Insulator Fault Identification Based on Image Fusion Technology and Improved Faster R-CNN

GENG Zhanquan; SU Fei

doi:10.13296/j.1001-1609.hva.2025.10.014

您当前的位置：

首页 >

文章列表页 >

Insulator Fault Identification Based on Image Fusion Technology and Improved Faster R-CNN

Digital Twin Technology and Applications for Multi-scenario in New Power System | 更新时间：2025-10-21

- Insulator Fault Identification Based on Image Fusion Technology and Improved Faster R-CNN
- High Voltage Apparatus Vol. 61, Issue 10, Pages: 128-138(2025)
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：10.13296/j.1001-1609.hva.2025.10.014
  CLC：
- Received：02 March 2025，
  
  Revised：2025-06-01，
  
  Published：16 October 2025
- 稿件说明：
移动端阅览
GENG Zhanquan, SU Fei. Insulator Fault Identification Based on Image Fusion Technology and Improved Faster R-CNN[J]. High voltage apparatus, 2025, 61(10): 128-138.
DOI：

GENG Zhanquan, SU Fei. Insulator Fault Identification Based on Image Fusion Technology and Improved Faster R-CNN[J]. High voltage apparatus, 2025, 61(10): 128-138. DOI： 10.13296/j.1001-1609.hva.2025.10.014.

摘要

为提高地铁接触网绝缘子故障检测精度，文中提出一种融合双流注意力生成对抗网络（DSAGAN）与改进Faster R-CNN的检测框架。针对多源图像特征融合难题，通过DSAGAN实现接触网绝缘子红外与可见光图像的高效融合：生成器采用注意力机制强化关键特征提取，判别器引入最小二乘损失函数替代传统交叉熵，有效保留纹理细节并提升网络稳定性。在故障识别阶段，改进Faster R-CNN模型采用ResNet101替代VGG16作为主干网络，结合残差连接缓解梯度异常问题，并在RPN层嵌入非极大值抑制（NMS）算法消除冗余候选框。实验表明：DSAGAN在

SSIM

、

和

5项指标上较5种主流方法平均提升20.78%，融合图像质量显著优化；改进Faster R-CNN模型在平均精度均值

mAP

@0.5和

mAP

@0.5∶0.95上分别达到96.5%和78.3%，较原模型分别提升了7.1%和8.1%；融合图像对破损（96%）、闪络（93%）、缺失（94%）的检测准确率分别较单一可见光/红外图像提高8%～12%。文中方法为复杂场景下的电力设备智能巡检提供了有效解决方案。

Abstract

For improving the fault detection accuracy of insulators in subway catenaries

in this paper a detection framework integrating the dual-stream attention-guided generative adversarial network (DSAGAN) and the improved Faster R-CNN is proposed. Aiming at the challenge of multi-source image feature fusion

DSAGAN is used to achieve efficient fusion of infrared and visible-light images of catenary insulators: The generator employs an attention mechanism to enhance the extraction of key features

and the discriminator introduces the least-squares loss function to replace the traditional cross-entropy

which effectively preserving texture details and improving network stability. In fault identification stage

the improved Faster R-CNN model uses ResNet101 instead of VGG16 as the backbone network

combines residual connections to alleviate gradient anomalies

and embeds the non-maximum suppression (NMS) algorithm in the region proposal network (RPN) layer to eliminate redundant candidate boxes. Experiments show that DSAGAN has an average improvement of 20.78% over five mainstream methods in terms of such five indicators as structural similarity index measure (

SSIM

)

entropy (

)

standard deviation (

)

mutual information (

)

and spatial frequency (

)

the quality of fused images significantly optimized；the improved Faster R-CNN model reaches 96.5% and 78.3% in terms of the mean average precision

mAP

@0.5 and

mAP

@0.5∶0.95 respectively

which are 7.1% and 8.1% higher than those of the original model；and the detection accuracies of the fused images for damage (96%)

flashover (93%)

and absence (94%) are 8%-12% higher than those of si

ngle visible-light/infrared images respectively. The method in this paper has provided an effective solution for intelligent inspection of power equipment in complex scenarios.

关键词

Keywords

references

霍红刚 , 周蠡 , 蔡杰 , 等 . 基于先验知识Faster R-CNN的输电线路无人机图像识别方法 [J ] . 智慧电力 , 2024 , 52 ( 6 ): 108 - 115 .

