Joint inversion of gravity and gravity gradient data are of paramount importance in geophysical exploration, as the integration of these datasets enhances subsurface resolution and facilitates the accurate delineation of ore body shapes and boundaries. Conventional regularization methods, such as the L₂-norm, frequently yield excessively smooth solutions, which complicates the recovery of sharp boundaries. Furthermore, disparities in data units, magnitudes, and noise levels introduce additional complexities in selecting appropriate weighting functions and inversion parameters. To address these challenges, this study proposes a greedy inversion method based on cosine similarity, which identifies the most relevant cells and reduces the complexity involved in data weighting and parameter selection. Additionally, it incorporates prior information on density limits to achieve a high-resolution and sparse solution. To further enhance the stability and accuracy of the inversion process, a pruning mechanism is introduced to dynamically detect and remove erroneously selected cells, thereby suppressing error propagation. Synthetic model experiments demonstrate that incorporating the pruning mechanism significantly improves inversion accuracy. The method not only accurately resolves models of varying volumes while avoiding local convergence issues in the presence of major anomalies, but also exhibits strong robustness against noise, successfully delineating clear boundaries even when applied to complex composite models contaminated with 10% Gaussian noise. Finally, when applied to the joint inversion of measured gravity and gravity gradient tensor data from the Vinton salt dome, the results closely align with previous studies and actual geological observations.

Although various magnetic inversion techniques have been developed in geophysics, traditional methods are often constrained by inherent limitations such as low computational efficiency and pronounced non-uniqueness. In 3D magnetic inversion, multi-dimensional deep learning methods have shown promise in numerical simulations; however, their generalization capabilities and practical effectiveness in real-world geological applications, particularly in complex settings like gold exploration, remain underexplored. This study introduces MAGNETPRO, a residual convolutional neural network based on an encoder–decoder architecture, designed to accurately invert 2D magnetic field data into 3D magnetic susceptibility structures. To enhance the model’s generalization ability and inversion accuracy, an innovative data construction strategy was implemented to create a highly randomized training dataset incorporating complex geological features. Theoretical model tests demonstrate that MAGNETPRO achieves inversion accuracies of 97% across the entire region and 80% within magnetic structure areas, highlighting its excellent spatial resolution and structural recognition capabilities. To further validate its practical effectiveness, the method was applied to real exploration data from a gold mining area in Fujian Province. The results show a high degree of consistency between the inversion outcomes and drilling data, confirming the method’s reliability and practical value under real geological conditions.