基于UNet++技术的油田管线泄漏监测
Oilfield Pipeline Leakage Monitoring Based on UNet++
朱苏青 1段志刚 2周宇成 3李汉周 2李兴 4周新墙1
作者信息
- 1. 中国石油化工股份有限公司江苏油田分公司科技与信息管理部,江苏扬州 225009
- 2. 中国石油化工股份有限公司江苏油田分公司石油工程技术研究院,江苏扬州 225009
- 3. 中国石油化工股份有限公司江苏油田分公司,江苏扬州 225009
- 4. 中国石油化工股份有限公司江苏油田分公司采油一厂,江苏扬州 225265
- 折叠
摘要
针对石油泄漏初期目标较小、石油颜色特征在昏暗场景下不容易区分、残留的泄漏痕迹难以完全清除等易造成误检漏检问题,提出一种基于UNet++技术的油田管线泄漏监测网络.该方法采用Kind++图像增强,增加图像照度;引入SiLU激活函数,缓解梯度消失,保留更多特征信息,加快网络收敛速度;引入EMA注意力机制聚合多尺度空间结构信息,增强特征融合质量.实验结果表明,该方法准确率为95.5％,精确率为95.5％,召回率为95.2％,平均交并比为91.3％,可以准确分割出昏暗环境、遮挡环境、小目标的石油泄漏区域,能实现油田管线泄漏监测工作.
Abstract
Aiming at the problems of initial target of oil leakage is small,the color characteristics of oil is not easy to distinguish in the dark scene,and the residual leakage traces are difficult to be completely removed,and easy to cause misdetection and omission issues.An oil pipeline leakage monitoring network based on UNet++is proposed.The method uses Kind++image enhancement to increase image illumination.SiLU activation function is introduced to ease gradient disappearance,retain more feature information,accelerate network convergence speed,and EMA attention mechanism is introduced to aggregate multi-scale spatial structure information to enhance feature fusion quality.The experimental results show that the accuracy rate of the method is 95.5％,the precision rate is 95.5％,the recall rate is 95.2％,and the average crossover ratio is 91.3％.The method can accurately segment the oil leakage area in dim environment,shielded environment and small target,and can be qualified for pipeline leakage monitoring in oil fields.
关键词
泄漏监测/SiLU激活函数/EMA注意力机制Key words
leak monitoring/SiLU activation function/EMA attention mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据