《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (3): 969-979.DOI: 10.11772/j.issn.1001-9081.2025030340
黄萍1(
), 李清1, 邱海枫1, 王程斯1, 黄安子1, 樊龙2
收稿日期:2025-04-03
修回日期:2025-06-11
接受日期:2025-06-12
发布日期:2025-06-27
出版日期:2026-03-10
通讯作者:
黄萍
作者简介:李清(1988—),女,湖南衡阳人,高级工程师,硕士,主要研究方向:电力信息化
Ping HUANG1(
), Qing LI1, Haifeng QIU1, Chengsi WANG1, Anzi HUANG1, Long FAN2
Received:2025-04-03
Revised:2025-06-11
Accepted:2025-06-12
Online:2025-06-27
Published:2026-03-10
Contact:
Ping HUANG
About author:LI Qing, born in 1988, M. S., senior engineer. Her research interests include power informatization.摘要:
作为电力系统的核心输配电载体,高压输电线路的运行状态直接关系到电网安全。针对传统人工巡检效率低和漏检率高的问题,提出一种基于两阶段多模态注意力机制与动态特征解耦的轻量化输电线路缺陷检测方法。在第一阶段,基于改进型轻量检测网络Light-YOLO实现关键组件的精准定位;在第二阶段,构建基于双分支对比学习的缺陷检测网络Dual-DifferNet实现缺陷的精确分类与识别。在Light-YOLO的设计中,引入分层可分离视觉Transformer(SepViT)与深度可变形卷积网络(DCN)的混合结构,并通过交替堆叠局部感知卷积层与全局注意力Transformer块,在降低计算量的同时,增强模型对长程依赖关系的建模能力,从而有效提升绝缘子和导线接头等小目标的检测精度。针对缺陷分类任务,Dual-DifferNet采用双分支结构在每个分支中嵌入空间-通道双重注意力(SCDA)模块,利用交叉注意力机制促进双模态特征交互,从而提高缺陷识别的鲁棒性与泛化能力。实验结果表明,所提方法的平均精度均值(mAP@50)达到了96.9%,较基准模型YOLOv8提升16.1个百分点,同时浮点运算量降低了56.73%,充分验证了该方法在保证高精度检测的同时,具备优异的计算效率与部署潜力。
中图分类号:
黄萍, 李清, 邱海枫, 王程斯, 黄安子, 樊龙. 轻量化输电线路缺陷检测方法[J]. 计算机应用, 2026, 46(3): 969-979.
Ping HUANG, Qing LI, Haifeng QIU, Chengsi WANG, Anzi HUANG, Long FAN. Lightweight method for transmission line defect detection[J]. Journal of Computer Applications, 2026, 46(3): 969-979.
| 模型 | 精确率/% | 召回率/% | mAP@50/% | mAP@[50:95]/% | 浮点运算量/GFLOPs | 参数量/106 |
|---|---|---|---|---|---|---|
| Faster R-CNN[ | 56.6 | 53.4 | 59.2 | 51.5 | 134.38 | 41.76 |
| SSD[ | 67.2 | 64.6 | 70.1 | 62.8 | 34.86 | 35.64 |
| I2D-Net[ | 89.6 | 86.1 | 83.1 | 71.2 | 79.50 | 87.50 |
| YOLOv3[ | 88.7 | 89.1 | 91.6 | 74.3 | 283.10 | 103.70 |
| YOLOv5 | 91.5 | 84.9 | 90.6 | 72.3 | 64.40 | 25.10 |
| YOLOv8[ | 80.4 | 76.5 | 80.8 | 64.7 | 69.10 | 24.90 |
| YOLOv10[ | 88.9 | 86.4 | 90.6 | 75.1 | 64.10 | 16.50 |
| YOLOv11n[ | 88.9 | 86.8 | 89.5 | 70.0 | 6.50 | 2.59 |
| YOLOv11s | 88.0 | 88.7 | 90.7 | 67.8 | 21.60 | 9.43 |
| YOLOv11m | 88.9 | 87.5 | 90.6 | 75.1 | 68.00 | 16.11 |
| YOLOv11l | 89.7 | 89.0 | 92.1 | 77.0 | 86.70 | 25.29 |
| YOLOv11x | 90.1 | 89.5 | 92.6 | 77.3 | 195.60 | 56.89 |
| Light-YOLOx | 98.2 | 95.2 | 97.4 | 87.6 | 120.50 | 27.00 |
| Light-YOLO | 97.4 | 91.5 | 96.9 | 85.3 | 29.90 | 11.24 |
表1 不同模型的输电线路组件检测性能的对比
Tab. 1 Comparison of detection performance of transmission line components by different models
