Page 96 - 无损检测2025年第三期

试验研究
试验研究

DOI：10.11973/wsjc240282

基于 YOLOv5 的管道环焊缝缺陷目标检测

算法分析

彭云超 ，李亚平 ，齐 峰 ，饶连涛 ，刘九宏 ，徐 杰 3
3
2
1
3
1
（1. 国家管网集团东部原油储运有限公司，徐州 221008；2. 国家管网集团科学技术研究总院分公司，廊坊 065000；
3. 中国矿业大学 材料与物理学院，徐州 221116）
摘 要：基于漏磁内检测技术，采用PyTorch框架，应用YOLOv5算法对实际管道环焊缝缺陷
漏磁信号图像进行了自动识别，通过对算法的进一步优化与改进，分析了其对自动识别准确率的影
响。试验结果表明，模型在图像数据混合增强后，各指标均有了显著提升，IoU阈值大于0. 5的平均
精度提升了近30%；通过增加小目标检测层，大幅降低了目标检测损失函数均值，改善了缺陷的目
标检测效果；显著提升了缺陷的识别率，最高提升11. 05%，获得了较好的自动识别结果。该方法
为管道环焊缝信号异常数据判读提供了高效的方法和技术手段，对于管道智能化检测实际生产作
业具有重要作用。
关键词：漏磁内检测；管道环焊缝；自动识别；YOLOv5算法；小目标检测
中图分类号：TG115.28；TE319 文献标志码：A 文章编号：1000-6656(2025)03-0062-09

Analysis of target detection algorithm for pipeline circumferential weld defects based on YOLOv5

PENG Yunchao , LI Yaping , QI Feng , RAO Liantao , LIU Jiuhong , XU Jie 3
1
1
3
3
2
（1. PipeChina Network Corporation Eastern Oil Storage and Transportation Co. Ltd., Xuzhou 221008, China; 2. PipeChina
Network Corporation Institute of Science and Technology, Langfang 065000, China; 3. School of Materials Science and Physics,
China University of Mining and Technology, Xuzhou 221116, China）
Abstract: Based on magnetic flux leakage internal detection technology, the PyTorch framework was used and the
YOLOv5 algorithm was applied to automatically identify defects in pipeline circumferential weld seam magnetic flux leakage
signal images. Through further optimization and improvement of the algorithm, its impact on the accuracy of automatic
recognition was analyzed. The experimental results indicated that after image data mixing and enhancement in the model,
all indicators were significantly improved and the average accuracy of IoU thresholds greater than 0.5 was improved by
nearly 30%. By adding small object detection layers, the average loss function of object detection was significantly reduced
and the object detection effect of defects in images was improved. The recognition rate of defects in images was significantly
improved, with a maximum increase of 11.05%, achieving good automatic recognition results. This proposed method
provided effective ideas and technical means for the interpretation of abnormal signal data of pipeline circumferential welds,
and played an important role in the intelligent detection of pipelines in actual production operations.
Key words: internal magnetic leakage testing; pipeline circumferential weld; automatic recognition; YOLOv5
algorithm; small object detection

收稿日期：2024-06-21 随着经济的发展，全球对于天然气、石油等能源
基金项目：国家管网集团东部原油储运有限公司揭榜挂帅科技项 的需求与日俱增。管道输送是与铁路、公路、水运和
目（GWHT20220042619）
航空并列的五大运输方式之一，具有一次性投资少、
作者简介：彭云超（1975－） ，男，高级工程师，学士，主要从事油
运输成本低、安全性高、利于环保等独特优势，在经济
气管道完整性管理技术研究工作
通信作者：徐 杰，j.xu@cumt.edu.cn 发展及能源安全中有着举足轻重的地位 [1-2] 。2021年
62
2025 年 第 47 卷 第 3 期
无损检测