To break through the drawbacks of traditional damage test methods that cannot obtain the full-plate mechanical property distribution of pipeline steel and are laboratory-dependent and costly, multimicromagnetic nondestructive testing (NDT) technology and machine learning technology are employed. The micromagnetic multiparameter microstructure and stress analysis (3MA) technique is used to extract 11 micromagnetic characteristics of pipeline steel. The Lorentz force-based electromagnetic ultrasound (EMAT) technique is applied to offset the defects of the 3MA technique in testing depth and extract ultrasonic velocity characteristics of pipeline steel as a new characteristic. Based on machine learning techniques, 12 characteristics are used for information fusion to evaluate the mechanical properties of the entire pipeline steel. The prediction results show that, compared with multilayer perceptron (MLP), random forest (RF), and deep neural networks (DNNs), the testing accuracy based on the extreme gradient boosting (XGBoost) algorithm was found to be the highest (PR10 over 93%). Meantime, based on the importance of parameters during the training of the XGBoost model, the modeling mechanism of the surface hardness and tensile strength models is investigated, providing a theoretical basis for analyzing the fusion of micromagnetic feature information based on machine learning (black box).