| 模型 | 精确率/% | 召回率/% | mAP@50/% | mAP@[50:95]/% | 浮点运算量/GFLOPs | 参数量/106 |
|---|---|---|---|---|---|---|
| Faster R-CNN[ | 56.6 | 53.4 | 59.2 | 51.5 | 134.38 | 41.76 |
| SSD[ | 67.2 | 64.6 | 70.1 | 62.8 | 34.86 | 35.64 |
| I2D-Net[ | 89.6 | 86.1 | 83.1 | 71.2 | 79.50 | 87.50 |
| YOLOv3[ | 88.7 | 89.1 | 91.6 | 74.3 | 283.10 | 103.70 |
| YOLOv5 | 91.5 | 84.9 | 90.6 | 72.3 | 64.40 | 25.10 |
| YOLOv8[ | 80.4 | 76.5 | 80.8 | 64.7 | 69.10 | 24.90 |
| YOLOv10[ | 88.9 | 86.4 | 90.6 | 75.1 | 64.10 | 16.50 |
| YOLOv11n[ | 88.9 | 86.8 | 89.5 | 70.0 | 6.50 | 2.59 |
| YOLOv11s | 88.0 | 88.7 | 90.7 | 67.8 | 21.60 | 9.43 |
| YOLOv11m | 88.9 | 87.5 | 90.6 | 75.1 | 68.00 | 16.11 |
| YOLOv11l | 89.7 | 89.0 | 92.1 | 77.0 | 86.70 | 25.29 |
| YOLOv11x | 90.1 | 89.5 | 92.6 | 77.3 | 195.60 | 56.89 |
| Light-YOLOx | 98.2 | 95.2 | 97.4 | 87.6 | 120.50 | 27.00 |
| Light-YOLO | 97.4 | 91.5 | 96.9 | 85.3 | 29.90 | 11.24 |
| 实验组 | 模型配置 | mAP@50/% | 帧率/(frame·s-1) |
|---|---|---|---|
| Baseline | YOLOv8 | 80.8 | 118 |
| +SepViT | 替换C2f为SepViT | 87.2 | 105 |
| +DCN | 添加可变形卷积 | 92.1 | 103 |
| +BR-FPN | 替换FPN为BR-FPN | 96.9 | 98 |
表2 Light-YOLO模型中各模块的有效性
Tab. 2 Effectiveness of each module in Light-YOLO model
| 实验组 | 模型配置 | mAP@50/% | 帧率/(frame·s-1) |
|---|---|---|---|
| Baseline | YOLOv8 | 80.8 | 118 |
| +SepViT | 替换C2f为SepViT | 87.2 | 105 |
| +DCN | 添加可变形卷积 | 92.1 | 103 |
| +BR-FPN | 替换FPN为BR-FPN | 96.9 | 98 |
| +SepViT@-1 | +SepViT@-2 | +SepViT@-3 | 浮点运算量/GFLOPs | mAP@50/% |
|---|---|---|---|---|
| √ | 23.5 | 87.4 | ||
| √ | √ | 26.7 | 89.3 | |
| √ | √ | √ | 29.7 | 91.5 |
表3 消融实验结果
Tab. 3 Ablation experiment results
| +SepViT@-1 | +SepViT@-2 | +SepViT@-3 | 浮点运算量/GFLOPs | mAP@50/% |
|---|---|---|---|---|
| √ | 23.5 | 87.4 | ||
| √ | √ | 26.7 | 89.3 | |
| √ | √ | √ | 29.7 | 91.5 |
| mAP@50/% | 浮点运算量/GFLOPs | 参数量/106 | |
|---|---|---|---|
| 0.0 | 96.9 | 29.9 | 11.24 |
| 0.1 | 96.5 | 26.9 | 10.12 |
| 0.2 | 96.0 | 22.4 | 9.80 |
| 0.3 | 90.2 | 19.8 | 9.20 |
| 0.4 | 89.1 | 18.1 | 8.60 |
表4 通道稀疏阈值对模型性能的影响
Tab. 4 Impact of channel sparsity threshold on model performance
| mAP@50/% | 浮点运算量/GFLOPs | 参数量/106 | |
|---|---|---|---|
| 0.0 | 96.9 | 29.9 | 11.24 |
| 0.1 | 96.5 | 26.9 | 10.12 |
| 0.2 | 96.0 | 22.4 | 9.80 |
| 0.3 | 90.2 | 19.8 | 9.20 |
| 0.4 | 89.1 | 18.1 | 8.60 |
| 类别 | DifferNet | Dual-DifferNet |
|---|---|---|
| 平均 | 88.2 | 93.5 |
| 架线悬挂 | 91.7 | 95.1 |
| 玻璃绝缘子 | 82.8 | 92.4 |
| 避雷针悬挂 | 93.0 | 97.0 |
| 聚合物绝缘子上拉环 | 86.4 | 89.0 |
