Anomaly recognition of magnetic flux leakage pipeline based on YOLO-GLW model

Magnetic Flux Leakage (MFL) technology, widely adopted in oil and gas pipeline inspection due to its effectiveness in detecting high-permeability pipeline anomalies, enables integrity assessments through data analysis. However, traditional manual interpretation methods suffer from inefficiency and inaccuracy, and even with deep learning-aided automated defect detection, significant challenges persist, particularly in small target recognition and low-quality sample identification, hindering efficient and reliable inspection. To address these limitations, this study proposes a novel YOLO-GLW model. This model introduces the GhostConv module, achieving a lightweight design while enhancing sensitivity to small targets; integrates the LSKA attention mechanism to improve feature representation and reduce computational complexity; and employs the Wise-IoU loss function to optimize classification performance. We conducted experiments on a self-constructed dataset with six pipeline anomaly types, and the results show the YOLO-GLW model achieves a mean average precision (mAP@0.5) of 96.9%. Compared to the baseline YOLOv8n model, the number of parameters is reduced by 14.95%, floating-point operations down 9.76%, and mAP@0.5 is significantly improved by 2.6%, particularly excelling in the detection of small targets and low-quality samples. With its lightweight design, the model is suitable for deployment on resource-constrained devices, enhancing the efficiency and reliability of pipeline safety maintenance.