An Adaptive Voxelization Attribute Assignment and Dual-Model Collaborative Mineral Exploration Method and System Based on Multi-Source Data

Abstract

This application discloses an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method and system based on multi-source data, relating to the fields of artificial intelligence and mineral exploration prediction. The method includes: dividing the target area into three-dimensional voxel grids using an adaptive grid function; assigning attribute values ​​of multi-source mineralization-prospecting information to the three-dimensional voxel grids of the target area and known mining areas, and assigning attribute values ​​of ore grade to the three-dimensional voxel grids of known mining areas; calculating contribution weights using a logistic regression model, and obtaining the first mineralization score of the target area's three-dimensional voxel grids through linear combination calculation and mapping; training a deep learning model using multi-source mineralization-prospecting information as input and ore grade as a label, and determining the second mineralization score of the target area's three-dimensional voxel grids; determining the comprehensive mineralization score of each three-dimensional voxel grid in the target area using a comprehensive scoring function, and delineating the mineral exploration target area based on the comprehensive mineralization score. This application improves the accuracy and interpretability of mineral exploration in the target area.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and mineral exploration prediction technology, and in particular to an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method and system based on multi-source data. Background Technology

[0002] Mineral resources are an important material foundation for economic development. As surface prospecting becomes increasingly sophisticated, prospecting is gradually shifting towards deeper and concealed deposits, placing higher demands on three-dimensional prospecting prediction technology.

[0003] Voxelization, as a core foundation for 3D prediction, provides a unified structure for multi-source data fusion through regular grids. However, related technologies still have significant shortcomings: on the one hand, voxel sizes lack dynamic geological constraints and often adopt globally uniform sizes, leading to the loss of details in orebody distribution areas and data redundancy in surrounding rock areas; on the other hand, discrete variable assignment exhibits a sawtooth effect, and continuous variable interpolation methods are simplistic and cannot be dynamically adjusted according to data distribution. Crucially, current mineralization favorability prediction models are relatively simplistic, with highly subjective weight settings. While logistic regression offers interpretability, its linear assumption limits its accuracy; and although deep learning can characterize nonlinear spatial relationships, its "black box" nature results in poor interpretability.

[0004] Because the two advantages are difficult to integrate quantitatively, the overall mineral exploration prediction results are often not very accurate and have poor interpretability. Summary of the Invention

[0005] The purpose of this application is to provide an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method and system based on multi-source data, which effectively improves the accuracy and interpretability of mineral exploration in the target area.

[0006] To achieve the above objectives, this application provides the following solution.

[0007] Firstly, this application provides an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data, comprising: acquiring multi-source mineralization-prospecting information data of a target area and known mining areas, as well as ore body grade data of known mining areas; dividing a three-dimensional voxel grid using an adaptive mesh function based on the multi-source mineralization-prospecting information data of the target area and known mining areas; the adaptive mesh function being constructed based on local information entropy, structural curvature, and interface distance field; assigning attribute values ​​of multi-source mineralization-prospecting information to the three-dimensional voxel grids of the target area and known mining areas using the kriging method, the inverse power distance method, and a category function; and assigning ore body grade attribute values ​​to the three-dimensional voxel grids of known mining areas using a grade function; constructing a three-dimensional voxel grid database of the target area based on the three-dimensional voxel grids of the target area after the multi-source mineralization-prospecting information attribute assignment; and constructing a three-dimensional voxel grid database of the target area based on the three-dimensional voxel grids of the target area after the multi-source mineralization-prospecting information attribute assignment and the three-dimensional voxel grids of known mining areas after the ore body grade attribute assignment. A voxel grid is constructed to create a three-dimensional voxel grid database of known mining areas. Based on this database, a logistic regression model is used to calculate the contribution weights of multi-source mineralization and prospecting information to ore body grade. Based on the target area's three-dimensional voxel grid database and the contribution weights, a first mineralization score for each three-dimensional voxel grid in the target area is obtained through linear combination calculation and mapping. Using the known mining area's three-dimensional voxel grid database as input and ore body grade as a label, a deep learning model is trained using a binary cross-entropy loss function. Based on the target area's three-dimensional voxel grid database and the trained deep learning model, a second mineralization score for each three-dimensional voxel grid in the target area is obtained. Based on the first and second mineralization scores, a comprehensive scoring function is used to determine the comprehensive mineralization score for each three-dimensional voxel grid in the target area. The comprehensive mineralization score is compared with an optimal threshold to delineate the prospecting target area.

