An Adaptive Octree 3D Magnetic Inversion Method Integrating Anomaly Region Identification

Abstract

An adaptive octree-based 3D magnetic inversion method integrating anomaly region identification belongs to the field of 3D magnetic data inversion processing technology. It includes rapidly performing an initial inversion on a coarse grid to obtain the overall outline of the subsurface structure, and introducing a fuzzy c-means clustering algorithm to automatically extract target regions with significant magnetic anomalies from the coarse solution. A multi-level octree grid is constructed within the extracted anomaly regions to achieve fine subdivision of complex geological boundaries, and the fine model is used as a new starting point for high-precision inversion. The adaptive octree grid features high efficiency, a small number of grid cells, and low data fitting error. This "coarse inversion - intelligent identification - local refinement - fine inversion" process can be executed cyclically as needed, dynamically allocating computational resources. Compared with the traditional global fine-grid scheme, this method significantly reduces the number of grid cells while maintaining or even improving inversion accuracy.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional magnetic data inversion processing technology, and in particular to an adaptive octree three-dimensional magnetic inversion method that integrates anomaly region identification. Background Technology

[0002] Magnetic methods, a key tool in geophysical exploration, rely on magnetic anomalies caused by differences in the magnetic properties of subsurface media. By observing magnetic potential field data, they infer the distribution of subsurface structures and anomalies, and are widely used in mineral resource exploration and regional geological structure analysis. Magnetic inversion is the core step in quantitatively interpreting measured magnetic field data, aiming to reconstruct the location, shape, and physical properties of subsurface magnetic bodies. It serves as a crucial bridge connecting physical observation and geological interpretation. In three-dimensional magnetic inversion, subsurface space is typically discretized into a large number of grid cells, and the physical properties of each grid cell are estimated by minimizing the objective function. However, subsurface geological bodies have complex morphologies. High-precision characterization of anomaly boundaries using regular hexahedral meshes necessitates meticulous mesh partitioning in the study area, leading to a rapid increase in model degrees of freedom, a significant rise in computational complexity, and higher demands on storage resources. In recent years, numerous studies have focused on improving the computational efficiency of magnetic forward and inversion methods. Changing the model's mesh partitioning can fundamentally reduce model size and improve inversion efficiency. Locally refined meshing allows for fine meshing only within critical regions, while maintaining a coarse mesh over larger areas, significantly reducing computational resources used in non-critical regions. However, due to variations in mesh volume, locally refined meshing can lead to convergence instability during inversion and cannot utilize the smoothing constraint formulas for regular hexahedral meshes. x , y , z Applying difference operators between adjacent meshes to construct a smoothing matrix raises the crucial question of how to incorporate volumetric influence and apply smoothing constraints. Furthermore, local mesh refinement still presents the problem of "where to refine," with existing methods primarily focusing on extracting key regions in a specific direction (horizontal or vertical), falling short in terms of full-space 3D refinement. Therefore, determining the optimal method for extracting key regions to achieve efficient and high-precision 3D magnetic inversion remains a challenging problem. Summary of the Invention

[0003] The purpose of this invention is to solve the problems of low efficiency and computational waste in existing large-scale three-dimensional magnetic data inversion, and to provide an adaptive octree three-dimensional magnetic inversion method that integrates anomaly region identification.

[0004] An adaptive octree-based three-dimensional magnetic inversion method integrating anomaly region identification includes the following steps:

[0005] Step 1: Input magnetic observation data, input the magnetization intensity model of the initial coarse grid, input the number of grid refinements, initial regularization factor, focusing factor, and depth weighting parameters;

[0006] Step 2: Introduce a depth weighting function and a smoothing focusing regularization method to construct the objective function for 3D magnetic data inversion. ;

[0007] Step 3: Optimize the objective function of the 3D magnetic data inversion using the Gauss-Newton method to obtain the model iterative expression, and perform initial inversion iteration to obtain the initial coarse mesh inversion results;

[0008] Step 4: Combine fuzzy c-means clustering algorithm and adaptive extremum technique to separate target region and abnormal region, so as to realize intelligent extraction of abnormal region;

[0009] Step 5: Refine the abnormal region using an octree mesh to obtain a locally refined mesh model;

[0010] Step 6: For non-uniformly partitioned octree meshes, smoothing constraints are applied by applying difference operators to adjacent meshes respectively;

[0011] Step 7: Use the locally refined mesh model as the initial model for inversion iteration to obtain high-precision inversion results, and calculate the data fitting difference to determine whether to continue the "intelligent recognition-local refinement-fine inversion" process.

