表现，验证了其有效性和可行性。

（3）由于复杂背景对绝缘子缺陷特征提取时的强干扰问题，构建了模型级联网络（Cascade Network YOLO，CN-YOLO）来将 Ela-SeNet 分割定位后的绝缘子裁剪图像区域作为 CF-YOLOv7 缺陷检测网络的输入图像进行训练，检测出缺陷部分的具体位置。通过与其他学者提出的绝缘子缺陷检测网络以及最新的先进目标检测算法进行详尽的对比实验分析，CN-YOLO 级联网络在应对具有复杂背景的绝缘子数据集时，表现出了非常良好和可靠的检测效果，并且在准确率、参数量和检测速度方面可以达到最佳平衡，充分说明了该级联网络的可行性。

本文对航拍图像下的复杂背景绝缘子缺陷进行分析，通过基于级联网络的方式，针对绝缘子缺陷特征在复杂图像中易丢失的难点，分别在检测精度和检测速度上进行了研究。经过与其它复杂背景下的绝缘子缺陷检测网络进行了对比实验，验证了本文所提方法能够满足电力巡检的实时准确的要求，具有理论意义和研究价值。

With the rapid development of aerial photography technology, aerial images play an important role in insulator defect detection. However, complex backgrounds such as buildings, trees, streets, and towers often exist in aerial images, which poses a huge challenge to the detection of insulator defects. Due to the complexity of the background, it is difficult to highlight insulation defects in the image, which increases the complexity of detection capability. Timely detection of insulation defects is crucial for the safe operation of the power system. Therefore, there is an urgent need for research on defect detection technology for insulators with complex backgrounds. This article aims to explore and propose an innovative method that fully utilizes image features and contextual information through a cascaded network structure to improve the accuracy and robustness of insulator defect detection, providing a reliable guarantee for the safe operation of the power system. The research work of this article is summarized as follows:

(1) In the stage of insulator target localization, an Ela-SeNet segmentation and localization algorithm based on double-layer semantic encoding and superpixel edge enhancement was designed to address the issues of multi-scale and unclear boundary information of insulators. Racbam-MobileNetV3 was used as the feature extraction network in the algorithm to reduce network parameters; By using a double-layer encoder to enhance the multi-scale target feature information of insulators, and integrating superpixels to process boundary blurring, the edge information is clear. Based on the Pytorch framework, the network model was trained and predicted forward to demonstrate its superiority in pixel segmentation accuracy and average intersection to union ratio metrics. Verified through ablation experiments and comparative experiments with different segmentation networks.

(2) In the process of insulator defect detection, a defect detection algorithm based on CF-YOLOv7 is adopted to address the issues of direct fusion of conflicting information from different feature layers and imbalance between difficult and easy samples and positive and negative samples. The Cross Scale Feature Pyramid Network (CSFPN) with feature refinement structure is used to suppress conflicting information features from different feature layers and improve the multi-scale expression ability of the network; The Focal SIOU Loss regression loss function combines the weight adjustment mechanism of Focal Loss with the structural similarity of SIOU Loss to improve the accuracy and stability of the regression process. Meanwhile, the use of REP-GhostConv and Convext lightweight modules reduces the number of model parameters and improves detection speed. Through experimental verification of each module and comparative experiments with the original YOLOv7 detection algorithm, the defect detection network proposed in this paper has achieved good performance in terms of loss results and average accuracy, verifying its effectiveness and feasibility.

(3) Due to the strong interference problem of complex backgrounds in insulator defect feature extraction, a Cascade Network YOLO (CN-YOLO) is proposed to train the insulator region cropped image of the Ela-SeNet segmentation and localization algorithm as the input image of the CF-YOLOv7 defect detection network to detect the specific location of the defect. Through detailed comparative experimental analysis with insulator defect detection networks proposed by other scholars and the latest advanced object detection algorithms, the method explored in this paper demonstrates very good and reliable detection results when dealing with insulator datasets with complex backgrounds, and can achieve the best balance in accuracy, parameter quantity, and detection speed, fully demonstrating the feasibility of the cascaded network.