[0008] Secondly, this application also provides a computer system, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data as described in the first aspect.

[0009] Based on the specific embodiments provided in this application, the following technical effects are disclosed.

[0010] First, this application acquires multi-source mineralization and prospecting information data from the target area and known mining areas, as well as ore body grade data from known mining areas, ensuring the richness of data dimensions and the reliability of comparative analysis. Second, this application constructs an adaptive grid function based on local information entropy, tectonic curvature, and interface distance field, and uses the adaptive grid function to achieve dynamic partitioning of the three-dimensional voxel grid, effectively overcoming the boundary jaggedness effect caused by a fixed grid and more realistically reflecting the spatial distribution details of geological bodies. Then, this application also employs a category function and a grade function to improve the accuracy of assigning values ​​to the multi-source information and ore body grade attributes of the three-dimensional voxel grid. Finally, this application uses a logistic regression model to obtain a first mineralization score, a deep learning model to obtain a second mineralization score, and fuses the two through a comprehensive score function to obtain a comprehensive mineralization score. After comparing this comprehensive mineralization score with the optimal threshold, an effective prospecting target area can be delineated, and subsequent precise prospecting can be carried out around the delineated prospecting target area. Based on the above design, this application effectively improves the accuracy and interpretability of prospecting in the target area. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of an embodiment of the adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data in this application.

[0013] Figure 2 This is a rendering of a three-dimensional voxel mesh in one embodiment of this application.

[0014] Figure 3 This is an internal structure diagram of a computer system according to another embodiment of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] In one exemplary embodiment, an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data is provided, such as... Figure 1 As shown, the method is as follows.

[0018] Step S1: Obtain multi-source mineralization and prospecting information data for the target area and known mining areas, as well as ore body grade data for known mining areas.

[0019] In this embodiment, the multi-source mineralization-prospecting information data includes: basic geological data (strata, lithology, faults), geophysical data (magnetic susceptibility, polarizability), geochemical data (soil geochemical content), and remote sensing alteration data (alteration mineral types), etc. The ore body grade data is the controlled three-dimensional spatial location and grade of the gold ore body.

[0020] Step S2: Based on the multi-source mineralization and prospecting information data of the target area and known mineral areas, a three-dimensional voxel mesh is generated using an adaptive mesh function.

[0021] In this embodiment, an adaptive mesh function is constructed based on local information entropy, construction curvature, and interface distance field. The adaptive mesh function is as follows.

[0022] .

[0023] In the formula, The size of the three-dimensional voxel mesh; These are the base values ​​for the 3D voxel mesh; It is an adaptive strength coefficient; , and These are the weighting coefficients, and + + =1; The local information entropy is calculated using the Gaussian weighted kernel density method; To construct the curvature, the gradient structure tensor method is used for calculation; The interface distance field is calculated using the fast travel method.

[0024] Step S3: Using the Kriging method, the inverse power distance method, and the category function, assign attribute values ​​of multi-source mineralization-prospecting information to the three-dimensional voxel grids of the target area and known mining areas. Using the grade function, assign attribute values ​​of ore body grade to the three-dimensional voxel grids of known mining areas.

[0025] In this embodiment, step S3 is as follows.

[0026] (1) For continuous variables (geophysical data, geochemical data, etc.) in multi-source mineralization-prospecting information, the attribute values ​​of the three-dimensional voxel grid of the target area and known mineral areas are calculated by using the Kriging method and the distance power inverse ratio method.

[0027] (2) For discrete variables (basic geological data, remote sensing alteration data, etc.) in multi-source mineralization-prospecting information, a category function is used to determine the attribute category of the three-dimensional voxel grid of the target area and known mining areas. The category function is constructed based on the geological category, Gaussian distance attenuation term, and volume ratio, and the specific category function is as follows.

[0028] .

[0029] In the formula, The attribute category of a three-dimensional voxel mesh (for a discrete variable); C Let be the set of geological categories for a discrete variable (all possibilities). m Indexing of the sub-3D voxel mesh; For the first The shortest distance from the center of a 3D voxel grid to the boundary of a geological body. This is the Gaussian distance decay term; This is the distance attenuation parameter; The Gaussian kernel function; This is the balance coefficient; For the first Within a 3D voxel mesh, the first The volume occupied by each geological category; For the first The volume of a 3D voxel mesh; This refers to the percentage of volume. For the first m The category to which a sub-3D voxel mesh belongs; For indicator functions; To select the category corresponding to the maximum value.