[0012] In step two, the objective function for the inversion of three-dimensional magnetic data is constructed, and its expression is as follows:

[0013] ;

[0014] Among them, the diagonal matrix Weighted matrix of data, Let be the normalized observation error at the i-th point; d represents the forward response; d represents the observed data; m represents the vector of physical property parameters; m ref The reference model is used; the regularization factor is used. Weighting matrix for the model;

[0015] Because the volume of the octree mesh varies, the inversion of the refined mesh is not very stable. Therefore, a depth-weighted function related to the volume is used, as shown below:

[0016] ;

[0017] in, V j For the first j The volume of each grid cell; rij For the first i The grid number to the first j The distance between observation points; β This is a constant used to control the strength of the weighting function, typically 0.5 < 0.5. β <1.5, N For the number of observation data, M The number of mesh subdivisions for the model;

[0018] Furthermore, a smooth focusing regularization method is employed to obtain inversion results with clearer anomaly boundaries while ensuring convergence stability. The expression for the model weighting matrix is ​​as follows:

[0019] ;

[0020] ;

[0021] ;

[0022] in, The focus portion of smooth cluster regularization is represented by ; RRR represents the smooth portion of smooth focus regularization; e R is the focus factor, used to control the degree of focus in the inversion model. x R y and R z They are respectively x , y , z The difference operator applied between adjacent grids in the direction.

[0023] In step three, the objective function formula (1) is optimized using the Gauss-Newton method, resulting in the following iterative model expression:

[0024] ;

[0025] Where J is the Jacobian matrix. For the first k The change in model parameters in each iteration;

[0026] To improve the efficiency of 3D magnetic inversion, an initial inversion is first performed on a coarse grid to quickly obtain the overall outline of the underground structure, preparing for subsequent intelligent identification of anomaly areas.

[0027] In step four, the fuzzy c-means clustering algorithm can automatically calculate the membership degree of each grid's physical property value relative to the cluster center based on the cluster center; the adaptive extremum technique can adaptively determine the initial cluster center and the number of cluster centers; through the fusion of the fuzzy c-means clustering algorithm and the adaptive extremum technique, the intelligent extraction of abnormal regions can be accurately achieved.

[0028] The objective function expression for the fuzzy c-means clustering algorithm is as follows:

[0029] ;

[0030] ;

[0031] in, n For the mesh subdivision fraction, p This represents the number of clusters. u ij Represented as the first i Model parameters m i Compared to the first j Cluster centers v j membership degree q For fuzzy factors;

[0032] The adaptive extremum technique first selects the values ​​of the region with the densest distribution of model property values ​​as background values ​​'a', and replaces the property values ​​close to 'a' with 'a' to enhance the data's discriminative power. Then, it extracts the extreme points in the model as the center points of outliers, and uses these to determine the value and number of cluster centers.

[0033] After determining the value and number of cluster centers, the membership matrix is ​​obtained by optimizing the objective function (7):

[0034] ;

[0035] ;

[0036] in, t ij For the first i Model parameters m i Compared to the first j Cluster centers v j The Euclidean distance; t ip For the first i Physical property values ​​of each grid m i Compared to the first p Cluster centers v p The Euclidean distance;

[0037] Finally, based on the principle of maximum membership, the physical property values ​​can be distinguished into background regions and abnormal regions, thus realizing intelligent extraction of abnormal regions from low-precision inversion models.

[0038] In step six, since one grid in the octree mesh may be adjacent to multiple grids and cannot be calculated by formula, we first find all the grids adjacent to each grid and then apply the difference operator to ensure the smoothness of the inversion result in any direction in space.

[0039] In step seven, the data fitting difference is used to quantitatively evaluate the inversion results, and its expression is as follows:

[0040] ;

[0041] Where, d obs For observation data, d inv To predict the forward response from the inversion data, err To account for observation errors, if the inversion result still does not meet the expected data fit difference requirement, the process of "intelligent identification - local refinement - fine inversion" will continue to achieve dynamic refinement of the grid and allocation of computing resources during the inversion process, so as to gradually improve the inversion accuracy.

[0042] The beneficial effects of this invention are:

[0043] 1. This invention is based on fuzzy c-means clustering technology and an adaptive cluster center determination method, which can accurately distinguish the background region and target region of the model and realize intelligent extraction of abnormal regions.