HUO Honggang , ZHOU Li , CAI Jie , et al . UAV image recognition method for transmission line based on prior knowledge Faster RCNN [J ] . Smart Power , 2024 , 52 ( 6 ): 108 - 115 .

ZHANG Zhaoyun , HUANG Shihong , LI Yanxin , et al . Image detection of insulator defects based on morphological processing and deep learning [J ] . Energies , 2022 , 15 ( 7 ): 2465 .

XIA Haiyang , YANG Baohua , LI Yunlong , et al . An improved CenterNet model for insulator defect detection using aerial imagery [J ] . Sensors , 2022 , 22 ( 8 ): 2850 .

唐小煜 , 黄进波 , 冯洁文 , 等 . 基于U-net和YOLOv4的绝缘子图像分割与缺陷检测 [J ] . 华南师范大学学报(自然科学版) , 2020 , 52 ( 6 ): 15 - 21 .

TANG Xiaoyu , HUANG Jinbo , FENG Jiewen , et al . Image segmentation and defect detection of insulators based on U-net and YOLOv4 [J ] . Journal of South China Normal University(Natural Science Edition) , 2020 , 52 ( 6 ): 15 - 21 .

ZHOU Ming , WANG Jue , LI Bo . ARG-Mask RCNN:An infrared insulator fault-detection network based on improved mask RCNN [J ] . Sensors , 2022 , 22 ( 13 ): 4720 .

ZHAO Zhenbing , XU Guozhi , QI Yincheng . Representation of binary feature pooling for detection of insulator strings in infrared images [J ] . IEEE Transactions on Dielectrics and Electrical Insulation , 2016 , 23 ( 5 ): 2858 - 2866 .

ZARIPOVA A D . Convolutional neural network to detect support and suspend insulators on infrared images [C ] // 2023 5th International Youth Conference on Radio Electronics, Electrical and Power Engineering(REEPE) . Russian : IEEE , 2023 : 1 - 5 .

石杰 , 张靖 , 钟汉华 . 基于梯度图像融合的接触网绝缘子故障检测 [J ] . 红外技术 , 2023 , 45 ( 10 ): 1106 - 1117 .

SHI Jie , ZHANG Jing , ZHONG Hanhua . Catenary insulator fault detection based on gradient image fusion [J ] . Infrared Technology , 2023 , 45 ( 10 ): 1106 - 1117 .

李特 , 王尧平 , 李响 , 等 . ±500 kV葛南线绝缘子自然积污特性研究 [J ] . 浙江电力 , 2023 , 42 ( 6 ): 98 - 104 .

LI Te , WANG Yaoping , LI Xiang , et al . A study of natural contamination characteristics of insulators on ±500 kV Gezhouba-Nanqiao HVDC transmission lines [J ] . Zhejiang Electric Power , 2023 , 42 ( 6 ): 98 - 104 .

CHEN Qi , YANG Jingjing , HUANG Ming , et al . ACT-GAN:Radio map construction based on generative adversarial networks with ACT blocks [J ] . IET Communications , 2024 , 18 ( 19 ): 1541 - 1550 .

YAN Jiayue , TAO Chenglong , WANG Yuan , et al . Application of enhanced weighted least squares with dark background image fusion for inhomogeneity noise removal in brain tumor hyperspectral images [J ] . Applied Sciences , 2024 , 15 ( 1 ): 321 - 321 .

NUSANTIKA R N , XIAO Jin , HU Xiaoguang . An enhanced multiscale retinex, oriented FAST and rotated BRIEF(ORB), and scaleinvariant feature transform(SIFT) pipeline for robust key point matching in 3D monitoring of power transmission line icing with binocular vision [J ] . Electronics , 2024 , 13 ( 21 ): 4252 - 4252 .

JONGIL P , CHANGHYUN P . Harmonic contribution assessment based on the random sample consensus and recursive least square methods [J ] . Energies , 2022 , 15 ( 17 ): 6448 - 6448 .

ZRIRA N , JIMI A , NARDO D M , et al . GCBAM-UNet:Sun glare segmentation using convolutional block attention module [J ] . Applied System Innovation , 2024 , 7 ( 6 ): 128 - 128 .

