电子测量与仪器学报

彭道刚,潘俊臻,王丹豪,胡捷.基于改进 YOLO v5 的电厂管道油液泄漏检测[J].电子测量与仪器学报,2022,36(12):200-209

基于改进 YOLO v5 的电厂管道油液泄漏检测

Oil leakage detection of pipelines of power plants based on improved YOLO v5

DOI：

英文关键词:pipeline leak detection YOLO v5 algorithm CBAM attention mechanism

基金项目:上海市"科技创新行动计划"高新技术领域项目（21511101800）资助

作者	单位
彭道刚	1. 上海电力大学自动化工程学院
潘俊臻	1. 上海电力大学自动化工程学院
王丹豪	1. 上海电力大学自动化工程学院
胡捷	2. 宝山钢铁股份有限公司电厂

Author	Institution
Peng Daogang	1. College of Automation Engineering, Shanghai University of Electric Power
Pan Junzhen	1. College of Automation Engineering, Shanghai University of Electric Power
Wang Danhao	1. College of Automation Engineering, Shanghai University of Electric Power
Hu Jie	2. Power plant of Baoshan Iron &Steel Co. , Ltd.

摘要点击次数: 701

全文下载次数: 983

中文摘要:

针对电厂油库、化水车间等关键区域油液管道时常发生泄漏问题,本文提出了一种基于改进 YOLO v5 的电厂关键区域管道油液泄漏检测方法,通过融入 CBAM 注意力机制模块,加强对管道油液泄漏区域图像的特征学习与特征提取,同时弱化复杂背景对检测结果的影响;在此基础上运用了双向特征金字塔网络进行多尺度特征融合,减少冗余计算,同时提升算法对小目标的检测能力;最后采用 Focal EIoU Loss 作为损失函数,使回归过程更加专注于高质量锚框,加快收敛速度,提高模型的回归精度和鲁棒性。实验结果表明,本文所提出的改进算法在真实样本中表现良好,平均准确率达 79. 6%,较原 YOLO v5s 目标检测算法提高了 38. 4%,在电厂复杂背景下的误报率和漏报率明显下降,可有效应用于实际生产环境中。

英文摘要:

In view of the frequent leakage of oil pipelines in key areas such as power plant oil depots and chemical water workshops, a pipeline leak detection method in key areas of power plants based on improved YOLO v5 is proposed. The improved YOLO v5 detection algorithm first incorporates CBAM module to strengthen the learning of regional features of pipeline oil leakage images. The CBAM makes the model more focused on the extraction of pipeline leakage features, and weakens the influence of complex backgrounds on detection results. Secondly, a bidirectional feature pyramid network is used for multi-scale feature fusion. It also reduces redundant calculation, and improves the detection ability of the algorithm for small targets. Finally, Focal EIoU Loss is used as the loss function to make the regression process more focused on high-quality anchor boxes. It improves the regression accuracy, speeds up the convergence speed, and improve the robustness of the model. The experimental results show that the improved algorithm performs well in real samples, with an average accuracy rate of 79. 6%, which is 38. 4% higher than the original YOLO v5s algorithm. The false positive rate and the false negative rate in the complex background of the power plant are significantly reduced. It shows that the improved YOLO v5 detection algorithm can be effectively applied in the actual production environment.

查看全文查看/发表评论下载PDF阅读器