A method and system for reduction to the pole based on three components of aeromagnetic

Abstract

The application discloses a kind of based on aeromagnetic three-component polarizability method and system, it is related to the field of magnetic prospecting, including: obtaining target area aeromagnetic three-component data and carrying out anomaly extraction;The three-component anomaly data extracted is grid processing, and equivalent rectangular body data is constructed;Based on the equivalent rectangular body data, inversion is carried out using pre-trained inversion model, and the total magnetization intensity three-component data of each rectangular body is obtained;According to the total magnetization intensity three-component data, aeromagnetic three-component polar calculation is carried out, and the final polar result is obtained.The application greatly improves the reliability and engineering applicability of magnetic anomaly polarizability, and provides more efficient and stable technical support for mineral exploration, geological structure analysis and deep geological exploration.

Description

Technical Field

[0001] This invention relates to the field of magnetic exploration technology, and more specifically to a polarization method and system based on the three components of aeromagnetic magnetics. Background Technology

[0002] Magnetic exploration is an important geophysical method for mineral resource exploration, geological structure analysis, and deep geological exploration. Airborne magnetic surveys, with their advantages of wide detection range, high operational efficiency, and minimal interference from terrain, have become the mainstream detection method for large-area regional magnetic field surveys. Magnetic anomaly polarization is a core and fundamental step in magnetic survey data processing. Its purpose is to convert magnetic anomalies under inclined magnetization conditions into vertical magnetization anomalies, eliminating the anomalous offsets and morphological distortions caused by inclined magnetization, and visually reconstructing the spatial location and boundary morphology of underground magnetic geological bodies, providing reliable data support for geological interpretation and resource assessment.

[0003] However, existing magnetic anomaly polarization techniques still have many technical shortcomings. First, most current polarization algorithms rely solely on aeromagnetic total field data for calculations, failing to fully utilize the northward, eastward, and vertical magnetic field components, resulting in low data utilization and loss of magnetic field spatial vector characteristics. For complex geological bodies containing remnant magnetism, it is difficult to accurately determine the true magnetization state, leading to large errors in magnetization direction estimation and easily causing polarization results to shift and become distorted. Second, traditional frequency-domain polarization methods have poor adaptability in low-latitude regions. When the geomagnetic dip angle is small, high-frequency noise is easily amplified, causing problems such as contour stretching and morphological distortion, resulting in insufficient polarization stability. Furthermore, existing magnetic anomaly processing techniques containing remnant magnetism mostly use traditional iterative inversion methods to solve for magnetization parameters. This method has slow convergence speed, high computational load, and is easily affected by observation noise and initial parameters, resulting in poor inversion stability and difficulty in adapting to large-scale aeromagnetic data processing.

[0004] Therefore, how to effectively reduce polarization offset, distortion and aberration problems, and efficiently and stably output polarization results is a problem that urgently needs to be solved by those skilled in the art.

[0005] Problems include low efficiency in inversion operations and imprecise formula logic.

[0006] Therefore, there is an urgent need to develop a magnetic anomaly polarization method and system that relies on aeromagnetic three-component data, has a unified coordinate system, rigorous calculation logic, is adaptable to all latitudes, and can stably correct the influence of residual magnetism. Summary of the Invention

[0007] In view of this, the present invention provides a polarization method and system based on aeromagnetic three-component data. By relying on aeromagnetic three-component data, the method effectively reduces the offset, distortion and aberration problems of polarization results, especially in low-latitude regions, and outputs polarization results efficiently and stably.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: This invention discloses a pole-forming method based on the three aeromagnetic components, comprising: Step 1: Obtain the three-component aeromagnetic data of the target area and perform anomaly extraction; Step 2: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Step 3: Based on the equivalent rectangular body data, perform inversion using a pre-trained inversion model to obtain the three-component data of the total magnetization intensity of each rectangular body; Step 4: Based on the total magnetization three-component data, perform aeromagnetic three-component polarization calculation to obtain the final polarization result.

[0009] Furthermore, step 1 specifically includes: acquiring the original observation data of the northward, eastward, and vertical components of the aeromagnetic survey of the target area; subtracting the normal field component of the international geomagnetic reference field from the original observation data to obtain the three-component anomaly data, as shown in the formula: ; in, , and These are the northward, eastward, and vertical components of the original aeromagnetic observation data, respectively. , and These are the northward, eastward, and vertical components of the international geomagnetic reference field, respectively. , and These are the northward, eastward, and vertical magnetic anomaly components, respectively.

