[1]简路生,龙飞飞.基于改进 YOLOv8 的储罐焊缝缺陷检测方法[J].机械与电子,2026,44(05):41-48.
　JIAN Lusheng,LONG Feifei.A Storage Tank Weld Defect Detection Method Based on an Improved YOLOv8 Model[J].Machinery & Electronics,2026,44(05):41-48.
点击复制

基于改进 YOLOv8 的储罐焊缝缺陷检测方法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年05期
页码:
41-48
栏目:
智能检测
出版日期:
2026-05-27

文章信息/Info

Title:
A Storage Tank Weld Defect Detection Method Based on an Improved YOLOv8 Model
文章编号:
1001-2257 ( 2026 ) 05-0041-08
作者:
简路生; 龙飞飞
大连民族大学机电工程学院,辽宁 大连 116600
Author(s):
JIAN Lusheng ;  LONG Feifei
( School of Mechanical and Electrical Engineering , Dalian Minzu University , Dalian 116600 , China )
关键词:
焊缝缺陷检测;  YOLOv8n 模型; 部分卷积; GAM 注意力机制;  EIoU 损失函数
Keywords:
weld defect detection ;  YOLOv8n model ;  partial convolution ;  GAM attention mechanism ;  EIoU loss function
分类号:
TP391.4 ;TP242.2
文献标志码:
A
摘要:
针对储罐焊缝缺陷检测对高精度、轻量化和实时性的三重需求,结合储罐焊缝检测机器人的实际应用场景,提出一种基于 YOLOv8n 模型改进的 YOLOv8n-PConv-GAM-EIoU 焊缝缺陷检测方法。在实验室模拟真实储罐焊缝环境采集 320 张原始图像,经数据增强将数据集扩充至 6 400 张,构建涵盖多种结构类型的焊缝缺陷数据集用于模型深度学习训练。选用 YOLOv8n 作为基础模型,在主干网络中引入 FasterNet 核心算子 PConv 重构网络,替换原模型 C2f 模块的标准卷积,添加 GAM 强化微小缺陷特征,并采用 EIoU 损失函数优化边界框回归以提升定位精度。改进后模型精确率达 98.0% ,较原始 YOLOv8n 模型提升 10.7 百分点, mAP@0.5 达 90.4% 、 mAP@0.5 : 0.95 达 68.3% ,参数量降至 1.7×106 ,减少 43% ,每秒浮点运算次数降至 5.0×109 ,降低 38% ,实验环境下检测帧率达 58.7 帧/ s ,为后续部署至储罐焊缝检测机器人嵌入式设备奠定轻量化与实时性基础。
Abstract:
To satisfy the triple requirements of high precision , lightweight design , and real time per- formance in storage tank weld defect detection , this study proposes an improved YOLOv8n model , designated as YOLOv8n PConv GAM EIoU , tailored for deployment on weld inspection robots.A specialized dataset was constructed by collecting 320 raw images in a simulated environment , which was subsequently expanded to 6 , 400 samples through data augmentation to encompass diverse structural types of weld defects.In present work , YOLOv8n is selected as the baseline model.The backbone network was reconstructed by incorporating the Partial Convolution ( PConv ) operator from FasterNet to replace standard convolutions within the C2f modules , thereby reducing redundant computations.Furthermore , a Global Attention Mechanism ( GAM ) was integrated to enhance the feature representation of micro defects , while the EIoU loss function is adopted to optimize bounding box regression for superior localization accuracy. The experimental results indicate that the proposed model achieves a precision of 98.0% , representing a 10.7 percentage point improvement over the baseline.The mAP@0.5 and mAP@0.5 : 0.95 reached 90.4% and 68.3% , respectively.Meanwhile , the number of model parameters was reduced to 1.7×106 ( a 43% decrease ), and the computational complexity was lowered to 5.0 GFLOPs ( a 38% reduction ) .With a detection rate of 58.7 FPS , the model provides a robust lightweight and real-time foundation for embedded deployment on storage tank inspection robots.

参考文献/References:

[ 1 ] 王春妮,李健,白真权,等 . 油气输送管道失效分析技术现状及研究进展[ J ] . 石油管材与仪器, 2024 , 10 ( 1 ): 1-8.