| 可变张力夹 | 86.2 | 94.0 |
表5 在不同输电线路组件上AUROC结果对比 (%)
Tab. 5 Comparison of AUROC results on different transmission line components
| 类别 | DifferNet | Dual-DifferNet |
|---|---|---|
| 平均 | 88.2 | 93.5 |
| 架线悬挂 | 91.7 | 95.1 |
| 玻璃绝缘子 | 82.8 | 92.4 |
| 避雷针悬挂 | 93.0 | 97.0 |
| 聚合物绝缘子上拉环 | 86.4 | 89.0 |
| 可变张力夹 | 86.2 | 94.0 |
| 方法 | Jetson NX | 无人机 | |
|---|---|---|---|
推理速度/ (frame·s-1) | 吞吐量/ (frame·min-1) | 续航/min | |
| 基准方法 | 58 | 120 | 30 |
| 本文方法 | 82 | 159 | 33.0~34.5 |
表6 在边缘端部署下的性能比较
Tab. 6 Performance comparison under edge deployment
| 方法 | Jetson NX | 无人机 | |
|---|---|---|---|
推理速度/ (frame·s-1) | 吞吐量/ (frame·min-1) | 续航/min | |
| 基准方法 | 58 | 120 | 30 |
| 本文方法 | 82 | 159 | 33.0~34.5 |
| [1] | SZEGEDY C, TOSHEV A, ERHAN D. Deep neural networks for object detection [C]// Proceedings of the 27th International Conference on Neural Information Processing Systems — Volume 2. Red Hook: Curran Associates Inc., 2013: 2553-2561. |
| [2] | FAN L, ZHANG T, DU W. Optical-flow-based framework to boost video object detection performance with object enhancement [J]. Expert Systems with Applications, 2021, 170: No.114544. |
| [3] | 周立明,李士心,朱致仁,等. 基于深度学习的输电线路缺陷检测研究综述[J]. 计算机科学与应用, 2023, 13(6): 1235-1243. |
| ZHOU L M, LI S X, ZHU Z R, et al. A review of transmission line defect detection research based on deep learning [J]. Computer Science and Application, 2023, 13(6): 1235-1243. | |
| [4] | 郝然,王红军,李天瑞. 基于双分支串行混合注意力的输电线路缺陷检测深度神经网络模型[J]. 计算机科学, 2024, 51(3): 135-140. |
| HAO R, WANG H J, LI T R, Deep neural network model for transmission line defect detection based on dual-branch sequential mixed attention [J]. Computer Science, 2024, 51(3): 135-140. | |
| [5] | 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. |
| [6] | ZHAO Z, QI H, QI Y, et al. Detection method based on automatic visual shape clustering for pin-missing defect in transmission lines[J]. IEEE Transactions on Instrumentation and Measurement, 2020, 69(9): 6080-6091. |
| [7] | FU Q, LIU J, ZHANG X, et al. A small-sized defect detection method for overhead transmission lines based on convolutional neural networks [J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: No.3524612. |
| [8] | 米增,连哲. 面向通用目标检测的YOLO方法研究综述[J]. 计算机工程与应用, 2024, 60(21): 38-54. |
| MI Z, LIAN Z. Review of YOLO methods for universal object detection [J]. Computer Engineering and Applications, 2024, 60(21): 38-54. | |
| [9] | 李洵,甘润东,钱俊凤,等. 改进YOLOv5混合样本训练的绝缘子伞盘脱落缺陷检测方法[J]. 计算机工程与应用, 2024, 60(4): 289-297. |
