基于特征聚焦扩散网络的电力巡检目标检测算法

doi:10.11996/JG.j.2095-302X.2025040719

图学学报 ›› 2025, Vol. 46 ›› Issue (4): 719-726.DOI: 10.11996/JG.j.2095-302X.2025040719

• 图像处理与计算机视觉 • 上一篇下一篇

基于特征聚焦扩散网络的电力巡检目标检测算法

郭瑞东¹(), 蓝贵文¹^,²(), 范冬林¹, 钟展¹, 徐梓睿¹, 任新月¹

1.桂林理工大学测绘地理信息学院，广西桂林 541006
2.广西空间信息与测绘重点实验室，广西桂林 541006

收稿日期:2024-10-09 修回日期:2025-01-07 出版日期:2025-08-30 发布日期:2025-08-11
通讯作者:蓝贵文(1977-)，男，教授，博士。主要研究方向为地理信息系统、地理空间智能等。E-mail：2009043@glut.edu.cn
第一作者:郭瑞东(1995-)，男，硕士研究生。主要研究方向为图像处理、计算机视觉等。E-mail：13028689662@163.com
基金资助:
国家自然科学基金(41861050);广西自然科学基金(2022GXNSFBA035637)

An object detection algorithm for powerline inspection based on the feature focus & diffusion network

GUO Ruidong¹(), LAN Guiwen¹^,²(), FAN Donglin¹, ZHONG Zhan¹, XU Zirui¹, REN Xinyue¹

1. College of Geomatics and Geoinformation, Guilin University of Technology, Guilin Guangxi 541006, China
2. Guangxi Key Laboratory of Spatial Information and Geomatics, Guilin Guangxi 541006, China

Received:2024-10-09 Revised:2025-01-07 Published:2025-08-30 Online:2025-08-11
First author：GUO Ruidong (1995-), master student. His main research interests cover image processing, computer vision, etc. E-mail：13028689662@163.com
Supported by:
National Natural Science Foundation of China(41861050);Guangxi Natural Science Foundation(2022GXNSFBA035637)

摘要/Abstract

摘要：

针对无人机电力巡检航拍图像通常背景复杂、待检目标尺寸较小，使用通用的特征提取网络往往导致漏检、误检率较高，提出了一种用于特征融合的特征聚焦扩散网络(FFDN)，利用FFDN对YOLOv8算法进行改进，设计了基于YOLOv8的电力巡检目标检测算法YOLOv8-SFD。在主干网络采用空间到深度非跨步卷积(SPDConv)保留更多小尺度特征，减少跨步卷积造成的特征损失，避免小目标以及遮挡目标特征提取过程中的特征丢失；在特征融合阶段使用FFDN代替传统的特征金字塔网络，利用特征聚焦模块融合多尺度特征并扩大感受野，将特征聚焦模块输出的多尺度特征图扩散传递至不同尺度，提升小目标的检测精度；将原始YOLOv8的头部替换为融合了尺度、空间和任务3种注意力机制的动态检测头(DyHead)，进一步提高模型检测性能。实验结果表明，YOLOv8-SFD准确率达到76.7%，召回率达到43.0%，平均精确率均值达到48.2%，分别比YOLOv8n提高了7.6%，2.0%和3.8%。YOLOv8-SFD有效提升了小目标和遮挡目标的检测精度，且检测速度达到119帧/秒，满足实时检测的需要。

关键词: 电力巡检, 特征融合, 目标检测, 特征聚焦扩散网络, YOLOv8

Abstract:

UAV images for powerline inspection usually have complex backgrounds, and often contain a lot of small targets, which may lead to a high rate of missed detections and false detections when processed by the general feature extraction networks for object detection. To address this, a feature focus & diffusion network (FFDN) was proposed for feature fusion, and an improved algorithm (YOLOv8-SFD) based on FFDN and YOLOv8 was designed for powerline component detection. Spatial-to-depth non-stride convolutions (SPDConv) were employed in the backbone network instead to preserve small-scale features and reduce feature loss caused by stride convolutions. The traditional feature pyramid network was replaced with the proposed FFDN. At the feature fusion stage, the feature focus modules in the FFDN were utilized to expand the receptive field and fuse multi-scale features, and the output feature maps by them were then diffused across different scales to enhance small target detection accuracy. Finally, the original YOLOv8 head was replaced with a dynamic detection head (DyHead) that integrates three attention mechanisms (scale, space, and task), to further enhance the performance. Experimental results demonstrated that YOLOv8-SFD achieved an accuracy rate of 76.7%, which was 7.6% higher than YOLOv8n; a recall rate of 43.0%, which was 2.0% higher; and a MAP of 48.2%, which was 3.8%. YOLOv8-SFD effectively enhanced the detection precision for small and obscured targets, and the detection speed reached 119 FPS, which satisfied real-time detection requirements.

