Gravity inversion has become the main method to obtain the 3D density distribution in the exploration of petroleum and mineral resources as well as the study of geological structure. In recent years, with rapid development of deep learning algorithms, gravity intelligent inversion based on convolutional neural networks have also achieved certain good results. However, in the previous study on 3D gravity intelligent inversion, some issues in "data-driven" inversion still need to be faced, that is, insufficient number and lack of the diversity of data sets. This paper selects a suitable U-net network with four-layer down-sampling structure, designs typical block and inclined plate and their combined models to construct the data sets. In order to reduce the overfitting of the network, the data augmentation is introduced to improve the training process for datasets. The data augmentation method greatly alleviates the limited number and types of samples, allowing the network to learn more situations and improve the generalization ability. Theoretical model data tests show that the data augmentation method of Mixup and flip can effectively improve the generalization ability of U-net network, and the noise-adding augmentation has good inversion result for noisy data. The tests on the measured gravity data also verify the effectiveness of the proposed method, which indicates that data augmentation does play a significant role in the improvement of gravity intelligent inversion.

3-D inversion for gravity data is of great significance in geophysical quantitative interpretation. However, due to the complexity of the density distribution on the subsurface and nonuniqueness of inverse problems, the classical inversion still has many shortages. The smooth constrained inversion results in redundant structures, whereas the results of the compact inversion are too focused, and excessively rely on the choice of target density parameters. In view of the above issues, we develop a 3-D combined compact and smooth inversion for gravity and gravity gradient data. The objective function is built by the combination of the minimization of field source volume and the misfit between observed and predicted data. The prior information is treated as the “target density” in the inversion process. Furthermore, in order to solve the problem of overreliance on “target density,” we improve a 3-D gravity inversion algorithm where the model constraints include the combination of compact and smooth factors. Model tests show the method can better recover the source distribution when the parameter of target density is not given. Also, we make a comparative analysis of the inversion results of different gradient components and conclude that the inversion results of the gradient data $g_{xy}$ are better than those of other gradient components. Finally, we apply the method to the real gravity data. The inferred results of the anomalous rock mass are consistent with the previous geologic knowledge.

Apparent magnetization mapping is a technique to estimate magnetization distribution in the subsurface magnetic layer from the observed magnetic data, of benefit in identifying lithologic units and delineating magnetic geologic boundaries. The conventional approaches for apparent magnetization mapping usually neglect effects of remanence, resulting in large geologic deviation and the occurrence of negative magnetization when the magnetic layer contains strong remanent magnetization. We have developed a space-domain inversion approach for apparent magnetization mapping based on the amplitude of magnetic anomaly (AMA), the analytic signal (AS), and the normalized source strength (NSS) to reduce effects of remanent magnetization. The AMA, AS, and NSS are three common quantities insensitive or weakly sensitive to the remanence transformed from the magnetic total field anomaly or components. The magnetic layer underground is first divided into a regular grid of vertical rectangular prisms, each having a cross-sectional area of one grid square and a uniform magnetization. Then, an iterative algorithm is adopted to invert each quantity of the AMA, AS, and NSS to obtain an optimum value of magnetization of each prism in the magnetic layer. The inversion approach permits the top and bottom surfaces of the magnetic layer to be constant or variable in depth, and requires no prior information of magnetization directions. Our tests on the synthetic and real data from the metallic ores area in the southern margin of North China have proved the feasibility and robustness of the presented inversion approach. All of the AMA, AS, and NSS inversions produced nonnegative magnetization distribution in the magnetic layer. Also, the AS and NSS inversions produced a better resolution of magnetization distribution than that of the AMA.