[0010] Furthermore, step 2 specifically includes: taking the extracted northward, eastward, and vertical magnetic anomaly component data as a reference, using the Kriging interpolation method or the minimum curvature interpolation method to perform regular gridding processing, generating plane grid data with equal spacing, and determining the X and Y axis coordinates and grid spacing of the plane grid nodes; Based on the spatial range, grid spacing, and preset depth of the planar grid data, a layer of three-dimensional equivalent rectangles of equal thickness is divided along the Z-axis, and the three-dimensional spatial coordinates and geometric dimensions of each equivalent rectangle are defined. The aeromagnetic three-component anomaly data in the planar grid are matched and assigned to the corresponding three-dimensional equivalent rectangles according to the spatial correspondence, thus completing the construction of the equivalent rectangle data.

[0011] Furthermore, step 3 specifically includes: Based on the equivalent rectangular body data, an input feature matrix with dimensions H×W×1×3 is constructed, where H is the number of rows of the planar grid and W is the number of columns of the planar grid; the input feature matrix is ​​flattened into a two-dimensional input sequence with a sequence length L=H×W and a feature dimension of 3; the two-dimensional input sequence is input into a pre-trained inversion model, and the model outputs the three-component data of the total magnetization intensity of each equivalent rectangular body; The pre-trained inversion model is an improved Transformer model, comprising: an input embedding layer, a position encoding layer, a Transformer encoder stack layer, a global feature fusion layer, and an output mapping layer, which are connected in sequence. The input embedding layer maps the two-dimensional input sequence to a high-dimensional hidden layer feature space, converting it into embedded features of the hidden layer dimension specified by the model. The location encoding layer superimposes a learnable two-dimensional location code onto the embedded features; The Transformer encoder stack layer consists of N layers of encoders with the same structure connected in series, capturing the long-range spatial dependence between different equivalent rectangular bodies and extracting the correlation features between magnetic anomalies and magnetization intensity. The global feature fusion layer: adaptively weights and aggregates all features output by the encoder to generate globally perceptual features; The output mapping layer maps the globally perceived features into three components of total magnetization.

[0012] Furthermore, the single-layer Transformer encoder includes: a normalized LN layer, a multi-head self-attention MHA layer, a first residual connection layer, a layer normalized LN layer, a multilayer perceptron MLP layer, and a second residual connection layer connected in sequence. The standardized LN layer performs layer normalization on the input features; The multi-head self-attention (MHA) layer projects features to multiple subspaces and independently calculates attention weights, fusing spatial correlation features of different scales and directions. The first residual connection layer adds the output features of the multi-head self-attention (MHA) layer to the input features of the layer element by element; The multilayer perceptron (MLP) layer consists of two fully connected layers and an activation function, which performs nonlinear transformation and high-order feature mapping on the features after the first residual connection. The second residual connection layer performs element-wise residual superposition of the output features of the multilayer perceptron (MLP) layer with the input features of the same layer.

[0013] Furthermore, the input embedding layer is represented as: ; in, This indicates the embedded output features. Represents a two-dimensional input sequence. Represents the embedding weight matrix. Represents the embedding bias vector; The location encoding layer is represented as follows: ; in, The output of the position coding layer, This indicates a learnable two-dimensional positional encoding; The single-layer Transformer encoder is represented as: ; ; in, Indicates input features, This represents the output features of the multi-head self-attention (MHA) layer. This represents the final output feature of the encoder; This indicates standardized processing. This indicates multilayer perceptron processing. The formula for multi-head self-attention processing is as follows: ; ; ; in, This represents the mean of the input features along the feature dimension. This represents the variance of the input features along the feature dimension. This is the numerical stability constant, with a value of 10. -6 , This indicates a learnable scaling parameter. This represents the learnable offset parameter. This represents the activation function. and These represent the weight matrices of the first and second layers of the MLP, respectively. and These represent the bias vectors of the first and second layers of the MLP, respectively. This indicates a feature concatenation operation. This represents the multi-head output projection matrix. Indicates the first The output of each attention head; The global feature fusion layer is represented as follows: ; ; in, Indicates the first Each encoder output feature This indicates that the shallow perceptron is learned by weights. Indicates the first Adaptive weights for each output feature Indicates the total number of encoders. Indicates global fusion features; The output mapping layer is represented as follows: ; in, , , These represent the first, second, and third fully connected layers, respectively. Indicates the output vector. , , and These represent the total magnetization at... axis, shaft and The components of the axis.