[0030] (3) For a three-dimensional voxel grid of a known mining area that is completely located inside the ore body, the grade data of the corresponding position of the ore body is directly inherited.

[0031] (4) For a known ore area that is not completely located inside the ore body (also known as the boundary), the grade data of each 3D voxel grid is calculated using a grade function. The grade function is constructed based on the ore body grade data and the volume of the ore body within the sub-3D voxel grid. The grade function is as follows.

[0032] .

[0033] In the formula, The final grade data for the (boundary) three-dimensional voxel mesh; For the first Original grade data of ore bodies within a three-dimensional voxel grid; For the first The volume of the ore body within a three-dimensional voxel grid.

[0034] See the final 3D voxel mesh effect after attribute assignment. Figure 2 .

[0035] Step S4: Based on the three-dimensional voxel grid of the target area after assigning attribute values ​​to the multi-source mineralization-prospecting information, construct a three-dimensional voxel grid database of the target area. Based on the three-dimensional voxel grid of the known mining area after assigning attribute values ​​to the multi-source mineralization-prospecting information and ore body grade, construct a three-dimensional voxel grid database of the known mining area.

[0036] In this embodiment, the three-dimensional voxel grid database of the target area and known mining areas stores multi-source mineralization and prospecting information attribute fields, including four major categories and seven fields: basic geological data (strata, lithology, faults), geophysical data (magnetic susceptibility, polarizability), geochemical data (Au element soil geochemical content), and remote sensing alteration data (alteration mineral types). The three-dimensional voxel grid database of known mining areas also stores one field, such as ore body grade data. In addition to storing the above attribute fields, dual quality control is required, namely geometric accuracy and attribute consistency verification.

[0037] Step S5: Based on the known three-dimensional voxel grid database of the mining area, use the logistic regression model to calculate the contribution weight of multi-source mineralization-prospecting information to the ore body grade. Based on the three-dimensional voxel grid database of the target area and the contribution weight, obtain the first mineralization score of each three-dimensional voxel grid in the target area through linear combination calculation and mapping.

[0038] In this embodiment, the data in the known three-dimensional voxel grid database of the mining area needs to be randomly divided into a training set and a validation set. The training set is used to train the elastic network logistic regression to obtain the final weight coefficient vector. Finally, the weights are applied to the target area, and by calculating the linear combination and mapping it to the probability, the first mineralization score of each three-dimensional voxel grid is obtained.

[0039] Step S6: Based on the known three-dimensional voxel grid database of the mining area, with multi-source mineralization and prospecting information as input and ore body grade as label, the deep learning model is trained using the binary cross-entropy loss function. Based on the target area three-dimensional voxel grid database and the trained deep learning model, the second mineralization score of each three-dimensional voxel grid in the target area is obtained.

[0040] In this embodiment, the deep learning model is a 4-layer 3D encoder-decoder network with a 12-channel feature tensor as input. The decoder is symmetrical to the encoder, and attention gating is added at the skip connections. The training uses the Adam optimizer, and a spatial smoothing term is added to the loss function. After training, the target area is forward propagated to obtain the second mineralization score of each 3D voxel grid.

[0041] Step S7: Based on the first mineralization score and the second mineralization score, use the comprehensive score function to determine the comprehensive mineralization score of each three-dimensional voxel grid in the target area, compare the comprehensive mineralization score with the optimal threshold, and delineate the mineral exploration target area.

[0042] In this embodiment, a comprehensive scoring function is constructed based on the first mineralization score, the second mineralization score, the global AUC value, and the local correction factor. The comprehensive scoring function is as follows.

[0043] .

[0044] .

[0045] In the formula, 3D voxel mesh x The overall mineralization score; 3D voxel mesh x First mineralization score; 3D voxel mesh x The second mineralization score; This represents the global AUC value of the logistic regression model on the validation set. This represents the global AUC value of the deep learning model on the validation set. This is a local correction factor for the logistic regression model; This is a local correction factor for the deep learning model; For the validation set; To verify the total number of 3D voxel meshes in the set; 3D voxel mesh Multi-source mineralization-prospecting information feature vectors, for any three-dimensional voxel grid in the target area ,That It includes seven fields: stratigraphy, lithology, faults, Au element soil geochemical content, magnetic susceptibility, polarizability, and alteration mineral type. To verify the first set A feature vector of multi-source mineralization and prospecting information in a three-dimensional voxel grid; For bandwidth parameters, ; It is a logistic regression model; For deep learning models; The Gaussian kernel function; denoted as Euclidean distance between two points in the feature space.