Key words: powerline inspection, feature fusion, object detection, feature focus & diffusion network, YOLOv8

中图分类号:

郭瑞东, 蓝贵文, 范冬林, 钟展, 徐梓睿, 任新月. 基于特征聚焦扩散网络的电力巡检目标检测算法[J]. 图学学报, 2025, 46(4): 719-726.

GUO Ruidong, LAN Guiwen, FAN Donglin, ZHONG Zhan, XU Zirui, REN Xinyue. An object detection algorithm for powerline inspection based on the feature focus & diffusion network[J]. Journal of Graphics, 2025, 46(4): 719-726.

图/表 9

参考文献 25

[1]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2014: 580-587.
[2]	GIRSHICK R. Fast R-CNN[C]// 2015 IEEE International Conference on Computer Vision. New York: IEEE Press, 2015: 1440-1448.
[3]	REN S Q, HE K M, 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. DOI PMID
[4]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 779-788.
[5]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]// The 14th European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[6]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(2): 318-327.
[7]	冯珺, 潘司晨, 赵帅, 等. 基于改进RPN的孪生小样本电力目标检测[J]. 河北科技大学学报, 2023, 44(1): 67-73.
	FENG J, PAN S C, ZHAO S, et al. Research on few-shot power detection of Siamese network based on improved RPN[J]. Journal of Hebei University of Science and Technology, 2023, 44(1): 67-73 (in Chinese).
[8]	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(9): 3089-3096.
	GU C Y, LI Z, SHI J T, et al. Detection for pin defects of overhead lines by UAV patrol image based on improved faster-RCNN[J]. High Voltage Engineering, 2020, 46(9): 3089-3096 (in Chinese).
[9]	黄芹芹, 董洁, 陈玥, 等. 一种改进SSD算法的输电线路目标检测方法[J]. 电工电气, 2021(6): 51-55.
	HUANG Q Q, DONG J, CHEN Y, et al. A transmission line target detection method with improved SSD algorithm[J]. Electrotechnics Electric, 2021(6): 51-55 (in Chinese).
[10]	郝帅, 赵新生, 马旭, 等. 基于TR-YOLOv5的输电线路多类缺陷目标检测方法[J]. 图学学报, 2023, 44(4): 667-676. DOI
	HAO S, ZHAO X S, MA X, et al. Multi-class defect target detection method for transmission lines based on TR-YOLOv5[J]. Journal of Graphics, 2023, 44(4): 667-676 (in Chinese). DOI
[11]	李利霞, 王鑫, 王军, 等. 基于特征融合与注意力机制的无人机图像小目标检测算法[J]. 图学学报, 2023, 44(4): 658-666. DOI
	LI L X, WANG X, WANG J, et al. Small object detection algorithm in UAV image based on feature fusion and attention mechanism[J]. Journal of Graphics, 2023, 44(4): 658-666 (in Chinese).
[12]	苏凯第, 赵巧娥. 基于YOLOv5算法的无人机电力巡检快速图像识别[J]. 电力科学与工程, 2022, 38(4): 43-48. DOI
	SU K D, ZHAO Q E. Fast image recognition of UAV power inspection based on YOLOv5 algorithm[J]. Electric Power Science and Engineering, 2022, 38(4): 43-48 (in Chinese). DOI
[13]	奉志强, 谢志军, 包正伟, 等. 基于改进YOLOv5的无人机实时密集小目标检测算法[J]. 航空学报, 2023, 44(7): 327106. DOI
	FENG Z Q, XIE Z J, BAO Z W, et al. Real-time dense small object detection algorithm for UAV based on improved YOLOv5[J]. Acta Aeronautica et Astronautica Sinica, 2023, 44(7): 327106 (in Chinese).
[14]	冯辉, 蒋成鑫, 徐海祥, 等. 基于多特征聚合的水面遮挡目标检测算法[J]. 华中科技大学学报(自然科学版), 2024, 52(4): 76-81.
	FENG H, JIANG C X, XU H X, et al. Multi feature fusion-based water occlusion object detection algorithm[J]. Journal of Huazhong University of Science and Technology (Natural Science Edition), 2024, 52(4): 76-81 (in Chinese).
[15]	HUANG H Q, LAN G W, WEI J, et al. TLI-YOLOv5: a lightweight object detection framework for transmission line inspection by unmanned aerial vehicle[J]. Electronics, 2023, 12(15): 3340.
[16]	翟永杰, 郭聪彬, 王乾铭, 等. 基于隐含空间知识融合的输电线路多金具检测方法[J]. 图学学报, 2023, 44(5): 918-927. DOI
	ZHAI Y J, GUO C B, WANG Q M, et al. Multi-fitting detection method for transmission lines based on implicit spatial knowledge fusion[J]. Journal of Graphics, 2023, 44(5): 918-927 (in Chinese).
[17]	冯欣, 胡成杭. 一种自监督掩码图像建模的遮挡目标检测方法[J]. 重庆理工大学学报(自然科学), 2024, 38(6): 186-193.
	FENG X, HU C H. An occlusion object detection method based on self-supervised mask image modeling[J]. Journal of Chongqing University of Technology (Natural Science), 2024, 38(6): 186-193 (in Chinese).
[18]	ZHAO Z B, PAN Y T, GUO G X, et al. YOLO‐AFPN: marrying YOLO and AFPN for external damage detection of transmission lines[J]. IET Generation, Transmission & Distribution, 2024, 18(9): 1935-1946.
[19]	YANG G Y, LEI J, ZHU Z K, et al. AFPN: asymptotic feature pyramid network for object detection[C]// 2023 IEEE International Conference on Systems, Man, and Cybernetics. New York: IEEE Press, 2023: 2184-2189.
[20]	DING X H, ZHANG X Y, ZHOU Y G, et al. Scaling up your kernels to 31×31: revisiting large kernel design in CNNs[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 11953-11965.
[21]	WANG C Y, YEH I H, LIAO H Y M. YOLOv9:learning what you want to learn using programmable gradient information[C]// The 18th European Conference on Computer Vision. Cham: Springer, 2024: 1-21.
[22]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 7464-7475.
[23]	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 8759-8768.
[24]	SUNKARA R, LUO T. No more strided convolutions or pooling: a new CNN building block for low-resolution images and small objects[C]// European Conference on Machine Learning and Knowledge Discovery in Databases. Cham: Springer, 2022: 443-459.
[25]	DAI X Y, CHEN Y P, XIAO B, et al. Dynamic head: unifying object detection heads with attentions[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 7369-7378.

