Abstract: In-line inspection technology of magnetic flux leakage (MFL) plays a critical role in the detection of defects in long-distance oil and gas pipelines. During actual operation, the data collected by in-line inspection tools from pipelines inevitably contain abnormal data. Identifying these anomalies is essential for accurately assessing pipeline performance and is also a key indicator of whether the detection equipment is functioning properly. Given the vast amount of inspection data, traditional manual interpretation methods are time-consuming and no longer meet current industrial demands. To improve recognition efficiency, a lightweight anomaly detection method for MFL data based on a one-dimensional convolutional neural network (1D-CNN) is proposed. The collected MFL data is segmented into multiple sub-sequences along the radial, axial, and circumferential directions, and a dataset is established using the anomaly conditions in the MFL data as labels. By directly inputting the one-dimensional signals into the model, the convolutional neural network (CNN) is employed to automatically extract the characteristic information of the MFL anomalies. The model incorporates batch normalization (BatchNorm) and an optimized Dropout regularization method, and is iteratively trained and evaluated to achieve anomaly recognition in MFL data. This method has a high recognition accuracy, with recall, precision, and accuracy all exceeding 96% when the Dropout ratio is set to 0.3. This demonstrates the method's ability to efficiently handle large-scale data and its applicability in identifying anomalies in MFL inspection data. The proposed deep learningbased method for MFL anomaly detection provides an efficient solution for recognizing abnormal data in long-distance oil and gas pipelines, effectively extracting key features from the MFL data. The experimental results further demonstrate that this deep learning model is accurate and effective, holding great promise for application.