This article analyzes the complex background insulator defects in aerial images, and studies the difficulty of insulator defect features being easily lost in complex images through a cascaded network approach. The detection accuracy and speed are studied eparately. Through comparative experiments with other insulator defect detection networks in complex backgrounds, it has been verified that the method proposed in this paper can meet the real-time and accurate requirements of power inspection, and hasmtheoretical significance and research value.

[1]杨德昌, 廖文, 李洪雨. 关于超高压输电线路运行管理问题及优化对策探析[J]. 科技创新与应用, 2017, 14(36): 100-101.

[2]李映国, 杨宏, 徐郁等. 基于Alexnet网络的绝缘子自爆无人机巡检技术研究[J]. 智慧电力, 2021, 49(08): 104-109.

[3]陈俊杰, 叶东华, 产焰萍等. 基于Faster R-CNN模型的绝缘子故障检测[J]. 电工电气, 2020, 268(04): 56-60.

[4]She L, Fan Y, Wang J, et al. Insulator surface breakage recognition based on multi-scale residual neural network[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70(14): 1-9.

[5]Feng S, Sheng H, Yong L, et al. Rsin-dataset: an uav-based insulator detection aerial images dataset and benchmark[J]. Drones, 2023, 7(2): 125-125.

[6]刘高. 面向输电线路精细化巡检的无人机航线自动规划方法[J]. 电子设计工程, 2021, 29(04): 33-37．

[7]汤丹, 方挺. 基于核相关滤波的多尺度绝缘子目标跟踪[J]. 电瓷避雷器, 2019, 19(06): 205-209.

[8]沈杰, 陈玉权, 吴媚等. 基于图像分割技术的玻璃绝缘子缺陷检测[J]. 信息技术, 2022, 8(05): 165-176.

[9]Tang B, Qin Q, Huang L. Transmission line aerial image recognition of insulator strings based on color model and texture features[J]. Journal of Electric Power Science and Technology, 2020, 35(4): 13-19.

[10]Huang X, Zhang Z. A method to extract insulator image from aerial image of helicopter patrol[J]. Power System Technology, 2010, 34(1): 194-197.

[11]Gerald Z, David G, Martin B, et al. Gabor filter-based segmentation of railroad radargrams for improved rail track condition assessment: preliminary studies and future erspectives[J]. Remote Sensing, 2021, 13(21): 4293-4293.

[12]程伟涛, 郭来功, 汪强. 基于FPGA的多方向阈值自适应Sobel算法实现[J].无线互联科技, 2023, 20(09): 142-145.

[13]张红民, 彭闯, 郭泓邑等. 一种基于SIFT-BOW模型的绝缘子图像识别方法[J]. 激光杂志, 2020, 41(11): 96-99.

[14]Zuo D, Hu H, Qian R, et al. An insulator defect detection algorithm based on computer vision[C]//Proceedings of the International Conference on Information and Automation(ICIA), 2017: 361-365 .

[15]Jabid T, Ahsan T. Insulator detection and defect classification using rotation invariant local directional pattern[C]//Proceedings of the International Journal of Advanced Computer Science and Applications(IJACSA), 2018: 265-272.

[16]Tang W, Niu Z, Wu Q, et al. Research and application of data-driven artificial intelligence technology for condition analysis of power equipment[J]. Engineering, Computer Science, 2020, 46(18): 2985–2999 .

[17]Zhang Y, Qiu C, Yang F, et al. Overview of application of deep learning with image data and spatio-temporal data of power grid[J]. Power System Technology, 2019, 43(06): 1865-1873 .

[18]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies foraccurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), 2014: 580-587 .

[19]Li J, Liang X, Shen S, et al. Scale-aware fast R-CNN for pedestrian detection[J]. IEEE Transactions on Multimedia, 2018, 20(4): 985-996 .

[20]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.

[21]Zhou Z. Study of insulator self-burst defects based on multiscale feature fusion Faster R-CNN[J]. Hydropower Energy Science, 2020, 38(11): 187-189 .

[22]Kang G, Gao S, Yu L, et al. Deep architecture for high-speed railway insulator surface defect detection: denoising auto encoder with multitask learning[J]. IEEE Transactions on Instrumentation and Measurement, 2019, 68(8): 2679-2690.