[0046] In addition, the optimal threshold needs to be determined by maximizing the Youden index on the validation set. This involves iterating through all thresholds from 0 to 1 on the validation set with a step size of 0.01, calculating the sensitivity and specificity for each threshold, and selecting the threshold where the Youden index reaches its maximum value as the optimal threshold. Finally, three-dimensional voxel grids with a comprehensive mineralization score greater than or equal to the optimal threshold are marked as potential ore bodies, while the rest are considered background, thus delineating multiple target areas for subsequent mineral exploration.

[0047] In addition, the following steps may be performed after step S7.

[0048] (1) Calculate the comprehensive evaluation value of each target area using the target area evaluation function. The target area evaluation function is constructed based on the comprehensive mineralization score, target area volume, target area morphological complexity, and distance from the target area center to the nearest known ore body. The specific target area evaluation function is as follows.

[0049] .

[0050] In the formula, For the first k The comprehensive evaluation value of each target area; For the first k The average of the comprehensive mineralization scores of all three-dimensional voxel grids within a target area; For the first k The volume of each target region; For reference volume; For the first k The distance from the center of each target area to the nearest known ore body; Considering the maximum distance; For the first k The morphological complexity of each target region; This represents the maximum morphological complexity.

[0051] (2) Based on the comprehensive evaluation value, all target areas are sorted from largest to smallest.

[0052] (3) Conduct (precise) mineral exploration for each target area according to the sorting results.

[0053] In another exemplary embodiment, a practical application scenario of an adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data is provided. Taking a gold mine in a certain region as an example, the study area has well-developed faults and widespread metamorphic rocks, and multiple gold deposits (points) have been discovered. This region includes the target area and known mining areas.

[0054] (1) Obtain multi-source mineralization-prospecting information data of the target area and known mining areas, and ore body grade data of known mining areas. Based on the distribution area and periphery of known deposits (points), the modeling space is determined as follows: X direction 39340000~39352000m, Y direction 4126000~4131000m, Z direction 650~1700m. The multi-source mineralization-prospecting information data collected this time includes: basic geological data (1:10,000 geological map, 30 borehole columnar sections, 10 exploration line profiles), geophysical data (1:10,000 high-precision magnetic survey, induced polarization gradient survey data), geochemical data (1:10,000 soil geochemical survey data), remote sensing alteration data (1:50,000 hyperspectral remote sensing hydroxyl, iron alteration), and ore body grade data are the three-dimensional spatial location and grade of the controlled gold ore body.

[0055] (2) Based on the multi-source mineralization and prospecting information data of the target area and known mining areas, the size of the three-dimensional voxel grid is divided by an adaptive grid function, and the three-dimensional voxel grid is constructed with the three-dimensional coordinates as the primary key and each three-dimensional voxel grid is encoded in sequence.

[0056] Adaptive mesh function is used: .

[0057] Local information entropy H Using the Gaussian weighted kernel density method, the average stratum thickness in the study area is approximately 40m. The sliding window size was set to 200m × 200m × 40m, and the Gaussian kernel bandwidth was 50m. (This is for areas with complex lithology.) ; homogeneous gneiss area Construction curvature Using the gradient structure tensor method, the fault surface inferred from geological maps and magnetic inversion is calculated near the fault zone. Far from the fault zone Interface distance field The fast-travel method is used to infer fault structures based on geological maps and magnetic inversion, and then the Laplace operator is calculated for the area near the fault surface. ,far away Take weights Basic 3D voxel mesh size Adaptive strength coefficient Then: the size of the three-dimensional voxel mesh in the uniform region m; Size of the three-dimensional voxel mesh in the fracture region m; Size of the three-dimensional voxel mesh in the mineralized zone m. A three-dimensional voxel mesh is generated according to the above-mentioned size division principles. Approximately 900,000 three-dimensional voxel meshes are found in the known mining area, and approximately 700,000 in the target area. Each voxel mesh is assigned a code sequentially. .