[0014] Furthermore, the training process of the inversion model is as follows: A simulated aeromagnetic three-component dataset containing different magnetization directions, remanence, and three-dimensional equivalent rectangular spatial distributions was constructed. The dataset was divided into training, validation, and test sets in a 7:1.5:1.5 ratio. The optimization objective was the L1 loss between the model's predicted total magnetization three components and the true values. Iterative training was performed using the Adam optimizer, and the model weights with the minimum validation set loss were used as the final pre-trained inversion model. The loss function was: ; in, This represents the total number of equivalent rectangular volumes for a single sample. , and The first Predicted magnetization for a rectangular prism , and Axial components, , and This corresponds to the actual value.

[0015] Furthermore, step 4 specifically includes: First, the total magnetization amplitude is calculated based on the three components of the total magnetization intensity of each equivalent rectangular body. Then, based on the total magnetization amplitude of each equivalent rectangular body, the magnetic anomaly pole data of each equivalent rectangular body are calculated; Finally, the magnetic anomaly polarization data of each equivalent rectangular body are superimposed to obtain the global polarization result of the observation point.

[0016] Furthermore, the formula for calculating the total magnetization amplitude is as follows: ; Indicates the first The total magnetization amplitude of the rectangular body , and These represent the total magnetization at... , and The components of the axis; The formula for calculating the magnetic anomaly polarization data is: ; in, Indicates the first A rectangular object at the observation point Magnetic anomaly polarization data; The magnetic permeability of vacuum; Indicates the distance from the observation point to the th The distance between the rectangles; Indicates the first The positions of the rectangles vary within the following ranges: and , Indicates the first The coordinates of the center point of each rectangle For a rectangular body in axis, shaft and Side length along the axial direction; The formula for calculating the global polarization result is as follows: ; This indicates that all rectangles are at the observation point. The superimposed global polarization result, This represents the total number of rectangles.

[0017] This invention also discloses a polarization system based on the three aeromagnetic components, comprising: Data acquisition module: Acquires aeromagnetic three-component data of the target area and performs anomaly extraction; Data preprocessing module: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Inversion module: Based on the equivalent rectangular body data, the inversion is performed using a pre-trained inversion model to obtain the three-component data of the total magnetization intensity of each rectangular body; Pole shifting calculation module: Based on the three-component data of the total magnetization intensity, perform aeromagnetic three-component pole shifting calculation to obtain the final pole shifting result.

[0018] As can be seen from the above technical solution, compared with the prior art, the present invention provides a pole-forming method and system based on the three aeromagnetic components, which has the following beneficial effects: This invention fully utilizes aeromagnetic observation data in the north, east, and vertical directions, completely preserving magnetic field vector characteristics and significantly improving data utilization and anomaly interpretation accuracy. By improving the Transformer model, it achieves accurate inversion of magnetization intensity vectors containing remanent magnetization, eliminating the need for manual assumptions about magnetization direction and effectively correcting pole shift and distortion problems caused by remanent magnetization. Employing equivalent rectangular 3D modeling and physical forward calculation, it avoids the defects of traditional frequency-domain pole shifting in low-latitude regions, such as noise amplification and morphological distortion, achieving stable pole shifting across all latitudes. Simultaneously, the model boasts high computational efficiency and strong anti-interference capabilities, adaptable to rapid processing of massive aeromagnetic data over large areas, with unified spatial coordinates and rigorous computational logic. This invention significantly improves the reliability and engineering applicability of magnetic anomaly pole shifting, providing more efficient and stable technical support for mineral exploration, geological structure analysis, and deep geological exploration. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention discloses a pole-forming method based on the three aeromagnetic components, such as... Figure 1 As shown, it includes: Step 1: Obtain the three-component aeromagnetic data of the target area and perform anomaly extraction; Step 2: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Step 3: Based on the equivalent rectangular body data, use the pre-trained inversion model to perform remanent magnetization vector inversion to obtain the three-component data of the total magnetization of each rectangular body; Step 4: Based on the three-component data of total magnetization, perform aeromagnetic three-component polarization calculation to obtain the final polarization result.