[0044] 2. In order to solve the instability problem caused by the volume difference of the octree mesh, this invention introduces a depth weighting matrix that takes into account the volume effect and a smoothing focusing regularization method, which improves the stability of the inversion while ensuring the accuracy of the inversion.

[0045] 3. The adaptive octree mesh partitioning process of the present invention can dynamically adjust the range of the extracted abnormal regions during the inversion iteration process, and gradually refine the inversion mesh to achieve full-space refinement of the mesh. Attached Figure Description

[0046] Figure 1 This is a flowchart of an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram illustrating the application of smooth constraints to the regular hexahedral mesh described in the background section of this invention.

[0048] Figure 3 A schematic diagram illustrating the application of smoothing constraints to an octree mesh according to an embodiment of the present invention;

[0049] Figure 4 This is a three-dimensional schematic diagram of the V-shaped model according to an embodiment of the present invention;

[0050] Figure 5This is a top view schematic diagram showing the distribution of measuring points on the V-shaped model according to an embodiment of the present invention;

[0051] Figure 6 This is a diagram showing the correspondence between the real model and the magnetic response to be fitted in the V-shaped model in this embodiment, where a is a schematic diagram of the real model in this embodiment and b is a diagram of the magnetic response to be fitted in the real model.

[0052] Figure 7 The correspondence between the inversion result slice of the uniform coarse grid at x=5km and the predicted magnetic response of the uniform coarse grid inversion result in the V-shaped model of the present invention is shown in the figure. Here, a is the slice of the inversion result of the uniform coarse grid at x=5km and b is the predicted magnetic response of the inversion result of the uniform coarse grid.

[0053] Figure 8 This is a diagram showing the correspondence between the inversion results of the uniform fine mesh at x=5km in the V-shaped model of this invention and the predicted magnetic response of the uniform fine mesh inversion results, where a is a slice of the inversion results of the uniform fine mesh at x=5km and b is the predicted magnetic response of the inversion results of the uniform fine mesh.

[0054] Figure 9 This is a diagram showing the correspondence between the inversion result slice of the adaptive octree mesh at x=5km in the V-shaped model of this invention and the predicted magnetic response of the adaptive octree mesh inversion result, where a is a slice of the inversion result of the adaptive octree mesh at x=5km and b is the predicted magnetic response of the inversion result of the adaptive octree mesh. Detailed Implementation

[0055] Please see Figure 1 , Figures 3 to 9 The image shown is an embodiment of the present invention.

[0056] An adaptive octree-based three-dimensional magnetic inversion method integrating anomaly region identification includes the following steps:

[0057] Step 1: Input magnetic observation data, input the magnetization intensity model of the initial coarse grid, input the number of grid refinements, initial regularization factor, focusing factor, and depth weighting parameters;

[0058] Step 2: Introduce a depth weighting function and a smoothing focusing regularization method to construct the objective function for 3D magnetic data inversion. ;

[0059] Step 3: Optimize the objective function of the 3D magnetic data inversion using the Gauss-Newton method to obtain the model iterative expression, and perform initial inversion iteration to obtain the initial coarse mesh inversion results;

[0060] Step 4: Combine fuzzy c-means clustering algorithm and adaptive extremum technique to separate target region and abnormal region, so as to realize intelligent extraction of abnormal region;

[0061] Step 5: Refine the abnormal region using an octree mesh to obtain a locally refined mesh model;

[0062] Step 6: For non-uniformly partitioned octree meshes, smoothing constraints are applied by applying difference operators to adjacent meshes respectively;

[0063] Step 7: Use the locally refined mesh model as the initial model for inversion iteration to obtain high-precision inversion results, and calculate the data fitting difference to determine whether to continue the "intelligent recognition-local refinement-fine inversion" process.

[0064] In step two, the objective function for the inversion of three-dimensional magnetic data is constructed, and its expression is as follows:

[0065] ;

[0066] Among them, the diagonal matrix Weighted matrix of data, Let be the normalized observation error at the i-th point; d represents the forward response; d represents the observed data; m represents the vector of physical property parameters; m ref For reference model; α As a regularization factor; Weighting matrix for the model;

[0067] Because the volume of the octree mesh varies, the inversion of the refined mesh is not very stable. Therefore, a depth-weighted function related to the volume is used, as shown below:

[0068] ;

[0069] in, V j For the first j The volume of each grid cell; r ij For the first i The grid number to the first j The distance between observation points; β This is a constant used to control the strength of the weighting function, typically 0.5 < 0.5. β <1.5, N For the number of observation data, M The number of mesh subdivisions for the model;

[0070] Furthermore, a smooth focusing regularization method is employed to obtain inversion results with clearer anomaly boundaries while ensuring convergence stability. The expression for the model weighting matrix is ​​as follows:

[0071] ;

[0072] ;

[0073] ;

[0074] Among them, R e The focus portion of smooth cluster regularization is represented by ; RRR represents the smooth portion of smooth focus regularization; e R is the focus factor, used to control the degree of focus in the inversion model. x R y and R z They are respectively x , y , z The difference operator applied between adjacent grids in the direction.

