Abstract: Drill pipes are subjected to various loads such as tension, torsion, and vibration during drilling operations, making them prone to damage leading to severe accidents. Therefore, damage detection for stress concentration-induced defects in drill pipes is critical. To address the issues of low accuracy and imprecise defect-type prediction in existing drill pipe inspection methods, this study developed a magnetic flux leakage (MFL) detection system for drill pipes based on Gramian angular field (GAF) transformation. The system employed eight probe arrays for omnidirectional scanning. A user-friendly LabVIEW-based human-machine interface was also developed to enhance operational efficiency. Furthermore, this paper proposed a ResNet-50-based MFL detection method for wellhead drill pipe defects. By transforming MFL signals into 2D features using GAF and leveraging ResNet-50 for multi-model feature fusion, the method achieved 3D inversion of various defect types in drill pipes. Experimental results demonstrated that the mean relative error (MRE) of 3D inversion for two defect types was within 3.5%, validating the robustness of the method and its capability for quasi-3D inversion of drill pipe defects. This approach significantly improved the accuracy of defect identification and enabled precise multi-channel defect classification.