Aeromagnetic measurement plays an important role in mineral exploration, but unmanned aerial vehicles generate maneuvering noise during aerial flight, which negatively impacts the accuracy of aeromagnetic measurement data. Therefore, aeromagnetic compensation is an indispensable step in aeromagnetic data processing. The multicollinearity of variables in the aeromagnetic compensation model based on linear regression affects its accuracy, resulting in a large difference in the compensation effect of the same group of compensation coefficients in different directions. In order to obtain high-quality aeromagnetic data, this study proposes an adaptive model-based method for suppressing aeromagnetic maneuvering noise. First, due to the fact that the variables that cause multiple collinearity in the compensation model are related to the flight heading, the model variables are adaptively assigned to each heading based on the characteristics of the variable data for different headings. The compensation model is optimized and improved, and the impact of multiple collinearity is thus suppressed. In adaptive modeling, variables with greater significance and smaller multicollinearity are automatically allocated to build the optimal heading model, and then high-precision compensation coefficients are obtained. This algorithm was applied to the data collected by a certain unmanned aerial vehicle aeromagnetic measurement platform in Ma’anshan and compared with traditional methods. The experimental results show that the adaptive modeling-based aeromagnetic compensation algorithm is superior to traditional algorithms, with fewer errors and a higher improvement ratio. Hence, the method can effectively solve the ill-conditioned problem of a model affected by multicollinearity and further improve its compensation accuracy and robustness. Moreover, the feasibility and value of this algorithm were verified in actual mineral resource detection.

Quadrotor UAV-based aeromagnetic survey for mineral exploration has become a crucial solution in modern airborne geophysics due to its prominent advantages of cost-effectiveness and high efficiency. During the detection process, the magnetic anomaly interference generated by the quadrotor UAV itself reduces the signal-to-noise ratio (SNR) of the target signal, and some noise overlaps with the target signal in both time and frequency domains. Traditional methods exhibit poor compensation capability for such noise. To address these issues, this paper proposes an aeromagnetic compensation method based on a time–frequency joint representation neural network. This method combines continuous wavelet transform (CWT) and bidirectional long short-term memory (Bi-LSTM) to establish a prediction model. It uses wavelet transform to extract the frequency variation characteristics of the UAV’s magnetic interference, and it inputs these frequency characteristics along with the original time-domain data into the Bi-LSTM network to predict the UAV’s noise. Bi-LSTM can effectively extract the temporal logical connections in time-series signals, thereby improving the accuracy of the compensation model and ensuring high robustness. In this study, magnetic interference data from quadrotor UAV compensation flights were collected for experiments to evaluate the performance of the proposed method. Experimental results show that the neural network fused with time–frequency features, when applied to UAV aeromagnetic compensation, significantly enhances the accuracy and robustness of the compensation method. To verify the method’s effectiveness in removing UAV-generated noise during actual exploration, aeromagnetic survey data from a specific area were compensated using this method.

This paper proposes a physics-constrained 3D Swin Transformer (ST) for gravity inversion. By leveraging the self-attention mechanism in 3D ST, the method effectively models global dependencies within gravity data, enabling the network to reweight features globally and focus on critical anomalous regions. Additionally, prior gradient information is integrated into the loss function, and a hierarchical weight allocation strategy is adopted to guide the model in learning boundary information of density structures and deep-seated features more effectively. Synthetic experiments demonstrate that the proposed method achieves lower model errors, better boundary alignment, and higher inversion accuracy. The approach is further validated using gravity anomaly observations from the Gonghe Basin in Qinghai, yielding reliable and precise inversion results.

