基于轻量化改进YOLOv8的通信光缆缺陷检测

doi:10.11996/JG.j.2095-302X.2025010028

摘要/Abstract

摘要：

在全介质自承式(ADSS)通信光缆缺陷检测领域，针对跨尺度的电腐蚀缺陷做检测存在计算量要求高、检测精度低的问题，提出了一种改进YOLOv8的ADSS通信光缆缺陷检测方法。首先，对自建通信光缆缺陷数据集切片，防止光缆存在的电腐蚀缺陷在缩放的过程中丢失；其次，使用LS-FPN的结构代替传统的颈部结构，保留通道维度中的有利位置信息，解决光缆表面的缺陷尺度跨度问题的同时提高定位缺陷能力；再者，引用了可变形卷积的思想，对原有backbone网络的卷积进行替换，使得网络在特征提取的过程中能够更为关注周围缺陷信息；最后，用Focus-MPDIoU损失函数代替原有的CIoU，Focus-MPDIoU在处理边界情况时表现优异，能够避免过于偏激的损失梯度。实验结果表明，该方法在ADSS通信光缆缺陷数据集上进行实验验证，改进后的模型分别在mAP50-95和mAP50上达到50.6%和87.8%，相比YOLOv8n分别提升了2.1%和3.7%。同时，计算量GFLOPs降低至6.8，参数量降低至1.96 M，降低了检测设备的配置要求，贴近轻量化的工业需求。

关键词: 轻量化, YOLOv8n, 通信光缆, 缺陷检测, MPDIoU

Abstract:

In the field of defect detection of all dielectric self supporting (ADSS) communication cables, the detection of galvanic corrosion defects across scales has the problems of high computational demands and low detection accuracy. In this paper, a defect detection method for ADSS communication cables with improved YOLOv8 was proposed. Firstly, the self-built communication cable defect dataset was sliced to prevent the existence of galvanic corrosion defects in the fiber optic cable from being lost in the process of scaling; secondly, the structure of LS-FPN replaced the traditional necking structure, retaining the favorable positional information in the channel dimension, which resolved the defective scale-spanning problem on the surface of the fiber optic cable while enhancing defect localization capability; furthermore, the idea of deformable convolution was introduced, replacing the convolution in the original backbone network, allowing the network to better focus on the surrounding defect information in the process of feature extraction; finally, the original CIoU was replaced by the Focus-MPDIoU loss function, which excels in handling boundary cases and avoids overly radical loss gradients. The experimental results validated that the method on the ADSS communication fiber optic cable defect dataset, with the improved model achieving 50.6% and 87.8% on mAP50-95 and mAP50, respectively, reflecting increases of 2.1% and 3.7% compared to YOLOv8n. Meanwhile, the computational GFLOPs were reduced to 6.8 and the number of parameters decreased to 1.96 M, reducing the configuration requirements of the inspection equipment and meeting the lightweight industrial demand.

Key words: lightweight, YOLOv8n, communication fiber optic cable, defect detection, MPDIoU

中图分类号:

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.

WANG Zhidong, CHEN Chenyang, LIU Xiaoming. The defect detection method for communication optical cables based on lightweight improved YOLOv8[J]. Journal of Graphics, 2025, 46(1): 28-34.

图/表 12

图1 改进YOLOv8网络结构

Fig. 1 Improvement of YOLOv8 network structure

图2 可变形卷积操作

Fig. 2 Deformable convolutional operations

图3 C2f_DCNv2模块

Fig. 3 C2f_DCNv2 module

图4 分层尺度特征金字塔网络

Fig. 4 Layered scales-feature pyramid network

图5 选择性特征融合模块

Fig. 5 Select feature fusion module

图6 ADSS通信光缆缺陷类别图((a)局部坑蚀；(b)坑蚀；(c)断缆；(d)剥落；(e)电痕)

Fig. 6 ADSS fiber optic cable defect category chart ((a) Smallpitting; (b) Pitting; (c) Break; (d) Breakdown; (e) Tracking)

表1 YOLOv8不同型号的模型性能对比

Table 1 Comparison of model performance of different models of YOLOv8

Model	mAP50/%	mAP50-95/%	GFLOPs	Params/M
YOLOv8n	84.1	48.5	8.1	3.01
YOLOv8s	88.4	52.3	28.4	11.14
YOLOv8m	90.1	55.1	78.7	25.86

表2 消融实验

Table 2 Ablation experiments

A	B	C	mAP50/%	mAP50-95/%	GFLOPs	Params/M	Speed/ms
Baseline			84.1	48.5	8.1	3.01	2.1
√			84.5	49.4	8.0	3.03	1.5
	√		86.0	49.7	6.9	1.93	1.8
		√	84.9	49.0	8.1	3.01	2.4
√	√		86.4	50.3	6.8	1.96	2.2
√		√	85.4	49.5	8.0	3.03	2.3
	√	√	85.9	50.3	6.9	1.93	1.6
√	√	√	87.8	50.6	6.8	1.96	2.1

图7 热力对比图

Fig. 7 Heat comparison chart ((a) Baseline; (b) YOLOv8-LSFPN)

