基于窗口自注意力网络与YOLOv5融合的输电线路通道异物检测

doi:10.16183/j.cnki.jsjtu.2023.301

摘要/Abstract

摘要：

针对输电线路通道异物检测背景复杂以及小目标情况下检测效果不佳等问题,提出一种基于窗口自注意力网络与YOLOv5模型融合的输电线路通道安全检测算法.首先,选用窗口自注意力(S-T)网络优化主干网络,扩大模型感受视野,增强提取有效信息的能力.其次,改进自适应空间特征融合(ASFF)模块,增强多尺度特征融合能力.最后,考虑到真实框与预测框不匹配的问题,引入结构相似性交并比(SIoU),优化边界误差,提高小目标定位准确性.实验结果表明,本文模型对线路通道多目标入侵检测精度达到90.2%,且提升了小目标检测效果;与主流目标检测算法相比,可以更好地满足输电线路通道中的异物检测需求.

关键词: 智能化巡检, 输电线路通道, 目标检测, 窗口自注意力网络, 自适应空间特征融合

Abstract:

To address the challenges of complex detection background and poor detection performance for small targets, a transmission line channel security detection algorithm based on the fusion of window self-attention network and the YOLOv5 model is proposed. First, the Swin Transformer (S-T) is employed to optimize the backbone network, expanding the perception field of the model and enhancing its ability to extract effective information. Then, the adaptive spatial feature fusion (ASFF) module is improved to enhance the feature fusion ability of the model. Finally, considering the mismatch between the real frame and the predicted frame, the structural similarity intersection over union (SIoU) is introduced to optimize the boundary errors and improve the generalization ability of the model. The experimental results show that the model proposed achieves a multi-target intrusion detection accuracy of 90.2%, and with significant improvements in the detection of small targets. This approach better meets the requirements of foreign object detection in transmission line channels compared to other object detection algorithms.

Key words: intelligent inspection, transmission line channel, object detection, Swin Transformer (S-T), adaptive spatial feature fusion (ASFF)

中图分类号:

TM721

薛昂, 姜恩宇, 张文涛, 林顺富, 米阳. 基于窗口自注意力网络与YOLOv5融合的输电线路通道异物检测[J]. 上海交通大学学报, 2025, 59(3): 413-423.

XUE Ang, JIANG Enyu, ZHANG Wentao, LIN Shunfu, MI Yang. Detection of Foreign Bodies in Transmission Line Channels Based on Fusion of Swin Transformer and YOLOv5[J]. Journal of Shanghai Jiao Tong University, 2025, 59(3): 413-423.

