Abstract: Long-distance oil and gas pipelines are crucial in the global energy network. However, due to complex internal and external environments, defects can be formed on a pipeline’s surface, posing severe threats to structural safety. Aiming to detect surface defects, recent works have used magnetic flux leakage (MFL) inspection data for defect recognition and defect size estimation. Accurately locating and measuring defects based on the MFL data is essential for pipeline integrity assessment and safety maintenance. To obtain effective MFL data on pipeline defects, this study constructs an experimental pipeline at the Daxing pulling-through test site in Beijing. An ultra-high-definition MFL inspection robot is employed to collect defect data, which are then used to construct a defect detection and size estimation database. In addition, to achieve precise defect recognition and quantification, a cascaded method, which integrates a mature computer vision detection model, the YOLOv11 model, with a physics-informed and data-driven prior deep-learning quantification model, is proposed. Validation results show that, even for a limited amount of data, the proposed defect recognition model can achieve an AP50 of 92.1% at a confidence threshold of 0.6, a precision of 100%, a recall of 84.29%, and an F1-score of 91.47%, indicating high accuracy in identifying surface defects on pipelines. The quantification model can achieve the goodness of fit (Gof) values of 0.987, 0.979, and 0.994 for defect length, width, and depth, with the mean absolute percentage error (MAPE) of 7.97%, 8.52%, and 4.74%, respectively. Comparison analysis with different models confirms the superiority of the proposed cascaded recognition and quantification approach. The results also demonstrate that the proposed method can effectively identify and quantify defects in long-distance pipelines. Finally, it can improve the interpretation efficiency of MFL inspection data and provide reliable support for residual strength assessment and remaining life prediction of pipelines.