PAN Yue , PI Dechang , CHEN Junfu , et al . FDPPGAN:Remote sensing image fusion based on deep perceptual patchGAN [J ] . Neural Computing and Applications , 2021 , 33 ( 15 ): 9589 - 9605 .

刘影 , 彭鑫霞 , 王童 , 等 . 基于PCA-LSTM算法的非侵入式负荷辨识方法 [J ] . 电测与仪表 , 2023 , 60 ( 3 ): 53 - 58 .

LIU Ying , PENG Xinxia , WANG Tong , et al . Non-intrusive load monitoring method based on PCA-LSTM algorithm [J ] . Electrical Measurement&Instrumentation , 2023 , 60 ( 3 ): 53 - 58 .

KARIM S , TONG Geng , LI Jinyang , et al . Current advances and future perspectives of image fusion:A comprehensive review [J ] . Information Fusion , 2023 ( 90 ): 185 - 217 .

ZHU Xiaoliang , BAO Wenxing . Performance comparison of image fusion alternatives combining PCA with multi-resolution wavelet transforms [J ] . Journal of the Indian Society of Remote Sensing , 2024 , 52 ( 5 ): 943 - 956 .

LABBIHI I , EL MESLOUHI O , ELASSAD Z E A , et al . Hybrid 3D medical image segmentation using CNN and frequency transformer fusion [J ] . Arabian Journal for Science and Engineering , 2024 ( 9 ): 1 - 14 .

张帅 . 绝缘子红外图像自动检测及设备铭牌信息自动提取方法研究 [D ] . 西安 : 西安理工大学 , 2022 .

ZHANG Shuai . Research on automatic detection of insulator infrared image and automatic extraction method of equipment nameplate information [D ] . Xi'an : Xi'an University of Technology , 2022 .

黄浩川 . 基于稀疏表示算法的复合绝缘子憎水性等级识别研究 [D ] . 长沙 : 湖南大学 , 2013 .

HUANG Haochuan . Research on hydrophobic grade identification of composite insulators based on sparse representation algorithm [D ] . Changsha : Hunan University , 2013 .

LIU Bin , LI Jianfei , YANG Xue , et al . Diagnosis of primary clear cell carcinoma of the liver based on faster region-based convolutional neural network [J ] . Chinese Medical Journal , 2023 , 136 ( 22 ): 2706 - 2711 .

LEE H K , LEE W R , YUN S J , et al . Automated diagnosis of knee osteoarthritis using ResNet101 on a DEEP:PHI:Leveraging a nocode AI platform for efficient and accurate medical image analysis [J ] . Diagnostics , 2024 , 14 ( 21 ): 2451 - 2451 .

SARSHAR T N , SADEGHI S , KAMSARI M , et al . Advancing brain MRI image classification:Integrating VGG16 and ResNet50 with a multi-verse optimization method [J ] . BioMed , 2024 , 4 ( 4 ): 499 - 523 .

LIU G , WANG C . A novel multi-scale feature fusion method for re-gion proposal network in fast object detection [J ] . International Journal of Data Warehousing and Mining(IJDWM) , 2020 , 16 ( 3 ): 132 - 145 .

GUO Chunsheng , CAI Meng , YING Na , et al . ANMS:Attention-based non-maximum suppression [J ] . Multimedia Tools and Appli-cations , 2022 , 81 ( 8 ): 11205 - 11219 .

YINGJUN W , JUNFENG S , WENXUE M , et al . Bridge crack recog-nition method based on YOLOv5 neural network fused with atten-tion mechanism [J ] . International Journal of Intelligent Information Technologies(IJIIT) , 2024 , 20 ( 1 ): 1 - 25 .

WEIDONG L , ZHENYING L , CHISHENG W , et al . An improved SSD lightweight network with coordinate attention for aircraft target recognition in scene videos [J ] . Journal of Intelligent & Fuzzy Systems , 2024 , 46 ( 1 ): 355 - 368 .

GUO Xudong , XU Lei , LI Shengnan , et al . Cascade-EC network:Recognition of gastrointestinal multiple lesions based on Efficient-Net and CA stm retinanet [J ] . Journal of Imaging Informatics in Medicine , 2024 , 37 ( 5 ): 1 - 11 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

No data

Related Author

No data

Related Institution

No data

⁰