[0058] (3) Assign attribute values ​​to the three-dimensional voxel grids of the target area and known mining areas based on multi-source mineralization-prospecting information. For geophysical and geochemical data in the multi-source mineralization-prospecting information data, the attribute values ​​of the three-dimensional voxel grids are calculated using the kriging and distance power inverse ratio methods. For basic geological data and remote sensing alteration data in the multi-source mineralization-prospecting information data, the attribute categories of the three-dimensional voxel grids are determined using category functions.

[0059] For continuous variables such as geophysical and geochemical data, the three-dimensional voxels of the known mining area are used. , Taking Au element soil geochemical data as an example, with the center point of the voxel as the sphere center, the search radius is set to 200m, and the number of samples involved in the calculation must be ≥8. If there are fewer than 8 samples, the search radius is expanded in increments of 50m until the requirement is met. First, the local coefficient of variation is calculated using the standard coefficient of variation formula in statistics. After calculation, the three-dimensional voxels of the known mining area are... , The coefficients of variation for Au element soil geochemical data were 1.5 and 0.3, respectively. Next, the coefficients of variation were mapped to adaptive mixed weights using the Sigmoid function. During the calculation, the threshold parameter was set to 0.8 and the steepness parameter to 0.5. The calculated values ​​were obtained for the known three-dimensional voxels of the mining area. , The range mixing weights were 0.97 and 0.08, respectively. Then, the three-dimensional voxels of the known mining area were calculated using the inverse power distance method and the kriging method, respectively. , The soil geochemical estimate of Au element. For known three-dimensional voxels of the mining area. The estimated value was 30.5 ng / g calculated using the inverse power law of distance, and 23.2 ng / g calculated using the kriging method. For the known three-dimensional voxels of the mining area... The inverse power law method yielded 1.9 ng / g, while the kriging method yielded 2.4 ng / g. Finally, a weighted arithmetic mean was used to calculate the final adaptive hybrid interpolation result. As mentioned earlier, the three-dimensional voxels of the known mining area... , The range mixing weights are 0.97 and 0.08, respectively. For a known three-dimensional voxel area... The final soil geochemical result for Au is 0.97 × 30.5 + 0.03 × 23.2 = 30.3 ng / g. For the known three-dimensional voxels of the mining area... The final soil geochemical result for Au element is 0.08×1.9+0.92×2.4=2.36ng / g.

[0060] For discrete variables such as basic geological data and remote sensing alteration data, a category function is used: First, based on the distance from the fault, the samples are divided into five categories: 0m (within the fault zone), 0~20m, 20~50m, 50~100m, and >100m, respectively denoted as Category I, Category II, Category III, Category IV, and Category V. Second, the voting weights of the three-dimensional voxel grid are calculated, using the known three-dimensional voxels of the mining area. For example, this three-dimensional voxel is a boundary voxel, which is divided into 8 sub-three-dimensional voxels, denoted as... , , , , , , , .set up m, The weights of each sub-3D voxel are calculated sequentially. For example, the weights of each sub-3D voxel... The center distance from the fracture is 2m. The sub-3D voxel belongs to the 0~20m zone (Category II). The fracture volume within the sub-3D voxel accounts for 0.3%. Therefore, the weight... Finally, based on the category function, the 0~20m band (Category II) received the largest total weighted vote, and the voxel was ultimately assigned to the 0~20m band (Category II).

[0061] The grade of the ore body is assigned to the 3D voxel grid of a known mining area. For 3D voxel grids located entirely within the ore body, the grade data of the corresponding location of the ore body is directly inherited; for boundary 3D voxel grids, the grade data of each 3D voxel grid is calculated using a grade function.

[0062] Within the controlled three-dimensional space of the gold ore body, there are 4250 boundary three-dimensional voxel grids and 15320 internal three-dimensional voxel grids. For the internal three-dimensional voxel grids, the average grade of the corresponding location within the ore body is directly inherited. For the boundary three-dimensional voxel grids, a boundary three-dimensional voxel grid is first subdivided into 8 sub-three-dimensional voxel grids. The boundary voxel grade is then calculated by weighted averaging the volume and average grade of the ore body within each sub-three-dimensional voxel grid. For example, in a certain boundary voxel, 3 out of the 8 sub-voxels contain ore, with volumes of 2m³ and 2m³ respectively. 3 5m 3 3m 3 The corresponding average grades are 4.2 g / t, 3.8 g / t, and 4.5 g / t. Therefore, the weighted grade = (4.2 × 2 + 3.8 × 5 + 4.5 × 3) / (2 + 5 + 3) = 4.09 g / t.