Magnetic inversion through three-dimensional (3D) susceptibility reconstruction can effectively identify the deep extension characteristics and structural variations in faults. Therefore, the reliability of inversion results from magnetic anomaly data is a key issue that must be addressed in fault detection and quantitative evaluation of fault activity. In recent years, deep neural network-driven magnetic data inversion methods have rapidly become a research focus in the field of geophysical magnetic data inversion. However, existing methods primarily rely on convolutional neural networks (CNNs), whose inherent local feature extraction capabilities limit their ability to model the spatial continuity of large-scale subsurface magnetic structures. Moreover, the general lack of prior physical constraints in these network models often leads to unreliable inversion results. To address these limitations, this paper proposes a physics-informed multi-scale deep learning inversion method for magnetic anomaly data. The method designs a dual-stream Transformer-CNN fusion module (TCFM). It leverages the self-attention mechanism in Transformers to model global susceptibility correlations while efficiently capturing local geological features through CNN convolutional operations. This enables collaborative modeling of multi-scale subsurface magnetic structures, significantly enhancing inversion accuracy. Furthermore, by incorporating deep physical priors, we design a depth-aware weighted loss function. By strengthening optimization constraints in deep regions, it effectively improves the vertical resolution of inversion models for deep magnetic structures. Comparative experiments with U-Net++ and Transformer demonstrate that the proposed method achieves smaller errors and higher inversion accuracy. Applied to measured aeromagnetic data from the Dandong region of China, the method yields reliable inversion results. Variations in magnetic susceptibility within these results successfully delineate the spatial distribution of fault zones, providing a geophysical basis for regional seismic hazard monitoring and assessment.

Density interface inversion directly reveals the density distribution of underground structures, playing a significant role in the field of geophysics. Deep learning, due to its ability to extract complex and nonlinear density structures, has been widely applied to density interface inversion. Current convolution-based neural network methods for inverting density interfaces are limited by the finite receptive field of the convolution paradigm, neglecting long-range dependencies and posing challenges for high-precision inversion of density interfaces. Additionally, when using deep learning for density interface inversion, different channels contain different information, necessitating the effective utilization of this information variation in the channel dimension. To address these issues, this letter proposes a novel dual-path Transformer network for density interface inversion, named Inverseformer. The Local-Global branch introduces Transformer to establish long-range density dependencies and, combined with the U-net network, jointly captures multi-scale underground density interface features. The Channel branch introduces channel attention mechanism for the model to operate attentively between channels, effectively capturing density interface structural changes in the channels. Synthetic model experiments show that the proposed dual-branch network for inverting density interfaces has smaller model errors and better data fitting, resulting in higher inversion accuracy and more stable inversion results. Subsequently, its effectiveness and value were validated in actual data from the Brittany region of France.

Traditional magnetic data inversion methods are limited by linear approximation theories, which impose constraints on the inversion accuracy and stability. Deep learning (DL), with its powerful nonlinear modeling capabilities, can improve inversion accuracy. However, due to the lack of physical constraints, the inversion results often suffer from insufficient resolution and interpretability. To address these issues, this paper proposes an interpretable high-resolution physics-guided deep learning magnetic inversion method (PDM-Net). The method employs a dual-branch design with "gradient-guided—depth compensation" in its network structure: the first branch introduces a gradient-guided spatial feature attention module (GSFA) based on a convolutional neural network, explicitly embedding magnetic gradient information into the feature modeling process. This emphasizes geological boundaries and key anomaly areas, thus improving the accuracy of spatial structure representation. The second branch, based on a Transformer network, incorporates a depth-compensation Transformer module (DCTM) that integrates the physical law of magnetic response attenuation with depth into the self-attention mechanism, enhancing deep features with weighted reinforcement to improve vertical resolution. Synthetic data experiments show that PDM-Net outperforms other methods in boundary delineation and vertical structure recovery, providing more accurate and interpretable inversion results. In an example using measured magnetic data from the Eastern Himalayan Syntaxis region, PDM-Net reveals the vertical migration and uplift characteristics of high magnetic susceptibility materials in the deep crust, suggesting that this process is closely related to deep crustal uplift and supporting the dynamic mechanism of the "tectonic aneurysm" model.