Scholarlink

Corrosion in steel pipelines is a critical issue in the oil and gas industry. Early corrosion detection can limit the environmental damage and financial impact. To address this challenge, a new method was developed to segment corrosion defects from ultrasonic images representing variations in pipeline thickness. Different neural network architectures including 5-layers CNN, Mini U-Net, Deep U-Net and ResNet50 were evaluated. Among them, the Deep U-Net demonstrated superior performance with a Dice score of 0.845 and an IoU of 0.732 on validation data, with an IoU of 0.6410 using test data, outperforming existing methods. These results show the potential of Deep U-Net for segmenting corrosion in steel pipelines. These improvements may significantly facilitate the pipeline inspection process by reducing the required manual effort.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CorroSeg: U-Net Based Corrosion Segmentation in Steel Pipelines Using Ultrasonic Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mounir Salhi,
  • Moulay A. Akhloufi

摘要

Corrosion in steel pipelines is a critical issue in the oil and gas industry. Early corrosion detection can limit the environmental damage and financial impact. To address this challenge, a new method was developed to segment corrosion defects from ultrasonic images representing variations in pipeline thickness. Different neural network architectures including 5-layers CNN, Mini U-Net, Deep U-Net and ResNet50 were evaluated. Among them, the Deep U-Net demonstrated superior performance with a Dice score of 0.845 and an IoU of 0.732 on validation data, with an IoU of 0.6410 using test data, outperforming existing methods. These results show the potential of Deep U-Net for segmenting corrosion in steel pipelines. These improvements may significantly facilitate the pipeline inspection process by reducing the required manual effort.