| LI X, GAN R D, QIAN J F, et al. Improved YOLOv5 mixed sample training for detection of insulator umbrella plate falling defects [J]. Computer Engineering and Applications, 2024, 60(4): 289-297. | |
| [10] | SREE P B, BALAJI K, BANOTH N, et al. A lightweight mobile net SSD algorithm-based identification and detection of multiple defects in ceramic insulators [J]. Journal of Modern Technology, 2024, 1(1): 59-74. |
| [11] | BAO W, DU X, WANG N, et al. A defect detection method based on BC-YOLO for transmission line components in UAV remote sensing images [J]. Remote Sensing, 2022, 14(20): No.5176. |
| [12] | SHI C, LIN L, SUN J, et al. A lightweight YOLOv5 transmission line defect detection method based on coordinate attention[C]// Proceedings of the IEEE 6th Information Technology and Mechatronics Engineering Conference. Piscataway: IEEE, 2022: 1779-1785. |
| [13] | 孙阳,李佳. 基于通道剪枝的YOLOv7-tiny输电线路异物检测算法[J]. 计算机工程与应用, 2024, 60(14): 319-328. |
| SUN Y, LI J. YOLOv7-tiny transmission line foreign object detection algorithm based on channel pruning [J]. Computer Engineering and Applications, 2024, 60(14): 319-328. | |
| [14] | REIS D, KUPEC J, HONG J, et al. Real-time flying object detection with YOLOv8 [EB/OL]. [2024-10-02].. |
| [15] | LI W, WANG X, XIA X, et al. SepViT: separable Vision Transformer [EB/OL]. [2024-01-03].. |
| [16] | WANG R, SHIVANNA R, CHENG D, et al. DCN V2: improved deep & cross network and practical lessons for web-scale learning to rank systems [C]// Proceedings of the Web Conference 2021. New York: ACM, 2021: 1785-1797. |
| [17] | ANNAVARAPU C S R, PARISAPOGU, S A B, KEETHA N V, et al. A Bi-FPN-based encoder-decoder model for lung nodule image segmentation [J]. Diagnostics, 2023, 13(8): No.1406. |
| [18] | DEY R, SALEM F M. Gate-variants of Gated Recurrent Unit (GRU) neural networks [C]// Proceedings of the IEEE 60th International Midwest Symposium on Circuits and Systems. Piscataway: IEEE, 2017: 1597-1600. |
| [19] | LOMENICK B, HAO R, JONAI N, et al. Target identification using drug affinity responsive target stability [J]. Proceedings of the National Academy of Sciences of the United States of America, 2009, 106(51): 21984-21989. |
| [20] | RUDOLPH M, WANDT B, ROSENHAHN B. Same same but DifferNet: semi-supervised defect detection with normalizing flows[C]// Proceedings of the 2021 IEEE Winter Conference on Applications of Computer Vision. Piscataway: IEEE, 2021: 1906-1915. |