[0075] In step three, the objective function formula (1) is optimized using the Gauss-Newton method, resulting in the following iterative model expression:

[0076] ;

[0077] Where J is the Jacobian matrix. For the first k The change in model parameters in each iteration;

[0078] To improve the efficiency of 3D magnetic inversion, an initial inversion is first performed on a coarse grid to quickly obtain the overall outline of the underground structure, preparing for subsequent intelligent identification of anomaly areas.

[0079] In step four, the fuzzy c-means clustering algorithm can automatically calculate the membership degree of each grid's physical property value relative to the cluster center based on the cluster center; the adaptive extremum technique can adaptively determine the initial cluster center and the number of cluster centers; through the fusion of the fuzzy c-means clustering algorithm and the adaptive extremum technique, the intelligent extraction of abnormal regions can be accurately achieved.

[0080] The objective function expression for the fuzzy c-means clustering algorithm is as follows:

[0081] ;

[0082] ;

[0083] in, n For the mesh subdivision fraction, p This represents the number of clusters. u ij Represented as the first i Model parametersm i Compared to the first j Cluster centers v j membership degree q For fuzzy factors;

[0084] The adaptive extremum technique first selects the values ​​of the region with the densest distribution of model property values ​​as background values ​​'a', and replaces the property values ​​close to 'a' with 'a' to enhance the data's discriminative power. Then, it extracts the extreme points in the model as the center points of outliers, and uses these to determine the value and number of cluster centers.

[0085] After determining the value and number of cluster centers, the membership matrix is ​​obtained by optimizing the objective function (7):

[0086] ;

[0087] ;

[0088] in, t ij For the first i Model parameters m i Compared to the first j Cluster centers v j The Euclidean distance; t ip For the first i Physical property values ​​of each grid m i Compared to the first p Cluster centers v p The Euclidean distance;

[0089] Finally, based on the principle of maximum membership, the physical property values ​​can be distinguished into background regions and abnormal regions, thus realizing intelligent extraction of abnormal regions from low-precision inversion models.

[0090] In step six, since one grid in the octree mesh may be adjacent to multiple grids and cannot be calculated by formula, we first find all the grids adjacent to each grid and then apply the difference operator to ensure the smoothness of the inversion result in any direction in space.

[0091] In step seven, the data fitting difference is used to quantitatively evaluate the inversion results, and its expression is as follows:

[0092] ;

[0093] Where, d obs For observation data, dinv To predict the forward response from the inversion data, err To account for observation errors, if the inversion result still does not meet the expected data fit difference requirement, the process of "intelligent identification - local refinement - fine inversion" will continue to achieve dynamic refinement of the grid and allocation of computing resources during the inversion process, so as to gradually improve the inversion accuracy.

[0094] In an octree mesh, a single cell may be adjacent to multiple cells, making direct calculation impossible. Therefore, it's necessary to first identify all adjacent cells for each cell and then apply difference operators separately to ensure the smoothness of the inversion result in any spatial direction. Figure 3 As shown.

[0095] The adaptive octree three-dimensional magnetic inversion method for fusion anomaly region identification provided by this invention will be verified below.

[0096] Adopting such Figure 4 The V-shaped model shown was used for trial calculations. The study area, with a background magnetization of 0 A / m and a size of 10 km × 10 km × 6 km, was divided into 50 × 50 × 60 hexahedral units, and the magnetic tilt and magnetic declination were set to 70° and 60°, respectively. This included a V-shaped model with a magnetization of 1 A / m. x The directional distribution length is 1.6 km (the anomaly length is 1.6 km, and its x-axis coordinate in the coordinate system is between 4.2 km and 5.8 km). y The directional distribution length is 3.6km (that is, the length of the anomaly is 3.6km, and the y-axis coordinate in the coordinate system is between 3.2km and 6.8km), and the top burial depth is about 0.4km. Figure 5 The distribution of observation points is shown from a top-down perspective, with the 20×20 black dots representing observation points deployed on the ground.