[0063] (4) Construct a three-dimensional voxel grid database for the target area and known mining areas, storing attribute fields of multi-source mineralization and prospecting information, including four major categories and seven fields: basic geological data (strata, lithology, faults), geophysical data (magnetic susceptibility, polarizability), geochemical data (Au element soil geochemical content), and remote sensing alteration data (alteration mineral type). For known mining areas, it also includes one field such as ore body grade data. Ten profiles were randomly selected to calculate the geometric accuracy Jaccard coefficient, with an average value of 0.96, which is better than 0.95. The volume changes of major geological bodies before and after voxelization were statistically analyzed, with an error of 1.2% for metamorphic rocks, 1.8% for fault zones, an average attribute deviation of 3.2%, and an attribute consistency error (deviation) of less than 5%, thus passing quality control.

[0064] (5) In a known mining area, the contribution weight of multi-source mineralization-prospecting information to the ore body grade is calculated based on the logistic regression model, and the corresponding weight is applied to the target area to obtain the first mineralization score of each three-dimensional voxel grid in the target area.

[0065] First, the voxel data of the known mining areas are randomly divided into a training set (80%) and a validation set. (20%). In the training set, 2000 three-dimensional voxel grids of positive samples (ore bodies) and 8000 three-dimensional voxel grids of negative samples (surrounding rocks) were selected. Each three-dimensional voxel grid already contains feature vectors of multi-source mineralization-prospecting information and ore body information. f The sample data was normalized and subjected to elastic net logistic regression. To achieve feature selection and prevent overfitting, elastic net regularization was introduced. The search range for regularization strength was set to 0.01~10, and the search range for L1 proportionality coefficient was set to 0~1. 10,000 samples were randomly divided into 5 equal parts, and the optimal regularization strength of 0.1 and L1 proportionality coefficient of 0.6 were selected through 5-fold cross-validation. Then, using all 10,000 samples and the optimal regularization strength (0.1) and L1 proportionality coefficient (0.6), the elastic net logistic regression was retrained to obtain the final weight coefficient vector. The results showed that features such as Au element content (weight value 2.34), distance from the fracture (weight value 1.87), and polarizability (weight value 1.56) made significant contributions. Finally, the weights were applied to the target area, and the first mineralization score of each three-dimensional voxel grid in the target area was obtained through steps such as calculating linear combination (Logit) and mapping to probability (Sigmoid function). The first mineralization score for each three-dimensional voxel grid was calculated. The value range is [0.02, 0.93].

[0066] (6) Using the known ore body grade of the mining area as a label, input the multi-source (multi-channel) mineralization-prospecting information feature tensor, and use the binary cross-entropy loss function for end-to-end training to construct a deep learning model with a three-dimensional encoder-decoder structure. Input the multi-source mineralization-prospecting information feature tensor of the target area into the trained deep learning model to obtain the second mineralization score of each three-dimensional voxel grid in the target area.

[0067] A four-layer 3D encoder-decoder network was constructed, with a 12-channel feature tensor (64×80×32) as input. The number of convolutional kernels per layer in the encoder was [32, 64, 128, 256], and the decoder was symmetrical to the encoder. Attention gating was added at skip connections. Training was performed using the Adam optimizer with an initial learning rate of 0.001, a batch size of 4, and 200 training epochs. A spatial smoothing term (regularization coefficient of 0.01) was added to the loss function. After training, forward propagation was performed on the target region to obtain the second mineralization score for each 3D voxel grid in the target region. The value range is [0.01, 0.95].

[0068] (7) The comprehensive mineralization score of the three-dimensional voxel grid is obtained by using the comprehensive scoring function. The optimal threshold is determined by maximizing the Youden index on the validation set, the mineral exploration target area is delineated, and the target area evaluation function is used to carry out the comprehensive evaluation of the target area.

[0069] First, according to (5), the validation set contains approximately 180,000 three-dimensional voxel meshes. The global AUC values ​​of the two models are calculated on the validation set using numerical integration (trapezoidal method). The calculated values ​​for the logistic regression model are... Deep learning models .