[0023] In a specific embodiment, step 1 specifically includes: acquiring the original observation data of the northward, eastward, and vertical components of the aeromagnetic survey of the target area; subtracting the normal field component of the international geomagnetic reference field from the original observation data to obtain the three-component anomaly data, as shown in the formula: ; in, , and These are the northward, eastward, and vertical components of the original aeromagnetic observation data, respectively. , and These are the northward, eastward, and vertical components of the international geomagnetic reference field, respectively. , and These are the northward, eastward, and vertical magnetic anomaly components, respectively.

[0024] In a specific embodiment, step 2 specifically includes: taking the extracted northward, eastward, and vertical magnetic anomaly component data as a reference, using the Kriging interpolation method or the minimum curvature interpolation method to perform regular gridding processing, generating plane grid data with equal spacing, and determining the X and Y axis coordinates of the plane grid nodes and the grid spacing. Based on the spatial range, grid spacing, and preset depth of the planar grid data, a layer of three-dimensional equivalent rectangles of equal thickness is divided along the Z-axis, and the three-dimensional spatial coordinates (X, Y, Z) and geometric dimensions of each equivalent rectangle are defined. The aeromagnetic three-component anomaly data within the planar grid are matched and assigned to the corresponding three-dimensional equivalent rectangles according to the spatial correspondence, thus completing the construction of the equivalent rectangle data.

[0025] In one specific embodiment, step 3 specifically includes: Based on the equivalent rectangular body data, an input feature matrix with dimensions H×W×1×3 is constructed, where H is the number of planar grid rows, W is the number of planar grid columns, 1 indicates that only one layer of equivalent rectangular body is divided vertically, and 3 corresponds to the north, east, and vertical magnetic anomaly components. The input feature matrix is ​​flattened into a two-dimensional input sequence with a sequence length L=H×W and a feature dimension of 3. The two-dimensional input sequence is input into a pre-trained inversion model, and the model outputs the three-component data of the total magnetization intensity of each equivalent rectangular body, thus completing the inversion of the magnetization intensity vector including remanent magnetization. The pre-trained inversion model is an improved Transformer model, which includes: an input embedding layer, a position encoding layer, a stacked Transformer encoder layer, a global feature fusion layer, and an output mapping layer, which are connected in sequence. The input embedding layer maps the two-dimensional input sequence to a high-dimensional hidden layer feature space, converting it into embedded features of the hidden layer dimension specified by the model, thus completing the unification of feature dimensions; The positional encoding layer superimposes learnable two-dimensional positional codes onto the embedded features, endowing each equivalent rectangular volume with spatial positional information while preserving the planar grid topology. Transformer encoder stacking layer, composed of N layers of encoders with the same structure connected in series, captures the long-range spatial dependency between different equivalent rectangular bodies and extracts the correlation features between magnetic anomalies and magnetization intensity; Global Feature Fusion Layer: Adaptively weights and aggregates all features output by the encoder to enhance the representation of key region features and generate globally perceptual features; The output mapping layer maps the globally perceived features into three components of total magnetization with clearly defined physical meaning.

[0026] In one specific embodiment, a single-layer Transformer encoder includes: a normalized LN layer, a multi-head self-attention MHA layer, a first residual connection layer, a layer normalized LN layer, a multilayer perceptron MLP layer, and a second residual connection layer connected in sequence. Standardized LN layers perform layer normalization on the input features, eliminating feature distribution bias and scale differences, stabilizing the model training process, and improving the feature extraction stability and convergence speed of the multi-head self-attention module. The multi-head self-attention (MHA) layer is used to model multi-scale spatial dependencies in parallel. It projects features to multiple subspaces and independently calculates attention weights, integrates spatial correlation features of different scales and directions, and accurately captures the long-range spatial correlation and local detail features of aeromagnetic anomaly fields. The first residual connection layer adds the output features of the multi-head self-attention (MHA) layer to the input features of this layer element by element to construct a residual path, which alleviates the gradient vanishing problem in deep network training and preserves the integrity of the original magnetic anomaly features. The multilayer perceptron (MLP) layer consists of two fully connected layers and an activation function. It performs nonlinear transformation and high-order feature mapping on the features after the first residual connection to uncover the implicit physical relationship between magnetic anomalies and magnetization, thereby improving the feature representation capability. The second residual connection layer performs element-wise residual superposition of the output features of the multilayer perceptron (MLP) layer with the input features of this layer, further stabilizing network training, enhancing feature transfer efficiency, and ensuring the feature extraction effect of the deep encoder.