参数	具体配置
操作系统	Windows 10专业版22H2
CPU	Intel(R)Core(TM)i5-12400
内存	16 GB
GPU	NVIDIA GeForce RTX 3070 Ti
CUDA	11.6
Pytorch	1.13.1
开发环境	Pycharm
编程语音	Python 3.8.10

参数	具体配置
操作系统	Windows 10专业版22H2
CPU	Intel(R)Core(TM)i5-12400
内存	16 GB
GPU	NVIDIA GeForce RTX 3070 Ti
CUDA	11.6
Pytorch	1.13.1
开发环境	Pycharm
编程语音	Python 3.8.10

模型	Precision/%	Recall/%	F1	mAP@0.50/%	mAP@0.50∶0.95/%	训练时间/h
YOLOv3-tiny	0.724	0.381	0.498	0.409	0.196	6.28
YOLOv5n	0.699	0.373	0.486	0.408	0.210	6.16
YOLOv6n	0.665	0.406	0.504	0.418	0.220	6.16
YOLOv8n	0.694	0.410	0.516	0.444	0.240	6.11

模型	Precision/%	Recall/%	F1	mAP@0.50/%	mAP@0.50∶0.95/%	训练时间/h
YOLOv3-tiny	0.724	0.381	0.498	0.409	0.196	6.28
YOLOv5n	0.699	0.373	0.486	0.408	0.210	6.16
YOLOv6n	0.665	0.406	0.504	0.418	0.220	6.16
YOLOv8n	0.694	0.410	0.516	0.444	0.240	6.11

颈部结构	Precision/%	Recall/%	F1	mAP@0.50/%	mAP@0.50∶0.95/%
BiFPN	0.642	0.405	0.496	0.428	0.218
EfficientRepBiPAN	0.663	0.413	0.508	0.412	0.215
HSFPN	0.678	0.394	0.498	0.422	0.220
ASF	0.720	0.399	0.514	0.438	0.225
GFPN	0.693	0.411	0.516	0.433	0.233
AFPN	0.733	0.382	0.502	0.422	0.225
SDI	0.683	0.209	0.320	0.265	0.131
FFDN	0.720	0.413	0.524	0.453	0.232

