Using Multi-scale Detection Mechanism and Dynamic Monitor Defect Area for Pipeline Weld Identification

To effectively detect weld defects of various sizes, particularly small defects that are often overlooked and result in low detection accuracy, we propose a method for pipeline weld identification and dynamic monitoring of defect areas based on a multi-scale detection mechanism. This paper introduces multi-scale feature extraction, an Enhanced Multi-scale Attention (EMA) mechanism optimizer into the X-ray weld defect detection network. These enhancements aim to improve the model’s ability to capture features at different levels, address information loss from channel dimensionality reduction, and enhance the detection of both large targets and subtle details critical for identifying small defects. The model successfully identifies various weld defect types. Additionally, edge detection is employed to calculate the physical area of the cropping regions and non-edge areas, enabling the extraction of weld defect areas. Cubic spline smoothing is used to model the trends in defect area changes and predict the lifecycle of the pipeline. Experimental results demonstrate that this method effectively detects small and complex weld defects in real-time, achieving precision and recall rates of 96.9% for scratches and 90.4% for incomplete fusion. Furthermore, the correctly identified defects and the dynamic monitoring of defect areas provide a valuable decision-making framework for the safe operation and preventive maintenance of oil and gas pipelines.