[23]彭曙蓉, 刘登港, 何洁妮等. 基于改进YOLOv4的架空线路电力设备故障检测[J]. 电力科学与技术学报, 2023, 38(05): 169-176.

[24]Yang Y, Wang H, Jiang D, et al. Surface detection of solid wood defects based on SSD improved with resnet[J]. Forests, 2021, 12(10): 1419 -1430.

[25]Xia H, Yang B, Li Y, et al. An improved centernet model for insulator defect detection using aerial imagery[J]. Sensors, 2022, 22(8): 2850 -2863.

[26]Zheng J, Wu H, Zhang H, et al. Insulator-defect detection algorithm based on improved YOLOv7[J]. Sensors, 2022, 22(22): 1-23.

[27]叶力鸣, 陈蔚文. 一种结合语义分割和目标检测的级联式绝缘子缺陷检测方法[J]. 计算机与现代化, 2023, 37(06): 82-88.

[28]Wang S, Zhou Z, Zhao W. Semantic segmentation and defect detection of aerial insulators of transmission lines[J]. Journal of Physics: Conference Series, 2022, 2185(1): 012086-012098.

[29]Tao X, Zhang D, Wang Z, et al. Detection of power line insulator defects using aerial images analyzed with convolutional neural networks[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2020, 50(4): 1486- 1498.

[30]宋永, 刘德春. 基于多滤波器组合的光栅图像噪声抑制方法[J]. 激光杂志, 2023, 44(09): 125-129.

[31]Cheng S, Ma J. Smoke detection and trend prediction method based on Deeplabv3+ and generative adversarial network[J]. Journal of Electronic Imaging, 2019, 28(3): 033006-033017.

[32]Yang Y, Han J. Real-time object detector based MobileNetV3 for UAV applications[J]. Multimedia Tools and Applications, 2023, 82(11): 18709-18725.

[33]Lee C, So E, Saidy L, et al. Lung field segmentation in chest X-ray images using superpixel resizing and encoder-decoder segmentation networks[J]. Bioengineering, 2022, 9(8): 351-358.

[34]Delaigle A, Hall P. Nonparametric regression with homogeneous group testing data[J]. Annals of Statistics, 2012, 40(1): 131-158.

[35]Feng M, Hu J, Ou M, et al. Crowd counting algorithm based on depthwise separable of dilated convolution[J]. Procedia Computer Science, 2022, 208(14): 1245-1252 .

[36]Lee S, Chung J. The machine learning-based dropout early warning system for improving the performance of dropout prediction[J]. Applied Sciences, 2019, 9(15): 3093-3102.

[37]Sara S, Nicholas F, Geoffrey E H. Dynamic routing between capsules[C]. In 2017 Conference and Workshop on Neural Information Processing Systems, 2017: 1-17.

[38]Ullah S, Bhatti N, Zia M. Adaptive tuning of SLIC parameter K[J]. Multimedia Tools and Applications, 2021, 80(14): 25649–25672.

[39]Dong C, Tang Y, Zhang L. Higher efficient YOLOv7: a one-stage method for non-salient object detection[J]. Multimedia Tools and Applications, 2023, 42(23): 17185-17197.

[40]Cui M, Duan Y, Pan C, et al. Optimization for anchor-free object detection via scale-independent GIOU loss[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20(6), 1-5.

[41]Han G, Zhao L, Li Q, et al. A lightweight algorithm for insulator target detection and defect identification[J]. Sensors, 2023, 23(3): 106-112.

[42]Chen G, Zhou H, Li Z, et al. Multi-scale forest fire recognition model based on improved YOLOv5s[J]. Forests, 2023, 14(2): 267-275.

[43]Du J, Zhao S, Shang C, et al. Applying image analysis to build a lightweight system for blind obstacles detecting of intelligent wheelchairs[J]. Electronics, 2023, 12(21): 4472-4486.

[44]Chu X, Li L, Zhang B. Make RepVGG greater again: a quantization-aware approach[J]. arXiv - CS - Computer Vision and Pattern Recognition, 2023, 15(7): 126-133.

[45]Park S, Yeo Y, Shin Y. PConv: simple yet effective convolutional layer for generative adversarial network[J]. Neural Computing and Applications, 2022, 34(26): 7113-7124.