[0070] Secondly, calculate the local correction factor as follows.

[0071] .

[0072] Specifically, the target area is a three-dimensional voxel mesh. For example, the feature similarity between all samples in the validation set and the local correction factor is calculated. , .

[0073] Then, the comprehensive mineralization score of the three-dimensional voxel mesh is calculated as follows.

[0074] .

[0075] Calculation First mineralization score Second mineralization score Overall mineralization score The above calculations were performed on all 700,000 three-dimensional voxel grids in the target area, and the comprehensive mineralization score for each three-dimensional voxel grid was obtained.

[0076] Next, adaptive threshold determination is performed. The optimal threshold is determined by maximizing the Youden exponent on the validation set. All thresholds from 0 to 1 are traversed on the validation set with a step size of 0.01, and the sensitivity and specificity at each threshold are calculated, as shown in the table below.

[0077] Table 1. Key Threshold Related Data

[0078] When the threshold is 0.63, the Youden index reaches its maximum value of 0.72 (sensitivity 0.85, specificity 0.87). Therefore, 0.63 is chosen as the (globally) optimal threshold.

[0079] Finally, target area delineation and comprehensive evaluation were conducted. Voxels with a comprehensive mineralization score ≥0.63 were marked as potential ore bodies, and the rest were considered background, resulting in the delineation of five target areas, numbered K1, K2, K3, K4, and K5. A target area evaluation function was used to conduct a comprehensive evaluation of the target areas.

[0080] .

[0081] Taking target area K1 as an example, this target area is located east of the known ore body and contains 1250 three-dimensional voxel grids. The average comprehensive mineralization score is... target volume m 3 Reference volume value m 3 Distance from the center coordinates of the target area to the nearest known ore body m, maximum considered distance m, target surface area m 2 morphological complexity Maximum morphological complexity (Based on validation set statistics, 95th percentile) Overall evaluation value .

[0082] Similar to calculating all 5 target regions The values ​​are sorted from largest to smallest as follows.

[0083] Table 2 Target Area Ranking Data

[0084] The ranking results show that they are highly consistent with the experience of geological experts, prioritizing target areas with good mineralization conditions, regular shapes, and a certain scale.

[0085] In another exemplary embodiment, a computer system is provided, which may be a server or a terminal, and its internal structure diagram may be as follows: Figure 3 As shown, the computer system includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores multi-source mineralization-prospecting information data and ore body grade data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an adaptive voxelization attribute assignment and dual-model collaborative prospecting method based on multi-source data.

[0086] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer system to which the present application is applied. A specific computer system may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0087] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0088] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0089] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0090] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0091] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for adaptive voxelization attribute assignment and dual-model collaborative mineral exploration based on multi-source data, characterized in that, include: Acquire multi-source mineralization and prospecting information data for the target area and known mining areas, as well as ore body grade data for known mining areas; Based on multi-source mineralization and prospecting information data from the target area and known mining areas, a three-dimensional voxel mesh is generated using an adaptive meshing function. The adaptive meshing function is constructed based on local information entropy, structural curvature, and interface distance field. The adaptive meshing function is as follows: ； In the formula, The size of the three-dimensional voxel mesh; These are the base values ​​for the 3D voxel mesh; It is an adaptive strength coefficient; , and These are the weighting coefficients; Local information entropy; To construct curvature; For the interface distance field; The Kriging method, the inverse power distance method, and the category function are used to assign attribute values ​​of multi-source mineralization-prospecting information to the three-dimensional voxel grids of the target area and known mining areas. The grade function is used to assign attribute values ​​of ore body grade to the three-dimensional voxel grids of known mining areas. Based on the three-dimensional voxel grid of the target area after assigning attribute values ​​to the multi-source mineralization and prospecting information, a three-dimensional voxel grid database of the target area is constructed. Based on the three-dimensional voxel grid of the known mining area after assigning attribute values ​​to the multi-source mineralization and prospecting information and ore body grade, a three-dimensional voxel grid database of the known mining area is constructed. Based on the known three-dimensional voxel grid database of the mining area, the contribution weight of multi-source mineralization-prospecting information to the ore body grade is calculated using a logistic regression model. Based on the three-dimensional voxel grid database of the target area and the contribution weight, the first mineralization score of each three-dimensional voxel grid in the target area is obtained through linear combination calculation and mapping. Based on the known three-dimensional voxel grid database of the mining area, with multi-source mineralization-prospecting information as input and ore body grade as label, the deep learning model is trained using the binary cross-entropy loss function. Based on the target area three-dimensional voxel grid database and the trained deep learning model, the second mineralization score of each three-dimensional voxel grid in the target area is obtained. Based on the first mineralization score and the second mineralization score, a comprehensive mineralization score for each three-dimensional voxel grid in the target area is determined using a comprehensive scoring function. The comprehensive mineralization score is then compared with an optimal threshold to delineate the mineral exploration target area. The comprehensive scoring function is: ； ； In the formula, 3D voxel mesh x The overall mineralization score; 3D voxel mesh x First mineralization score; 3D voxel mesh x The second mineralization score; This represents the global AUC value of the logistic regression model on the validation set. This represents the global AUC value of the deep learning model on the validation set. This is a local correction factor for the logistic regression model; This is a local correction factor for the deep learning model; For the validation set; To verify the total number of 3D voxel meshes in the set; 3D voxel mesh Multi-source mineralization-prospecting information feature vector; To verify the first set A feature vector of multi-source mineralization and prospecting information in a three-dimensional voxel grid; For bandwidth parameters; It is a logistic regression model; For deep learning models; The distance is Euclidean. This is the Gaussian kernel function.

2. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 1, characterized in that, The category function is: ； For the attribute categories of the 3D voxel mesh; C A set of geological categories for discrete variables; For the first The shortest distance from the center of a 3D voxel grid to the boundary of a geological body; This is the distance attenuation parameter; The Gaussian kernel function; This is the balance coefficient; For the first Within a 3D voxel mesh, the first The volume occupied by each geological category; For the first The volume of a three-dimensional voxel mesh; For the first m The category to which a sub-3D voxel mesh belongs; For indicator functions; To select the category corresponding to the maximum value.

3. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 1, characterized in that, The grade function is: ； In the formula, The final grade data for the three-dimensional voxel mesh; For the first Original grade data of ore bodies within a three-dimensional voxel grid; For the first The volume of the ore body within a three-dimensional voxel grid.

4. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 1, characterized in that, The multi-source mineralization and prospecting information data includes: basic geological data, geophysical data, geochemical data, and remote sensing alteration data.

5. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 4, characterized in that, Kriging, inverse distance power ratio, and category functions are used to assign attribute values ​​of multi-source mineralization-prospecting information to the 3D voxel grids of the target area and known mining areas. A grade function is used to assign attribute values ​​of ore body grade to the 3D voxel grids of known mining areas. Specifically, this includes: For the geophysical and geochemical data in the multi-source mineralization-prospecting information, the attribute values ​​of the three-dimensional voxel grid of the target area and known mineral areas are calculated using the Kriging method and the inverse power distance method. For the basic geological data and remote sensing alteration data in the multi-source mineralization-prospecting information, the attribute categories of the three-dimensional voxel grids of the target area and known mining areas are determined by category functions; For a three-dimensional voxel grid of a known mining area that is entirely located inside the ore body, the grade data of the corresponding location of the ore body is directly inherited; For a three-dimensional voxel grid of a known ore area that is not entirely located within the ore body, the grade data of each three-dimensional voxel grid is calculated using a grade function.

6. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 1, characterized in that, After delineating the target area for mineral exploration, the following steps are also included: The comprehensive evaluation value of each target area is calculated using the target area evaluation function, which is constructed based on the comprehensive mineralization score, target area volume, target area morphological complexity, and the distance from the target area center to the nearest known ore body. Based on the comprehensive evaluation value, all target areas are sorted from largest to smallest; Mineral exploration is conducted in each target area according to the ranking results.

7. The adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data according to claim 6, characterized in that, The target evaluation function is: ； In the formula, For the first k The comprehensive evaluation value of each target area; For the first k The average of the comprehensive mineralization scores of all three-dimensional voxel grids within a target area; For the first k The volume of each target region; For reference volume; For the first k The distance from the center of each target area to the nearest known ore body; Considering the maximum distance; For the first k The morphological complexity of each target region; This represents the maximum morphological complexity.

8. A computer system, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the adaptive voxelization attribute assignment and dual-model collaborative mineral exploration method based on multi-source data as described in any one of claims 1-7.

Images