[0027] In a specific embodiment, the input embedding layer is represented as follows: ; in, This indicates the embedded output features. Represents a two-dimensional input sequence. Represents the embedding weight matrix. Represents the embedding bias vector; The position coding layer is represented as: ; in, The output of the position coding layer, This indicates a learnable two-dimensional positional encoding; A single-layer Transformer encoder is represented as: ; ; in, Indicates input features, This represents the output features of the multi-head self-attention (MHA) layer. This represents the final output feature of the encoder; This indicates standardized processing. This indicates multilayer perceptron processing. The formula for multi-head self-attention processing is as follows: ; ; ; in, This represents the mean of the input features along the feature dimension. This represents the variance of the input features along the feature dimension. This is the numerical stability constant, with a value of 10. -6 , This indicates a learnable scaling parameter. This represents the learnable offset parameter. Represents the Gaussian error activation function. and These represent the weight matrices of the first and second layers of the MLP, respectively. and These represent the bias vectors of the first and second layers of the MLP, respectively. This indicates a feature concatenation operation. This represents the multi-head output projection matrix. Indicates the first The output of each attention head is: ; For query matrix, For the key matrix, For value matrices, , and These are the corresponding linear transformation matrices. For the feature dimension of the attention head; The global feature fusion layer is represented as: ; ; in, Indicates the first Each encoder output feature This indicates that the shallow perceptron is learned by weights. Indicates the first Adaptive weights for each output feature Indicates the total number of encoders. Indicates global fusion features; The output mapping layer is represented as: ; in, , , These represent the first, second, and third fully connected layers, respectively. Indicates the output vector. , , and These represent the total magnetization at... axis, shaft and The components of the axis.

[0028] In one specific embodiment, the training process of the inversion model is as follows: A simulated aeromagnetic three-component dataset was constructed, containing different magnetization directions, remanence, and spatial distributions of a three-dimensional equivalent rectangular body. The dataset samples include north-, east-, and vertical magnetic anomaly components consistent with actual exploration data, along with corresponding ground truth labels for the total magnetization three-component. The dataset was divided into training, validation, and test sets in a 7:1.5:1.5 ratio. The optimization objective was the L1 loss between the model's predicted total magnetization three-component and the true values. Iterative training was performed using the Adam optimizer with a fixed learning rate of 1×10⁻⁶. -4 The model weights that minimize the loss on the validation set are used as the final pre-trained inversion model; the loss function is: ; in, This represents the total number of equivalent rectangular volumes for a single sample. , and The first Predicted magnetization for a rectangular prism , and Axial components, , and This corresponds to the actual value.

[0029] In one specific embodiment, step 4 specifically includes: First, the total magnetization amplitude is calculated based on the three components of the total magnetization intensity of each equivalent rectangular body. Then, based on the total magnetization amplitude of each equivalent rectangular body, the magnetic anomaly pole data of each equivalent rectangular body are calculated; Finally, the magnetic anomaly polarization data of each equivalent rectangular body are superimposed to obtain the global polarization result of the observation point.

[0030] In a specific embodiment, the formula for calculating the total magnetization amplitude is as follows: ; Indicates the first The total magnetization amplitude of the rectangular body , and These represent the total magnetization at... , and The components of the axis; The formula for calculating magnetic anomaly polarization data is: ; in, Indicates the first A rectangular object at the observation point Magnetic anomaly polarization data; The magnetic permeability of vacuum; Indicates the distance from the observation point to the th The distance between the rectangles; Indicates the first The positions of the rectangles vary within the following ranges: and , Indicates the first The coordinates of the center point of each rectangle For a rectangular body in axis, shaft and Side length along the axial direction; The formula for calculating the global polarization result is: ; This indicates that all rectangles are at the observation point. The superimposed global polarization result, This represents the total number of rectangles.

[0031] This invention also discloses a polarization system based on the three aeromagnetic components, comprising: Data acquisition module: Acquires aeromagnetic three-component data of the target area and performs anomaly extraction; Data preprocessing module: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Inversion module: Based on the equivalent rectangular data, the inversion is performed using a pre-trained inversion model to obtain the three-component data of the total magnetization of each rectangular body; Pole shifting calculation module: Based on the three components of the total magnetization, perform aeromagnetic three-component pole shifting calculation to obtain the final pole shifting result.