[0097] Using a uniform half-space model (0A / m) as the initial model, 3D inversion experiments were conducted with uniform coarse mesh, uniform fine mesh, and adaptive octree mesh to verify the effectiveness of the adaptive octree inversion. The uniform coarse mesh was divided into 20×20×15 hexahedral elements, with depth weighting parameters... β =1, Focusing factor e =0.02, the inversion iteration count is 30. The uniform fine mesh is divided into 50×50×40 hexahedral elements, depth weighting parameter β =1.2, Focusing Factor e=0.03, the inversion iteration count was 20. The adaptive octree mesh was refined twice. Its initial mesh was the same as the uniform coarse mesh. To ensure that the results obtained from the initial mesh inversion could more macroscopically reflect the distribution range of the anomaly and to prevent the extracted anomaly region from failing to cover the distribution range of the anomaly, the initial depth weighting parameter was set to 0.03. β =0.8, focusing factor e =0.05, initial iteration count is 10; depth-weighted parameter after refinement β =1, Focusing factor e =0.02, the number of iterations for each refined mesh is 5 more than the number of iterations for the previous mesh.

[0098] Figure 6 This section presents a schematic diagram of the real model and the corresponding magnetic response diagram to be fitted. Figures 7 to 9 The inversion model slices for each grid at x=5km are shown, with white lines representing the extent of the anomaly. The right side shows the predicted magnetic response map corresponding to the inversion model for each grid. Here, x and y represent the horizontal direction, z represents the vertical direction, κ represents the magnetization, and ΔT represents the magnetic anomaly response value.

[0099] Figure 6 In the diagram, a is a schematic diagram of the actual V-shaped model in an embodiment of the present invention, and b is the magnetic response diagram of the actual model to be fitted; the two correspond to each other.

[0100] Figure 7 In the diagram, a is a slice of the inversion result of the uniform coarse grid at x=5km in the V-shaped model of the present invention, and b is the predicted magnetic response diagram of the inversion result of the uniform coarse grid. The two are corresponding.

[0101] Figure 8 In the diagram, a is a slice of the inversion result of the uniform fine mesh at x=5km in the V-shaped model of the present invention, and b is the predicted magnetic response diagram of the inversion result of the uniform fine mesh. The two correspond to each other.

[0102] Figure 9 In the diagram, a is a slice of the inversion result of the adaptive octree mesh at x=5km in the V-shaped model of this invention, and b is the predicted magnetic response diagram of the inversion result of the adaptive octree mesh. The two are corresponding.

[0103] from Figures 6 to 9The results show that a uniform coarse mesh cannot accurately determine the extent of anomalies, a uniform fine mesh inverts anomalies that are lower in the direction of the anomaly, while an adaptive octree mesh more accurately characterizes the extent of the anomaly. Compared to a uniform fine mesh, the adaptive octree mesh (computation time: 173.8s; final number of inverted meshes: 9192; data fit difference: 0.594) is significantly more computationally efficient than the uniform fine mesh (computation time: 37291.1s; final number of inverted meshes: 100000; data fit difference: 0.693). It is easy to see that the adaptive octree mesh has the characteristics of high efficiency, fewer meshes, and smaller data fit difference.

Claims

1. An adaptive octree three-dimensional magnetic inversion method integrating anomaly region identification, characterized in that, Includes the following steps: Step 1: Input magnetic observation data, input the magnetization intensity model of the initial coarse grid, input the number of grid refinements, initial regularization factor, focusing factor, and depth weighting parameters; Step 2: Introduce a depth weighting function and a smoothing focusing regularization method to construct the objective function for 3D magnetic data inversion. ; Step 3: Optimize the objective function of the 3D magnetic data inversion using the Gauss-Newton method to obtain the model iterative expression, and perform initial inversion iteration to obtain the initial coarse mesh inversion results; Step 4: Combine fuzzy c-means clustering algorithm and adaptive extremum technique to separate target region and abnormal region, so as to realize intelligent extraction of abnormal region; Step 5: Refine the abnormal region using an octree mesh to obtain a locally refined mesh model; Step 6: For non-uniformly partitioned octree meshes, smoothing constraints are applied by applying difference operators to adjacent meshes respectively; Step 7: Use the locally refined mesh model as the initial model for inversion iteration to obtain high-precision inversion results, and calculate the data fitting difference to determine whether to continue the "intelligent recognition-local refinement-fine inversion" process.

