Abstract: Accurately quantifying metal loss defects in oil pipelines is crucial for ensuring structural integrity and operational safety. To improve the accuracy and robustness of defect profile estimation, this study proposes a defect profile estimation model named MFL-SPNet based on multimodal magnetic flux leakage (MFL) image fusion. First, an adaptive Dirichlet function was employed to encode one-dimensional axial and radial MFL signals into two-dimensional representations and fuse them complementarily, constructing a unified image representation with higher information density and stronger anti-interference capability. Second, a dynamic convolution-enhanced backbone network (DEM-backbone) was designed to adaptively enhance geometric features of defect edges and weak response regions through dynamic convolution, thereby improving the extraction capability for local morphology of complex defects. Finally, a parallel feature alignment pyramid network (PA-FPN) was constructed to suppress feature misalignment and detail attenuation during cross-scale fusion through a bidirectional calibration mechanism, ensuring the integrity of contour details. Experiments on a mixed dataset containing both simulated and real-world scenarios show that MFL-SPNet outperforms existing mainstream models in key metrics such as absolute error in defect length and comprehensive error in profile depth. The defect length error and profile depth comprehensive error were reduced by 13.0% and 11.2%, respectively. Ablation experiments and visualization results further validate the effectiveness of the multimodal fusion strategy and feature calibration modules. The MFL-SPNet model achieves high-fidelity parametric estimation of defect profiles, providing reliable technical support for pipeline integrity management and maintenance decision-making.