| [21] | SONG C H, HAN H J, AVRITHIS Y. All the attention you need: global-local, spatial-channel attention for image retrieval [C]// Proceedings of the 2022 IEEE/CVF Winter Conference on Applications of Computer Vision. Piscataway: IEEE, 2022: 439-448. |
| [22] | XI E, BING S, JIN Y. Capsule network performance on complex data [EB/OL]. [2024-09-02].. |
| [23] | MAZZIA V, SALVETTI F, CHIABERGE M. Efficient-CAPSNet: capsule network with self-attention routing [J]. Scientific Reports, 2021, 11: No.14634. |
| [24] | BUARQUE VIEIRA E SILVA A L, DE CASTRO FELIX H, SIMÕES F P M, et al. InsPLAD: a dataset and benchmark for power line asset inspection in UAV images [J]. International Journal of Remote Sensing, 2023, 44(23): 7294-7320. |
| [25] | REN S Q, 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. |
| [26] | LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector [C]// Proceedings of the 2016 European Conference on Computer Vision, LNCS 9905. Cham: Springer, 2016: 21-37. |
| [27] | REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. [2024-11-02].. |
| [28] | WANG A, CHEN H, LIU L, et al. YOLOv10: real-time end-to-end object detection [C]// Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 107984-108011. |
| [29] | KHANAM R, HUSSAIN M. YOLOv11: an overview of the key architectural enhancements [EB/OL]. [2025-01-14].. |
| [1] | 王璐, 刘东, 刘卫光. 可变形卷积网络的解释性研究及其在蝴蝶物种识别模型中的应用[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 261-274. |
| [2] | 赵志强, 马培红, 黑新宏. 基于双重注意力机制的人群计数方法[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2886-2892. |
| [3] | 张奕, 蔡钢生, 王真梅. 基于语义与全局双重注意力机制的长链非编码RNA-疾病关联预测模型[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2125-2132. |
| [4] | 姬张建, 张明, 王子龙. 基于改进VarifocalNet的高精度目标检测算法[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2147-2154. |
| [5] | 顾勇翔, 蓝鑫, 伏博毅, 秦小林. 基于几何适应与全局感知的遥感图像目标检测算法[J]. 《计算机应用》唯一官方网站, 2023, 43(3): 916-922. |
| [6] | 郭子豪, 董乐乐, 曲志坚. 基于改进Faster RCNN的节肢动物目标检测方法[J]. 《计算机应用》唯一官方网站, 2023, 43(1): 88-97. |
| [7] | 庄屹, 赵海涛. 面向三维点云单目标跟踪的提案聚合网络[J]. 《计算机应用》唯一官方网站, 2022, 42(5): 1407-1416. |
| [8] | 王品学, 张绍兵, 成苗, 何莲, 秦小山. 基于可变形卷积和自适应空间特征融合的硬币表面缺陷检测算法[J]. 《计算机应用》唯一官方网站, 2022, 42(2): 638-645. |
| [9] | 陈静, 毛莺池, 陈豪, 王龙宝, 王子成. 基于改进单点多盒检测器的大坝缺陷目标检测方法[J]. 计算机应用, 2021, 41(8): 2366-2372. |
| [10] | 边小勇, 江沛龄, 赵敏, 丁胜, 张晓龙. 基于多分支神经网络模型的弱监督细粒度图像分类方法[J]. 计算机应用, 2020, 40(5): 1295-1300. |
| 阅读次数 | ||||||
|
全文 |
|
|||||
|
摘要 |
|
|||||