[0032] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0033] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A pole-setting method based on the three aeromagnetic components, characterized in that, include: Step 1: Obtain the three-component aeromagnetic data of the target area and perform anomaly extraction; Step 2: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Step 3: Based on the equivalent rectangular body data, perform inversion using a pre-trained inversion model to obtain the three-component data of the total magnetization intensity of each rectangular body; Step 4: Based on the total magnetization three-component data, perform aeromagnetic three-component polarization calculation to obtain the final polarization result.

2. The pole-setting method based on the three aeromagnetic components according to claim 1, characterized in that, Step 1 specifically includes: acquiring the original observation data of the northward, eastward, and vertical components of the aeromagnetic survey of the target area; subtracting the normal field component of the International Geomagnetic Reference Field from the original observation data to obtain the three-component anomaly data, using the following formula: ； in, , and These represent the northward, eastward, and vertical components of the original aeromagnetic observation data, respectively. , and These are the northward, eastward, and vertical components of the international geomagnetic reference field, respectively. , and These are the northward, eastward, and vertical magnetic anomaly components, respectively.

3. The pole-setting method based on the three aeromagnetic components according to claim 1, characterized in that, Step 2 specifically includes: taking the extracted northward, eastward, and vertical magnetic anomaly component data as a reference, using the Kriging interpolation method or the minimum curvature interpolation method to perform regular gridding processing, generating plane grid data with equal spacing, and determining the X and Y axis coordinates and grid spacing of the plane grid nodes; Based on the spatial range, grid spacing, and preset depth of the planar grid data, a layer of three-dimensional equivalent rectangles of equal thickness is divided along the Z-axis, and the three-dimensional spatial coordinates and geometric dimensions of each equivalent rectangle are defined. The aeromagnetic three-component anomaly data in the planar grid are matched and assigned to the corresponding three-dimensional equivalent rectangles according to the spatial correspondence, thus completing the construction of the equivalent rectangle data.

4. The pole-setting method based on the three aeromagnetic components according to claim 1, characterized in that, Step 3 specifically includes: Based on the equivalent rectangular body data, an input feature matrix with dimensions H×W×1×3 is constructed, where H is the number of rows of the planar grid and W is the number of columns of the planar grid; the input feature matrix is ​​flattened into a two-dimensional input sequence with a sequence length L=H×W and a feature dimension of 3; the two-dimensional input sequence is input into a pre-trained inversion model, and the model outputs the three-component data of the total magnetization intensity of each equivalent rectangular body; The pre-trained inversion model is an improved Transformer model, comprising: an input embedding layer, a position encoding layer, a Transformer encoder stack layer, a global feature fusion layer, and an output mapping layer, which are connected in sequence. The input embedding layer maps the two-dimensional input sequence to a high-dimensional hidden layer feature space, converting it into embedded features of the hidden layer dimension specified by the model. The location encoding layer superimposes a learnable two-dimensional location code onto the embedded features; The Transformer encoder stack layer consists of N layers of encoders with the same structure connected in series, capturing the long-range spatial dependence between different equivalent rectangular bodies and extracting the correlation features between magnetic anomalies and magnetization intensity. The global feature fusion layer: adaptively weights and aggregates all features output by the encoder to generate globally perceptual features; The output mapping layer maps the globally perceived features into three components of total magnetization.

5. The pole-setting method based on the three aeromagnetic components according to claim 4, characterized in that, The single-layer Transformer encoder comprises: a normalized LN layer, a multi-head self-attention MHA layer, a first residual connection layer, a layer normalized LN layer, a multilayer perceptron MLP layer, and a second residual connection layer connected in sequence. The standardized LN layer performs layer normalization on the input features; The multi-head self-attention (MHA) layer projects features to multiple subspaces and independently calculates attention weights, fusing spatial correlation features of different scales and directions. The first residual connection layer adds the output features of the multi-head self-attention (MHA) layer to the input features of the layer element by element; The multilayer perceptron (MLP) layer consists of two fully connected layers and an activation function, which performs nonlinear transformation and high-order feature mapping on the features after the first residual connection. The second residual connection layer performs element-wise residual superposition of the output features of the multilayer perceptron (MLP) layer with the input features of the same layer.