图/表 17

图1

图2

图3

图4

图5

图6

图7

图8

图9

表1

图10

图11

图12

图13

表2

表3

图14

参考文献 25

[1]	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]. 中国电机工程学报, 2023, 43(19): 7423-7446.
	LIU Chuanyang, WU Yiquan. Research progress of vision detection methods based on deep learning for transmission lines[J]. Proceedings of the CSEE, 2023, 43(19): 7423-7446.
[2]	ZHAO Z Q, ZHENG P, XU S T, et al. Object detection with deep learning: A review[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(11): 3212-3232.
[3]	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(9): 3636-3648.
	SUI Yu, NING Pingfan, NIU Pingjuan, et al. Review on mounted UAV for transmission line inspection[J]. Power System Technology, 2021, 45(9): 3636-3648.
[4]	CHEN M H, TIAN Y N, XING S Y, et al. Environment perception technologies for power transmission line inspection robots[J]. Journal of Sensors, 2021, 2021: 5559231.
[5]	XIANG X Z, LV N, GUO X L, et al. Engineering vehicles detection based on modified faster R-CNN for power grid surveillance[J]. Sensors, 2018, 18(7): 2258.
[6]	ZHU J G, GUO Y, YUE F D, et al. A deep learning method to detect foreign objects for inspecting power transmission lines[J]. IEEE Access, 2020, 8: 94065-94075.
[7]	LEI X S, SUI Z H. Intelligent fault detection of high voltage line based on the Faster R-CNN[J]. Measurement, 2019, 138: 379-385.
[8]	魏业文, 李梅, 解园琳, 等. 基于改进Faster-RCNN的输电线路巡检图像检测[J]. 电力工程技术, 2022, 41(2): 171-178.
	WEI Yewen, LI Mei, XIE Yuanlin, et al. Transmission line inspection image detection based on improved Faster-RCNN[J]. Electric Power Engineering Technology, 2022, 41(2): 171-178.
[9]	XU C F, BO B, LIU Y, et al. Detection method of insulator based on single shot MultiBox detector[J]. Journal of Physics: Conference Series, 2018, 1069: 012183.
[10]	陈科圻, 朱志亮, 邓小明, 等. 多尺度目标检测的深度学习研究综述[J]. 软件学报, 2021, 32(4): 1201-1227.
	CHEN Keqi, ZHU Zhiliang, DENG Xiaoming, et al. Deep learning for multi-scale object detection: A survey[J]. Journal of Software, 2021, 32(4): 1201-1227.
[11]	XUE B, YI W J, JING F, et al. Complex ISAR target recognition using deep adaptive learning[J]. Engineering Applications of Artificial Intelligence, 2021, 97: 104025.
[12]	李振宇, 郭锐, 赖秋频, 等. 基于计算机视觉的架空输电线路机器人巡检技术综述[J]. 中国电力, 2018, 51(11): 139-146.
	LI Zhenyu, GUO Rui, LAI Qiupin, et al. Survey of inspection technology of overhead transmission line robot based on computer vision[J]. Electric Power, 2018, 51(11): 139-146.
[13]	邵延华, 张铎, 楚红雨, 等. 基于深度学习的YOLO目标检测综述[J]. 电子与信息学报, 2022, 44(10): 3697-3708.
	SHAO Yanhua, ZHANG Duo, CHU Hongyu, et al. A review of YOLO object detection based on deep learning[J]. Journal of Electronics & Information Technology, 2022, 44(10): 3697-3708.
[14]	张焕龙, 齐企业, 张杰, 等. 基于改进YOLOv5的输电线路鸟巢检测方法研究[J]. 电力系统保护与控制, 2023, 51(2): 151-159.
	ZHANG Huanlong, QI Qiye, ZHANG Jie, et al. Bird nest detection method for transmission lines based on improved YOLOv5[J]. Power System Protection and Control, 2023, 51(2): 151-159.
[15]	李登攀, 任晓明, 颜楠楠. 基于无人机航拍的绝缘子掉串实时检测研究[J]. 上海交通大学学报, 2022, 56(8): 994-1003. doi: 10.16183/j.cnki.jsjtu.2021.416
	LI Dengpan, REN Xiaoming, YAN Nannan. Real-time detection of insulator drop string based on UAV aerial photography[J]. Journal of Shanghai Jiao Tong University, 2022, 56(8): 994-1003.
[16]	谢静, 杜耀文, 刘志坚, 等. 基于轻量化改进型YOLOv5s的可见光绝缘子缺陷检测算法[J]. 电网技术, 2023, 47(12): 5273-5283.
	XIE Jing, DU Yaowen, LIU Zhijian, et al. Defect detection algorithm based on lightweight and improved YOLOv5s for visible light insulators[J]. Power System Technology, 2023, 47(12): 5273-5283.
[17]	GU J P, HU J J, JIANG L, et al. Research on object detection of overhead transmission lines based on optimized YOLOv5s[J]. Energies, 2023, 16(6): 2706.
[18]	游越, 伊力哈木·亚尔买买提. 基于改进YOLOv5在电力巡检中的目标检测算法研究[J]. 高压电器, 2023, 59(2): 89-96.
	YOU Yue, YILIHAMU Yaermaimaiti. Research on target detection algorithm based on improved YOLOv5 in power patrol inspection[J]. High Voltage Apparatus, 2023, 59(2): 89-96.
[19]	LIU Z, LIN Y, CAO Y, et al. Hierarchical vision Transformer using shifted windows[DB/OL]. (2021-08-17)[2023-04-12]. https://arxiv.org/abs/2103.1403.
[20]	高涛, 文渊博, 陈婷, 等. 基于窗口自注意力网络的单图像去雨算法[J]. 上海交通大学学报, 2023, 57(5): 613-623. doi: 10.16183/j.cnki.jsjtu.2022.032
	GAO Tao, WEN Yuanbo, CHEN Ting, et al. A single image deraining algorithm based on Swin Transformer[J]. Journal of Shanghai Jiao Tong University, 2023, 57(5): 613-623.
[21]	MA J Y, TANG L F, FAN F, et al. SwinFusion: Cross-domain long-range learning for general image fusion via swin transformer[J]. IEEE/CAA Journal of Automatica Sinica, 2022, 9(7): 1200-1217.
[22]	ZHANG C, WANG L J, CHENG S L, et al. SwinSUNet: Pure transformer network for remote sensing image change detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5224713.
[23]	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(6): 2319-2331.
	HAO Shuai, YANG Lei, MA Xu, et al. YOLOv5 transmission line fault detection based on attention mechanism and cross-scale feature fusion[J]. Proceedings of the CSEE, 2023, 43(6): 2319-2331.
[24]	XUE B, HE Y, JING F, et al. Dynamic coarse-to-fine ISAR image blind denoising using active joint prior learning[J]. International Journal of Intelligent Systems, 2021, 36(8): 4143-4166.
[25]	XUE B, HE Y, JING F, et al. Robot target recognition using deep federated learning[J]. International Journal of Intelligent Systems, 2021, 36(12): 7754-7769.

编辑推荐 0

Metrics

阅读次数

全文

428

HTML			PDF

最新录用	在线预览	正式出版	最新录用	在线预览	正式出版
0	0	5	390	0	33

来源	本网站	其他网站

次数	107	321
比例	25%	75%

摘要

1632

最新录用	在线预览	正式出版

196	0	1436

来源	本网站	其他网站

次数	91	1541
比例	6%	94%

配置名称	版本信息
系统	Windows11操作系统
处理器	Intel(R) Core(TM) i9-10900K
显卡	NVIDIA GeForce RTX 3070 8 GB内存
Cuda版本	11.3
CUDNN版本	8.2.1
Python版本	3.8.13
Torch版本	1.12.1
Torchvision版本	0.13.1
Numpy版本	1.23.1

名称	S-T	ASFF	SIOU	F₁/%	mAP@50%	FPS/(帧·s^-1)	模型大小/MB
YOLOv5				76.7	83.9	57	25.2
YOLOv5-S	√			81.1	87.5	52	26.1
YOLOv5-A		√		80.6	86.2	54	25.7
YOLOv5-SA	√	√		83.1	89.6	46	27.6
YOLOv5-SAS	√	√	√	84.4	90.2	48	27.6

检测方法	F₁/%	mAP/%	FPS/(帧·s^-1)	模型大小/MB
SSD	62.5	73.4	45	86.7
Fast R-CNN	69.3	77.2	15	207
YOLOv4	73.6	80.3	43	55.4
YOLOv5	76.7	83.9	57	25.2
YOLOv5-SAS	84.4	90.2	48	27.6