Adaptive feature fusion pyramid and attention mechanism-based method for transmission line insulator defect detection

doi:10.11996/JG.j.2095-302X.2025050950

Abstract

Abstract:

To address the challenges of complex background interference and varying defect region scales in transmission line insulator samples, a method for transmission line insulator defect detection based on an adaptive fusion feature pyramid and attention mechanism was proposed. First, an adaptive fusion module (AF) was introduced to process multi-scale feature information, which was integrated into the feature pyramid network to mitigate the inconsistencies of defect region scales in aerial images of insulators. Next, a defect feature refinement module (DFRM) based on an attention mechanism was designed to handle interference from complex background noise by expanding the receptive field and capturing the contextual features of defective regions. Finally, the improved algorithm was validated on a real-world transmission line insulator defect dataset. Experimental results demonstrated that the proposed method outperformed existing approaches in insulator defect detection, achieving a 5.7% improvement in accuracy compared to the baseline model. These findings offered an effective solution for intelligent inspection in power grid systems.

Key words: insulator defect, feature fusion, attention mechanism, object detection, multi-scale feature

CLC Number:

ZHAI Yongjie, ZHAI Bangchao, HU Zhedong, YANG Ke, WANG Qianming, ZHAO Xiaoyu. Adaptive feature fusion pyramid and attention mechanism-based method for transmission line insulator defect detection[J]. Journal of Graphics, 2025, 46(5): 950-959.

