High-resolution magnetic field data play a crucial role in intelligent system monitoring and fault diagnosis; however, time-consuming acquisition and strong noise interference limit their practical applications. To address this, this paper proposes a fast and accurate acquisition method capable of reconstructing complete high-resolution data from sparse measurements. The core contribution is a multi-scale information completion network that maps sparse data to multiple low-resolution scales for feature fusion. It fully exploits the spatial correlations among neighboring regions, the network filters noise and completes missing information. Unlike existing super-resolution networks that rely on paired sparse-high-resolution training data, the proposed network trains solely on sparse data, eliminating dependence on high-resolution ground truth and significantly reducing acquisition costs. Simulation and experimental results demonstrate that with only 625 sparse sampling points, the method accurately reconstruct 128×128 high-resolution data under both sparse and uniform sampling conditions, thereby significantly shortening acquisition time. Furthermore, compared to existing methods, it achieves substantial improvements in two key metrics: Peak signal-to-noise ratio and structural similarity index. These results fully verify the feasibility and superiority of the proposed method, offering a practical and effective solution for the rapid high-resolution magnetic field data acquisition.