[ 2 ] 孙东亮,李子杰 . 典型自然灾害引发常压储罐破坏的定量风险评估[ J ] . 安全与环境学报, 2022 , 22 ( 3 ): 1178-1186.
[ 3 ] 李安,陶波,丁汉 . 磁吸附爬壁机器人研究进展及其应用探索[ J ] . 机器人, 2025 , 47 ( 1 ): 123-144.
[ 4 ] Chetih N , Thelaidjia T , Ramou N , et al.Automatic welding defect detection of X-ray images by using adaptively regularized kernel fuzzy technique integrated with edge based level set function [ J ] .Russian Journal of Nondestructive Testing , 2024 , 60 ( 9 ):1051-1061.
[ 5 ] 卢鸿玮,周慧慧,刘婵,等 . 一种轻量级焊缝表面缺陷检测方法[ J ] . 中国激光, 2025 , 52 ( 24 ): 39-50.
[ 6 ] 苏志威,黄子涵,邱发生,等 . 基于改进 YOLOv8 的航空铝合金焊缝缺陷检测方法[ J ] . 航空动力学报, 2024 ,39 ( 6 ): 121-129.
[ 7 ] Li Jie , Jin Shanshan , Wang Cunjin , et al.Weld line recognition and path planning with spherical tank inspection robots [ J ] .Journal of Field Robotics , 2022 , 39 ( 2 ):131-152.
[ 8 ] 李云浩,李成铁,李秋明 . 一种轻量级实时焊缝缺陷分类模型[ J ] . 中国激光, 2025 , 52 ( 12 ): 44-50.
[ 9 ] Gajalakshmi K , Krishnan B S , Saravanan S.Automated weld defect detection using YOLOv8 algorithm [ J ] . Welding international , 2025 , 39 ( 9 ): 649-655.
[ 10 ] 岳剑峰,李伟明,宁黎华,等 . 基于 YOLO-DEFW 的焊缝缺陷实时检测算法研究[ J ] . 中国激光, 2025 , 52( 8 ): 64-76.
[ 11 ] 钮会昌,狄巨星,杨阳 . 基于改进 YOLOv8s 的焊缝缺陷检测算法[ J ] . 长江信息通信, 2025 , 38 ( 4 ): 69-72.
[ 12 ] 张润梅,潘晨飞,陈梓华,等 . 改进 YOLOv8 的焊缝表面缺 陷 检 测 算 法 [ J ] . 光 电 工 程, 2025 , 52 ( 3 ):112-125.
[ 13 ] 吴磊,储钰昆,杨洪刚,等 . 面向铝合金焊缝 DR 图像缺陷的 Sim-YOLOv8 目标检测模型[ J ] . 中国激光,2024 , 51 ( 16 ): 29-38.
[ 14 ] 宋杰三 . 基于改进 Yolov8 的焊缝缺陷检测研究[ J ] .中国设备工程,2024 ( 14 ): 196-199.
[ 15 ] 陈新,徐辉,张孜勉,等 . 基于改进 YOLOv8 的轻量化焊缝表面缺陷检测方法[ J ] . 现代信息科技,2025 , 9( 3 ): 44-49.
[ 16 ] 李 冠 胜,阮 景 奎,王 宸,等 . 基 于 改 进 FasterNet-YOLOv8 的焊缝表面缺陷检测算法[ J ] . 机电工程技术,2026 , 55 ( 2 ): 78-83.
[ 17 ] Sreejam M , Agilandeeswari L .Deep multimodal unmixing of hyperspectral images using Convolutional Block Attention Module ( CBAM ) and LiDAR features [ J ] .The Egyptian Journal of Remote Sensing and Space Sciences , 2025 , 28 ( 4 ): 666-680.
[ 18 ] Ni Youhao , Wang Hao , Mao Jianxiao , et al.Quantitative detection of typical bridge surface damages based on global attention mechanism and YOLOv7 network [ J ] .Structural Health Monitoring , 2025 , 24 ( 2 ): 941-962.

备注/Memo

备注/Memo:
收稿日期: 2026-02-26
基金项目:大连民族大学研究生创新项目资助( 0102-202201 )
作者简介:简路生 ( 1999- ),男,贵州遵义人,硕士研究生,研究方向为储罐爬壁机器人焊缝缺陷检测;龙飞飞 ( 1979- ),男,黑龙江大庆人,博士,教授,研究方向为工业结构中基于机器学习的检测数据分析技术、过程工业检测机器人研发。
更新日期/Last Update: 2026-08-25