Figures/Tables 12

References 39

[1]	王亚茹, 冯利龙, 宋晓轲, 等. TFD-YOLOv8: 一种用于输电线路的异物检测方法[J]. 图学学报, 2024, 45(5): 901-912. DOI
	WANG Y R, FENG L L, SONG X K, et al. TFD-YOLOv8: a transmission line foreign object detection method[J]. Journal of Graphics, 2024, 45(5): 901-912 (in Chinese). DOI
[2]	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. DOI
	ZHAO Z B, HAN Y, TANG C K. Cascade detection method for insulator defects in distribution lines based on improved YOLOv8[J]. Journal of Graphics, 2025, 46(1): 1-12 (in Chinese). DOI
[3]	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 Y, 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).
[4]	牛杭, 葛鑫雨, 赵晓瑜, 等. 基于改进YOLOv8的防振锤缺陷目标检测算法[J/OL]. 图学学报, 1-11 [2025-01-06]. http://kns.cnki.net/kcms/detail/10.1034.T.20241210.1142.002.html.
	NIU H, GE X Y, ZHAO X Y, et al. Vibration damper defect detection algorithm based on improved YOLOv8[J/OL]. Journal of Graphics, 1-11 [2025-01-06]. (in Chinese).
[5]	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
[6]	GHASEMI Y, JEONG H, CHOI S H, et al. Deep learning- based object detection in augmented reality: a systematic review[J]. Computers in Industry, 2022, 139: 103661.
[7]	戚银城, 武学良, 赵振兵, 等. 嵌入双注意力机制的Faster R-CNN航拍输电线路螺栓缺陷检测[J]. 中国图象图形学报, 2021, 26(11): 2594-2604.
	QI Y C, WU X L, ZHAO Z B, et al. Bolt defect detection for aerial transmission lines using Faster R-CNN with an embedded dual attention mechanism[J]. Journal of Image and Graphics, 2021, 26(11): 2594-2604 (in Chinese).
[8]	王韵琳, 冯天波, 孙宁, 等. 融合注意力与多尺度特征的电力绝缘子缺陷检测方法[J]. 高电压技术, 2024, 50(5): 1933-1942.
	WANG Y L, FENG T B, SUN N, et al. Defect detection method for power insulators based on attention and multi-scale context information[J]. High Voltage Engineering, 2024, 50(5): 1933-1942 (in Chinese).
[9]	魏良玉, 邹国锋, 赵新宇, 等. 面向航拍图像的绝缘子弱特征缺陷两阶段检测方法[J]. 国外电子测量技术, 2023, 42(10): 25-34.
	WEI L Y, ZOU G F, ZHAO X Y, et al. Two-stage detection method for weak feature defects of insulators in aerial images[J]. Foreign Electronic Measurement Technology, 2023, 42(10): 25-34 (in Chinese).
[10]	杨露露, 马萍, 王聪, 等. 结合特征重用与重建的YOLO绝缘子检测方法[J]. 计算机工程, 2024, 50(7): 303-313. DOI
	YANG L L, MA P, WANG C, et al. Insulator detection method using yolo combining feature reuse and reconstruction[J]. Computer Engineering, 2024, 50(7): 303-313 (in Chinese). DOI
[11]	白晓静, 谢雅祺, 赵淼, 等. 基于局部特征深度信息的绝缘子小样本缺陷检测[J]. 电网技术, 2024, 48(2): 740-749.
	BAI X J, XIE Y Q, ZHAO M, et al. Few-shot insulator defect detection based on deep information of local features[J]. Power System Technology, 2024, 48(2)3: 740-749 (in Chinese).
[12]	翟永杰, 赵晓瑜, 王璐瑶, 等. IDD-YOLOv7: 一种用于输电线路绝缘子多缺陷的轻量化检测方法[J]. 图学学报, 2024, 45(1): 90-101. DOI
	ZHAI Y J, ZHAO X Y, WANG L Y, et al. IDD-YOLOv7: a lightweight method for multiple defect detection of insulators in transmission lines[J]. Journal of Graphics, 2024, 45(1): 90-101 (in Chinese). DOI
[13]	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
[14]	GUO M H, LU C Z, LIU Z N, et al. Visual attention network[J]. Computational Visual Media, 2023, 9(4): 733-752.
[15]	CHEN K, WANG J Q, PANG J M, et al. MMDetection: open MMLab detection toolbox and benchmark[EB/OL]. [2024- 09-23]. https://arxiv.org/abs/1906.07155.
[16]	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.
[17]	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.
[18]	CAI Z W, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 6154-6162.
[19]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 2999-3007.
[20]	LU X, LI B Y, YUE Y X, et al. Grid R-CNN[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 7355-7364.
[21]	TIAN Z, SHEN C H, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 9626-9635.
[22]	ZHANG H K, CHANG H, MA B P, et al. Dynamic R-CNN: towards high quality object detection via dynamic training[C]// The 16th European Conference on Computer Vision. Cham: Springer, 2020: 260-275.
[23]	SUN P Z, ZHANG R F, JIANG Y, Et al. Sparse R-CNN: end-to-end object detection with learnable proposals[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 14449-14458.
[24]	FENG C J, ZHONG Y J, GAO Y, et al. TOOD: task-aligned one-stage object detection[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 3490-3499.
[25]	LYU C, ZHANG W, HUANG H, et al. Rtmdet: An empirical study of designing real-time object detectors[EB/OL]. (2022-12-14) [2024-09-23]. https://arxiv.org/abs/2212.07784.
[26]	JOCHER G. Ultralytics YOLOv5: version 7.0[EB/OL]. [2024-09-23]. https://docs.ultralytics.com/zh/models/yolov5/#citations-and-acknowledgements.
[27]	JOCHER G, CHAURASIA A, QIU J. Ultralytics YOLO:version 8.0.0[EB/OL]. [2024-09-23]. https://docs.ultralytics.com/zh/models/yolov8/#citations-and-acknowledgements.
[28]	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.
[29]	WANG A, CHEN H, LIU L H, et al. YOLOv10:real-time end-to-end object detection[EB/OL]. [2024-12-05]. https://dblp.uni-trier.de/db/conf/nips/neurips2024.html#WangCLCLHD24.
[30]	VIEIRA-E-SILVA A L B, de CASTRO FELIX H, de MENEZES CHAVES T, et al. STN PLAD: a dataset for multi-size power line assets detection in high-resolution UAV Images[C]// 2021 34th SIBGRAPI Conference on Graphics, Patterns and Images. New York: IEEE Press, 2021: 215-222.
[31]	ZHU X K, LYU S C, WANG X, et al. TPH-YOLOv5: improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 2778-2788.
[32]	LI L L, WANG Z F, ZHANG T T. GBH-YOLOv5: ghost convolution with BottleneckCSP and tiny target prediction head incorporating YOLOv5 for PV panel defect detection[J]. Electronics, 2023, 12(3): 561.
[33]	LI C Y, LI L L, JIANG H L, et al. YOLOv6:a single-stage object detection framework for industrial applications[EB/OL]. [2024-12-05]. https://arxiv.org/abs/2209.02976.
[34]	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. Piscataway: IEEE Press, 2023: 7464-7475.
[35]	JOCHER G, QIU J, CHAURASIA A, et al. Ultralytics YOLO version 8.3.0[EB/OL]. [2024-12-05]. https://github.com/ultralytics/ultralytics.
[36]	LIU S, QI L, QIN H Y, 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.
[37]	WANG J Q, CHEN K, XU R, et al. CARAFE: content-aware reassembly of features[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 3007-3016.
[38]	GHIASI G, LIN T Y, Le Q V. DropBlock: a regularization method for convolutional networks[C]// The 32nd International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2018: 10750-10760.
[39]	SUN K, ZHAO Y, JIANG B, et al. High-resolution representations for labeling pixels and regions[EB/OL]. (2019-04-09) [2024-12-05]. https://arxiv.org/abs/1904.04514.

