【摘要】 长输管道焊缝质量直接影响石油天然气运输安全。焊接工艺不当或环境因素可能导致焊缝缺陷，成为管道事故主要原因，因此缺陷检测至关重要。传统X射线检测依赖人工分析，效率低且主观性强。本文提出基于改进U-Net的焊缝缺陷像素级识别方法，采用预训练ResNet 18作为编码器，加快收敛并减少过拟合，同时在编码器和解码器间引入轻量级通道注意力机制，旨在自适应选择重要特征，进一步提升特征表达能力。实验表明该方法优于其他像素级检测方法，骰子系数（Dice）达0.917，交集大于联合（IoU）为0.849。更多还原

【Abstract】 The quality of welded joints in long-distance pipelines directly affects the safety of oil and gas transportation. Improper welding techniques or environmental factors may lead to weld defects, which constitute a primary cause of pipeline accidents. Therefore, defect detection is of paramount importance. Conventional X-ray inspection relies on manual analysis, resulting in low efficiency and high subjectivity. This paper proposes a pixellevel weld defect identification method based on an improved U-Net architecture. The approach employs a pre-trained ResNet-18 as the encoder to accelerate convergence and mitigate overfitting risks. Additionally, a lightweight channel attention mechanism is incorporated between the encoder and decoder to adaptively adjust feature importance and enhance feature representation capabilities. Experimental results demonstrate that the proposed method outperforms other pixel-level detection approaches, achieving a Dice coefficient(Dice) of 0.917 and an Intersection over Union(Io U) of 0.849.更多还原