基于特征聚焦扩散网络的电力巡检目标检测算法

An object detection algorithm for powerline inspection based on the feature focus & diffusion network

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 25

相关文章 15

编辑推荐

Metrics

本文评价

颈部网络结构	t
颈部网络结构	20%	30%	50%	99%
PAFPN	0.038	0.067	0.172	0.968
FFDN	0.064	0.112	0.228	0.968

[1]	闫卓越, 刘骊, 付晓东, 刘利军, 彭玮. 三维人体姿态和形状估计的分层注意力时空特征融合算法[J]. 图学学报, 2025, 46(4): 746-755.
[2]	廖国琼, 黄龙杰, 李清新, 辜勇, 李海波. 面向单目可见光环境的自适应双手重建网络[J]. 图学学报, 2025, 46(4): 837-846.
[3]	张立立, 杨康, 张珂, 魏薇, 李晶, 谭洪鑫, 张翔宇. 面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究[J]. 图学学报, 2025, 46(2): 249-258.
[4]	翟永杰, 王璐瑶, 赵晓瑜, 胡哲东, 王乾铭, 王亚茹. 基于级联查询-位置关系的输电线路多金具检测方法[J]. 图学学报, 2025, 46(2): 288-299.
[5]	赵振兵, 韩钰, 唐辰康. 基于改进YOLOv8的配电线路绝缘子缺陷级联检测方法[J]. 图学学报, 2025, 46(1): 1-12.
[6]	董佳乐, 邓正杰, 李喜艳, 王诗韵. 基于频域和空域多特征融合的深度伪造检测方法[J]. 图学学报, 2025, 46(1): 104-113.
[7]	程旭东, 史彩娟, 高炜翔, 王森, 段昌钰, 闫晓东. 面向域自适应目标检测的一致无偏教师模型[J]. 图学学报, 2025, 46(1): 114-125.
[8]	崔克彬, 耿佳昌. 基于EE-YOLOv8s的多场景火灾迹象检测算法[J]. 图学学报, 2025, 46(1): 13-27.
[9]	王志东, 陈晨阳, 刘晓明. 基于轻量化改进YOLOv8的通信光缆缺陷检测[J]. 图学学报, 2025, 46(1): 28-34.
[10]	苑朝, 赵明雪, 张丰羿, 冯晓勇, 李冰, 陈瑞. 基于点云特征增强的复杂室内场景3D目标检测[J]. 图学学报, 2025, 46(1): 59-69.
[11]	王杨, 马唱, 胡明, 孙涛, 饶元, 袁振羽. 基于多尺度特征融合的轻量型野外蝙蝠检测[J]. 图学学报, 2025, 46(1): 70-80.
[12]	孙前来, 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍. 基于元学习的小样本指针式仪表检测方法[J]. 图学学报, 2025, 46(1): 81-93.
[13]	卢洋, 陈林慧, 姜晓恒, 徐明亮. SDENet：基于多尺度注意力质量感知的合成缺陷数据评价网络[J]. 图学学报, 2025, 46(1): 94-103.
[14]	李琼, 考月英, 张莹, 徐沛. 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[15]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.

FFDN	SPDConv	Dyhead	Precision/%	Recall/%	mAP@0.50/%	mAP@0.50∶0.95/%	F1	FPS
			0.694	0.410	0.444	0.240	0.516	149
	√		0.738	0.420	0.463	0.236	0.534	188
		√	0.739	0.420	0.463	0.249	0.536	156
√			0.736	0.408	0.446	0.230	0.524	132
	√	√	0.732	0.438	0.478	0.259	0.548	136
√	√		0.711	0.431	0.460	0.242	0.536	135
√		√	0.717	0.422	0.470	0.243	0.532	119
√	√	√	0.767	0.430	0.482	0.254	0.552	119

FFDN	SPDConv	Dyhead	Precision/%	Recall/%	mAP@0.50/%	mAP@0.50∶0.95/%	F1	FPS
			0.694	0.410	0.444	0.240	0.516	149
	√		0.738	0.420	0.463	0.236	0.534	188
		√	0.739	0.420	0.463	0.249	0.536	156
√			0.736	0.408	0.446	0.230	0.524	132
	√	√	0.732	0.438	0.478	0.259	0.548	136
√	√		0.711	0.431	0.460	0.242	0.536	135
√		√	0.717	0.422	0.470	0.243	0.532	119
√	√	√	0.767	0.430	0.482	0.254	0.552	119