硬件名称	型号	数量
CPU	英特尔至强6148处理器	1
内存	三星DDR4内存条16 G	8
显卡	NVIDIA GeForce RTX 3090	1
硬盘	Samsung SSD 980 PRO 2TB	1

硬件名称	型号	数量
CPU	英特尔至强6148处理器	1
内存	三星DDR4内存条16 G	8
显卡	NVIDIA GeForce RTX 3090	1
硬盘	Samsung SSD 980 PRO 2TB	1

模型	mAP/%	mAP50/%	mAP75/%	AR1/%	AR100/%	AP50-damage/%	AP50-drop/%	FPS
Cascade R-CNN^[18]	40.7	70.2	42.3	42.5	48.2	51.7	88.7	27.2
SSD	39.9	72.2	39.6	42.8	49.5	55.9	88.5	27.1
RetinaNet^[19]	43.4	76.7	45.1	44.3	56.9	65.9	87.6	27.0
Grid R-CNN^[20]	40.0	72.3	39.1	41.7	49.1	56.0	88.5	27.2
FCOS^[21]	40.4	72.8	40.9	42.9	53.8	58.6	86.9	27.2
Dynamic R-CNN^[22]	40.1	74.8	40.2	41.7	46.8	61.6	88.0	27.2
Sparse R-CNN^[23]	41.7	77.3	40.0	43.4	60.2	64.1	90.5	27.2
TOOD^[24]	43.3	75.7	45.3	43.8	56.5	61.7	89.6	27.2
RTMDet^[25]	44.3	72.9	48.5	44.3	58.8	54.9	90.8	27.9
YOLOv5s^[26]	40.7	67.2	46.5	42.0	51.6	48.2	86.2	46.7
YOLOv8s^[27]	42.2	70.5	45.3	42.7	51.7	54.0	87.0	49.3
YOLOv9s^[28]	42.0	71.4	47.6	43.1	52.0	56.7	86.0	27.0
YOLOv10s^[29]	40.6	70.0	42.4	42.8	50.3	52.8	87.1	38.3
Faster R-CNN(基线)	38.8	75.0	34.0	41.1	47.4	58.7	91.3	23.7
AFAM-Net(本文方法)	40.3	80.7	37.5	42.9	47.6	69.5	91.9	27.2

模型	mAP/%	mAP50/%	mAP75/%	AR1/%	AR100/%	AP50-damage/%	AP50-drop/%	FPS
Cascade R-CNN^[18]	40.7	70.2	42.3	42.5	48.2	51.7	88.7	27.2
SSD	39.9	72.2	39.6	42.8	49.5	55.9	88.5	27.1
RetinaNet^[19]	43.4	76.7	45.1	44.3	56.9	65.9	87.6	27.0
Grid R-CNN^[20]	40.0	72.3	39.1	41.7	49.1	56.0	88.5	27.2
FCOS^[21]	40.4	72.8	40.9	42.9	53.8	58.6	86.9	27.2
Dynamic R-CNN^[22]	40.1	74.8	40.2	41.7	46.8	61.6	88.0	27.2
Sparse R-CNN^[23]	41.7	77.3	40.0	43.4	60.2	64.1	90.5	27.2
TOOD^[24]	43.3	75.7	45.3	43.8	56.5	61.7	89.6	27.2
RTMDet^[25]	44.3	72.9	48.5	44.3	58.8	54.9	90.8	27.9
YOLOv5s^[26]	40.7	67.2	46.5	42.0	51.6	48.2	86.2	46.7
YOLOv8s^[27]	42.2	70.5	45.3	42.7	51.7	54.0	87.0	49.3
YOLOv9s^[28]	42.0	71.4	47.6	43.1	52.0	56.7	86.0	27.0
YOLOv10s^[29]	40.6	70.0	42.4	42.8	50.3	52.8	87.1	38.3
Faster R-CNN(基线)	38.8	75.0	34.0	41.1	47.4	58.7	91.3	23.7
AFAM-Net(本文方法)	40.3	80.7	37.5	42.9	47.6	69.5	91.9	27.2

模型	AP50(杆塔)	AP50(绝缘子)	AP50(间隔棒)	AP50()防震锤)	AP50(塔牌)	mAP50
TPH-YOLOv5s^[31]	41.9	81.3	90.4	25.8	81.9	64.2
GBH-YOLOv5s^[32]	34.7	75.1	85.8	27.6	80.8	60.8
YOLOv6s^[33]	43.9	88.3	77.5	8.3	73.4	58.3
YOLOv7-tiny^[34]	37.6	72.6	79.8	13.8	71.8	55.1
YOLOv8s^[27]	46.4	78.5	91.8	27.1	78.2	64.4
YOLOv10s^[29]	39.2	89.0	91.4	24.1	73.9	63.5
YOLOv11s^[35]	42.7	88.7	88.0	25.4	83.9	65.7
Dynamic R-CNN^[22]	33.5	94.9	85.3	10.0	50.5	54.8
Sparse R-CNN^[23]	46.6	65.2	63.1	18.4	81.4	55.0
RetinaNet^[19]	36.0	83.6	82.7	27.7	78.8	61.8
TOOD^[24]	46.0	92.0	83.8	30.8	76.9	65.9
Faster R-CNN(基线)	27.5	93.9	82.3	7.9	50.5	52.4
AFAM-Net(本文方法)	31.8	95.8	84.3	43.3	85.6	68.1