[46]Park S, Shin Y. Generative convolution layer for image generation[J]. Neural Networks, 2022, 152(19): 370-379.

[47]Wang H, Qian H, Feng S, et al. L-SSD: lightweight SSD target detection based on depth-separable convolution[J]. Journal of Real-Time Image Processing, 2024, 21(33):0329-0338.

[48]Zhao X, Song Y. Improved ship detection with YOLOv8 enhanced with MobileViT and GSConv[J]. Electronics, 2023, 12(22): 4666-4679.

[49]Gennari M, Fawcett R, Prisacariu V. DSConv: efficient convolution operator[J].. arXiv - CS - Computer Vision and Pattern Recognition, 2019, 27(4): 3448-3456.

[50]Tian B, Chen H. Remote sensing image target detection method based on refined feature extraction[J]. Applied Sciences, 2023, 13(15): 8694-8703.

[51]Zhu X, Liu S, Fan B, et al. Forensics analysis of resampling via convnext block[J]. Journal of Circuits, Systems and Computers, 2023, 32(2): 2350033-2350047.

[52]Yu C, Shin Y. SAR ship detection based on improved YOLOv5 and bifpn[J]. ICT Express, 2024, 10(1): 28-33.

[53]Wang Q, Feng W, Yao L, et al. TPH-YOLOv5-air: airport confusing object detection via adaptively spatial feature fusion[J]. Remote Sensing, 2023, 15(15): 3883-3896.

[54]Zheng J, Wu H, Zhang H, et al. Insulator-defect detection algorithm based on improved YOLOv7[J]. Sensors, 2022, 22(22): 1-23.

[55]Raju S, Rao S. Colorectal cancer disease classification using mobilenetv2 based on deeplearning[J]. International Journal of Software Innovation, 2022, 10(1): 1-18.

[56]Wang C, He W, Nie Y, et al. Gold-YOLO: efficient object detector via gather-and-distribute mechanism[J]. arXiv - CS - Computer Vision and Pattern Recognition, 2023, 23(9): 11331-11342.

[57]Ding J, Cao H, Ding X, et al. High-accuracy insulator defect detection for overhead transmission lines based on improved YOLOv5[J]. Frontiers in Energy Research, 2022, 12(24): 1-13.

[58]Hao K, Chen G, Zhao L, et al. An insulator defect detection model in aerial images based on multiscale feature pyramid network[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71(9): 1-12.

[59]Nan Y, Zhang H, Zeng Y, et al. Faster and accurate green pepper detection using NSGA-II-based pruned YOLOv5l in the field environment[J]. Computers and Electronics in Agriculture, 2023, 205(33): 107563-107573.

[60]Han G, Li T, Li Q, et al. Improved algorithm for insulator and its defect detection based on YOLOX[J]. Sensors, 2022, 22(16): 6186-6198.

[61]Yi W, Ma S, Li R. Insulator and defect detection model based on improved YOlO-S[J]. IEEE Access, 2023, 11(29): 93215 - 93226.

[62]He M, Qin L, Deng X, et al. MFI-YOLO: multi-fault insulator detection based on an improved YOLOv8[J]. IEEE Transactions on Power Delivery, 2023, 39(1): 168-179.

[63]Chun W, Xin M, Xiang K, et al. Research on insulator defect detection algorithm of transmission line based on CenterNet[J]. PLOS ONE, 2021, 16(7): 0255135-0255146.

[64]Ping T, Xu L, Jin D, et al. Mask R-CNN and multifeature clustering model for catenary insulator recognition and defect detection[J]. Journal of Zhejiang University-SCIENCE A, 2022, 23(14): 745–756.

[65]Dong Z, Shi G, Long Y, et al. DefGAN: defect detection GANs with latent space pitting for high-speed railway insulator[J]. IEEE Transactions on Instrumentation and Measurement, 2020, 70(31): 3038008-3038020.

[66]Li X, Su H, Liu G. Insulator defect recognition based on global detection and local segmentation[J]. IEEE Access, 2020, 8(35): 59934-59946.

[67]Chen B, Zhang W, Wu W, et al. ID-YOLOv7: an efficient method for insulator defect detection in power distribution network[J]. Frontiers in Neurorobotics, 2024, 17(40): 1427-1439.