图8 训练对比图((a) mAP50精度对比；(b) mAP50-95精度对比)

Fig. 8 Training comparison chart ((a) mAP50 accuracy comparison; (b) mAP50-95 accuracy comparison)

表3 对比实验

Table 3 Comparative experiments

Model	mAP50/%	mAP50-95/%	GFLOPs	Params/M	FPS
YOLOv3-tiny	81.8	44.8	18.9	12.10	18.9
YOLOv5n	83.4	47.2	7.1	2.50	19.4
YOLOv6n	83.9	46.8	11.9	4.20	23.1
YOLOv8n	84.1	48.5	8.1	3.01	20.1
YOLOv9t	84.0	48.1	7.9	2.00	14.0
YOLOv10n	82.9	47.4	8.4	2.70	20.3
Gold-YOLO	84.6	48.2	10.2	5.90	15.9
RT-DETR	87.5	49.5	125.6	41.90	4.5
Ours	87.8	50.6	6.8	1.96	19.1

图9 检测结果对比图

Fig. 9 Comparison chart of test results ((a) YOLOv8; (b) YOLOv8-LSFPN)

参考文献 17

[1]	许一帆, 高文焕, 龙水铨. 全介质自承式光缆ADSS支架的设计[J]. 集成电路应用, 2023, 40(7): 288-290.
	XU Y F, GAO W H, LONG S Q. Design of ADSS bracket for all dielectric self-supporting optical cables[J]. Application of IC, 2023, 40(7): 288-290 (in Chinese).
[2]	杨绍哲, 纪萍, 刘喜军, 等. 全介质自承式光缆电腐蚀预防机制有限元分析[J]. 上海电气技术, 2023, 16(1): 37-43.
	YANG S Z, JI P, LIU X J, et al. Finite element analysis of electrical corrosion prevention mechanism of ADSS optical cable[J]. Journal of Shanghai Electric Technology, 2023, 16(1): 37-43 (in Chinese).
[3]	郑建军, 刘俊, 史贤达, 等. 在运ADSS光缆断裂原因分析[J]. 黑龙江电力, 2021, 43(5): 391-393.
	ZHANG J J, LIU J, SHI X D, et al. Cause analysis of ADSS optical cable fracture in operation[J]. Heilongjiang Electric Power, 2021, 43(5): 391-393 (in Chinese).
[4]	DU B X. Discharge energy and dc tracking resistance of organic insulating materials[J]. IEEE Transactions on Dielectrics and Electrical Insulation, 2001, 8(6): 897-901.
[5]	蔡嘉磊, 茅智慧, 李君, 等. 基于深度学习的目标检测算法与应用综述[J]. 网络安全技术与应用, 2023(11): 41-45.
	CAI J L, MAO Z H, LI J, et al. A review of deep learning based target detection algorithms and applications[J]. Network Security Technology & Application, 2023(11): 41-45 (in Chinese).
[6]	CHEN X H, LIU N, YOU B, et al. A novel method for surface defect inspection of optic cable with short-wave infrared illuminance[J]. Infrared Physics & Technology, 2016, 77: 456-463.
[7]	QU L Q, MU H B, ZOU X Y, et al. A method for online monitoring intermittent cable defects based on SSTDR[J]. Energy Reports, 2023, 9: 904-911.
[8]	WANG S R, CHU J W, SU X, et al. Cable defect detection technology based on low frequency method and damped oscillation wave[C]// The 3rd International Academic Exchange Conference on Science and Technology Innovation. New York: IEEE Press, 2021: 1717-1720
[9]	TABERNIK D, ŠELA S, SKVARČ J, et al. Segmentation-based deep-learning approach for surface-defect detection[J]. Journal of Intelligent Manufacturing, 2020, 31(3): 759-776.
[10]	崔克彬, 焦静颐. 基于MCB-FAH-YOLOv8的钢材表面缺陷检测算法[J]. 图学学报, 2024, 45(1): 112-125. DOI
	CUI K B, JIAO J Y. Steel surface defect detection algorithm based on MCB-FAH-YOLOv8[J]. Journal of Graphics, 2024, 45(1): 112-125 (in Chinese). DOI
[11]	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.
[12]	MA S L, XU Y. MPDIoU: a loss for efficient and accurate bounding box regression[EB/OL]. [2024-04-26]. https://arxiv.org/abs/2307.07662.
[13]	DAI J F, QI H Z, XIONG Y W, et al. Deformable convolutional networks[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 764-773.
[14]	ZHU X Z, HU H, LIN S, et al. Deformable ConvNets V2: more deformable, better results[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 9300-9308.
[15]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 936-944.
[16]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// The 15th European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[17]	熊举举, 徐杨, 范润泽, 等. 基于轻量化视觉Transformer的花卉识别[J]. 图学学报, 2023, 44(2): 271-279. DOI
	XIONG J J, XU Y, FAN R Z, et al. Flowers recognition based on lightweight visual transformer[J]. Journal of Graphics, 2023, 44(2): 271-279 (in Chinese). DOI