6. The pole-setting method based on the three aeromagnetic components according to claim 5, characterized in that, The input embedding layer is represented as follows: ； in, This indicates the embedded output features. Represents a two-dimensional input sequence. Represents the embedding weight matrix. Represents the embedding bias vector; The location encoding layer is represented as follows: ； in, The output of the position coding layer, This indicates a learnable two-dimensional positional encoding; The single-layer Transformer encoder is represented as: ； ； in, Indicates input features, This represents the output features of the multi-head self-attention (MHA) layer. This represents the final output feature of the encoder; This indicates standardized processing. This indicates multilayer perceptron processing. The formula for multi-head self-attention processing is as follows: ； ； ； in, This represents the mean of the input features along the feature dimension. This represents the variance of the input features along the feature dimension. This is the numerical stability constant, with a value of 10. -6 , This indicates a learnable scaling parameter. This represents the learnable offset parameter. This represents the activation function. and These represent the weight matrices of the first and second layers of the MLP, respectively. and These represent the bias vectors of the first and second layers of the MLP, respectively. This indicates a feature concatenation operation. This represents the multi-head output projection matrix. Indicates the first The output of each attention head; The global feature fusion layer is represented as follows: ； ； in, Indicates the first Each encoder output feature This indicates that the shallow perceptron is learned by weights. Indicates the first Adaptive weights for each output feature Indicates the total number of encoders. Indicates global fusion features; The output mapping layer is represented as follows: ； in, , , These represent the first, second, and third fully connected layers, respectively. Indicates the output vector. , , and These represent the total magnetization at... axis, shaft and The components of the axis.

7. The pole-setting method based on the three aeromagnetic components according to claim 1, characterized in that, The training process of the inversion model is as follows: A simulated aeromagnetic three-component dataset containing different magnetization directions, remanence, and three-dimensional equivalent rectangular spatial distributions was constructed. The dataset was divided into training, validation, and test sets in a 7:1.5:1.5 ratio. The optimization objective was the L1 loss between the model's predicted total magnetization three components and the true values. Iterative training was performed using the Adam optimizer, and the model weights with the minimum validation set loss were used as the final pre-trained inversion model. The loss function was: ； in, This represents the total number of equivalent rectangular volumes for a single sample. , and The first Predicted magnetization for a rectangular prism , and Axial components, , and This corresponds to the actual value.

8. The pole-setting method based on the three aeromagnetic components according to claim 1, characterized in that, Step 4 specifically includes: First, the total magnetization amplitude is calculated based on the three components of the total magnetization intensity of each equivalent rectangular body. Then, based on the total magnetization amplitude of each equivalent rectangular body, the magnetic anomaly pole data of each equivalent rectangular body are calculated; Finally, the magnetic anomaly polarization data of each equivalent rectangular body are superimposed to obtain the global polarization result of the observation point.

9. The pole-setting method based on the three aeromagnetic components according to claim 8, characterized in that, The formula for calculating the total magnetization amplitude is as follows: ； Indicates the first The total magnetization amplitude of the rectangular body , and These represent the total magnetization at... , and The components of the axis; The formula for calculating the magnetic anomaly polarization data is: ； in, Indicates the first A rectangular object at the observation point Magnetic anomaly polarization data; The magnetic permeability of vacuum; Indicates the distance from the observation point to the th The distance between the rectangles; Indicates the first The positions of the rectangles vary within the following ranges: and , Indicates the first The coordinates of the center point of each rectangle For a rectangular body in axis, shaft and Side length along the axial direction; The formula for calculating the global polarization result is as follows: ； This indicates that all rectangles are at the observation point. The superimposed global polarization result, This represents the total number of rectangles.

10. A pole-setting system based on aeromagnetic three-dimensional coordinates, employing the pole-setting method based on aeromagnetic three-dimensional coordinates as described in any one of claims 1 to 9, characterized in that, include: Data acquisition module: Acquires aeromagnetic three-component data of the target area and performs anomaly extraction; Data preprocessing module: The extracted three-component outlier data is processed into a grid to construct an equivalent rectangular data structure; Inversion module: Based on the equivalent rectangular body data, the inversion is performed using a pre-trained inversion model to obtain the three-component data of the total magnetization intensity of each rectangular body; Pole shifting calculation module: Based on the three-component data of the total magnetization intensity, perform aeromagnetic three-component pole shifting calculation to obtain the final pole shifting result.

Images