2. The adaptive octree three-dimensional magnetic inversion method for fusing anomaly region identification as described in claim 1, characterized in that: In step two, the objective function for the inversion of three-dimensional magnetic data is constructed, and its expression is as follows: ; Among them, the diagonal matrix Weighted matrix of data, For the first i Normalized observation error at each point; d represents the forward response; d represents the observed data; m represents the vector of physical property parameters; m ref This is the reference model; α is the regularization factor. Weighting matrix for the model; Because the volume of the octree mesh varies, the inversion of the refined mesh is not very stable. Therefore, a depth-weighted function related to the volume is used, as shown below: ； in, V j For the first j The volume of each grid cell; r ij For the first i The grid number to the first j The distance between observation points; β This is a constant used to control the strength of the weighting function, typically 0.5 < 0.

5. β <1.5, where N is the number of observation data and M is the number of model grid subdivisions; Furthermore, a smooth focusing regularization method is employed to obtain inversion results with clearer anomaly boundaries while ensuring convergence stability. The expression for the model weighting matrix is ​​as follows: ； ； ； in, The focus portion of smooth cluster regularization is represented by ; RRR represents the smooth portion of smooth focus regularization; e R is the focus factor, used to control the degree of focus in the inversion model. x R y and R z They are respectively x , y , z The difference operator applied between adjacent grids in the direction.

3. The adaptive octree three-dimensional magnetic inversion method for fusing anomaly region identification as described in claim 2, characterized in that: In step three, the objective function formula (1) is optimized using the Gauss-Newton method, resulting in the following iterative model expression: ； Where J is the Jacobian matrix. For the first k The change in model parameters in each iteration; To improve the efficiency of magnetic 3D inversion, an initial inversion is first performed on a coarse grid to quickly obtain the overall outline of the underground structure, preparing for subsequent intelligent identification of anomaly areas.

4. The adaptive octree three-dimensional magnetic inversion method for fusing anomaly region identification according to claim 1, characterized in that: In step four, the fuzzy c-means clustering algorithm can automatically calculate the membership degree of each grid's physical property value relative to the cluster center based on the cluster center; the adaptive extremum technique can adaptively determine the initial cluster center and the number of cluster centers; through the fusion of the fuzzy c-means clustering algorithm and the adaptive extremum technique, the intelligent extraction of abnormal regions can be accurately achieved. The objective function expression for the fuzzy c-means clustering algorithm is as follows: ； ； in, n For the mesh subdivision fraction, p This represents the number of clusters. u ij Represented as the first i Model parameters m i Compared to the first j Cluster centers v j membership degree q For fuzzy factors; The adaptive extremum technique first selects the values ​​of the region with the densest distribution of model property values ​​as background values ​​'a', and replaces the property values ​​close to 'a' with 'a' to enhance the data's discriminative power. Then, it extracts the extreme points in the model as the center points of outliers, and uses these to determine the value and number of cluster centers. After determining the value and number of cluster centers, the membership matrix is ​​obtained by optimizing the objective function (7): ; ; in, t ij For the first i Model parameters m i Compared to the first j Cluster centers v j The Euclidean distance; t ip For the first i Physical property values ​​of each grid m i Compared to the first p Cluster centers v p The Euclidean distance; Finally, based on the principle of maximum membership, the physical property values ​​can be distinguished into background regions and abnormal regions, thus realizing intelligent extraction of abnormal regions from low-precision inversion models.

5. The adaptive octree three-dimensional magnetic inversion method for fusing anomaly region identification according to claim 1, characterized in that: In step six, since one grid in the octree mesh may be adjacent to multiple grids and cannot be calculated by formula, we first find all the grids adjacent to each grid and then apply the difference operator to ensure the smoothness of the inversion result in any direction in space.

6. The adaptive octree three-dimensional magnetic inversion method for fusing anomaly region identification according to claim 1, characterized in that: In step seven, the data fitting difference is used to quantitatively evaluate the inversion results, and its expression is as follows: ; Where, d obs For observation data, d inv To predict the forward response from the inversion data, err To account for observation errors, if the inversion result still does not meet the expected data fit difference requirement, the process of "intelligent identification - local refinement - fine inversion" will continue to achieve dynamic refinement of the grid and allocation of computing resources during the inversion process, so as to gradually improve the inversion accuracy.

Images