Intelligent delineation method for gold-silver ore prospecting target area based on GIS multi-element information composite

Abstract

The application discloses a GIS multi-element information composite gold-silver ore prospecting target area intelligent delineation method, including the following steps: data preprocessing: collecting geological-geophysical-remote sensing multi-source geology information data of the working area, carrying out multi-scale gridding processing on the multi-source geology information data, simultaneously, extracting the occurrence information of the structural alteration zone for the medium-low temperature hydrothermal deposit and introducing the anisotropic variation function model for directional gridding, so that the data is correlated and interpolated along the bedding direction; feature extraction: extracting geometric features, texture features, lithological combination entropy and continuous fractal distance field; model construction: taking the extracted geometric features, texture features, lithological combination entropy, continuous fractal distance field and multiple fractal singular values of the geophysical data as an input feature set, inputting into a neural network for training; target area delineation: generating an ore-forming probability graph by using the trained neural network, and carrying out post-validation on the prediction result, and delineating the ore prospecting target area with different ore prospecting potential.

Description

Technical Field

[0001] This invention relates to the field of gold and silver mining technology, and in particular to a method for intelligent delineation of gold and silver prospecting target areas based on GIS multi-source information. Background Technology

[0002] Gold and silver deposits hold a core position in national strategic resources for energy security, economic development, and high-tech industries. However, traditional prospecting methods such as geological mapping and geophysical anomaly interpretation suffer from problems such as strong subjectivity, difficulty in integrating multi-source data, low target area delineation efficiency, and easy omission of concealed ore bodies. Therefore, this method aims to leverage the powerful spatial data integration and management capabilities of GIS, integrate heterogeneous information from multiple sources including geology, geophysics, geochemistry, remote sensing, and topography, and combine intelligent algorithms such as machine learning and data mining to construct a multi-dimensional information association model. This enables in-depth mining and comprehensive interpretation of prospecting information, thereby accurately identifying favorable gold and silver mineralization zones. The intelligent delineation of mineral exploration target areas is significant not only for improving the accuracy and efficiency of target area delineation, reducing exploration costs and cycles, and increasing the probability of gold and silver resource discovery and development efficiency, but also for promoting the deep integration and innovative application of GIS technology and artificial intelligence in the field of geological prospecting. This will facilitate technological innovation in prospecting theories and methods, enhance the country's strategic resource security capabilities, and reduce the environmental disturbance caused by traditional exploration methods, thereby achieving a green, intelligent, and efficient prospecting model. Ultimately, this will provide key technological support for the high-quality development of mineral resource exploration and development, and achieve a synergistic improvement in economic, resource, and environmental benefits.

[0003] Traditional methods employ uniform grid sizes for data meshing, failing to simultaneously address the need for fine grids to capture weak signals in porphyry gold-silver deposits with their disseminated veinlet distribution, and the requirement for macroscopic grid coverage in medium- and low-temperature hydrothermal deposits with their bedding-like distribution. This results in the loss of weak veinlet disseminated signals in areas of intense geochemical anomalies due to overly coarse grids, while causing computational redundancy and resource waste in areas with a uniform background field due to overly fine grids. Therefore, a method for intelligent delineation of gold-silver prospecting target areas based on GIS multi-source information is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for intelligent delineation of gold and silver prospecting target areas based on GIS multi-source information.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent delineation of gold and silver prospecting target areas based on GIS multi-source information includes the following steps: Data preprocessing: Collect geological, geophysical and geochemical, and remote sensing multi-source geoscience information data of the work area, and use a quadtree adaptive gridding strategy to perform multi-scale gridding on the multi-source geoscience information data. At the same time, extract various mineralization information for medium-low temperature hydrothermal deposits and introduce an anisotropic variability function model for directional gridding, so that the data are correlated and interpolated along the mineralized structural construction direction. The quadtree algorithm is applied to the geophysical and geochemical data points. The grid is dynamically divided according to the local variance threshold, and two sets of gridded data, grid A and grid B, are output. Grid A (detail-preserving version) uses nearest neighbor interpolation to preserve the local high-value anomalies of the original geochemical data. Grid B (smooth version) uses Kriging interpolation (including anisotropic variability function) to generate a smooth geochemical background field. Feature extraction: Based on the preprocessed grid data, the curvature and tortuosity of linear structures are automatically extracted as geometric features using high-precision remote sensing images or geophysical data. The energy, entropy, and contrast of the remote sensing images are extracted as texture features through gray-level co-occurrence matrix analysis. The lithological diversity index is calculated within the grid cell as the lithological assemblage entropy (the number of lithological types and the distribution area ratio of each type of lithology are counted in each grid cell, and the lithological diversity index is calculated according to the information entropy formula. The grid with the higher entropy value represents the lithological abrupt interface or tectonic mixed rock zone, indicating the favorable location for gold and silver mineralization). At the same time, the optimal buffer radius is determined by fractal box counting method based on the fracture data and a continuous fractal distance field is generated. Specifically: using the grid cells of grid B as a unified spatial index, for features requiring detail (texture features of subtle texture changes in local altered minerals and multifractal singular values), in each cell of grid B, high-resolution data of grid A at the corresponding location is called for calculation, and then the calculation results are assigned to the cells of grid B. Model Construction: The extracted geometric features, texture features, lithological combination entropy, continuous fractal distance field, and multifractal singular values ​​of geochemical data are used as the input feature set and fed into the neural network for training. A fractal regularization term is introduced into the loss function, which is the difference between the fractal dimension of the mineralization probability map predicted by the model and the standard fractal dimension of the known mineralization area as a penalty term to constrain the model output to conform to the geological fractal law. The construction of the fractal regularization term is as follows: During the neural network training process, CA fractal analysis is performed on each batch of output mineralization probability maps to calculate its fractal dimension. The absolute difference between the fractal dimension and the pre-statistically calculated standard fractal dimension of the known mineralization area is multiplied by the regularization coefficient and added to the total loss function, so that the model is constrained by the geological fractal law while fitting the training data. Target area delineation: The trained neural network is used to generate a mineralization probability map, and the prediction results are verified posteriorly by combining CA fractal theory to delineate mineral exploration target areas with different exploration potentials.

[0006] The above further includes: Furthermore, the quadtree adaptive mesh partitioning specifically includes: A threshold is set based on the local variance or coefficient of variation of the geochemical data. In areas exceeding the threshold, the data is recursively subdivided into small grids of 10m×10m to capture the weak, disseminated signals of veinlets in porphyry gold and silver deposits. In background field areas below the threshold, a large grid of 100m×100m is maintained. Simultaneously, during the gridding process, two sets of grid data, grid A and grid B, are generated for the geochemical data. Grid A uses nearest neighbor interpolation to retain local high-value anomalies in the original geochemical data, while grid B uses kriging interpolation to generate a smooth geochemical background field.

[0007] Furthermore, the method for constructing the anisotropic variation function model is as follows: based on the occurrence information of the tectonic alteration zone, the strike, dip and dip angle parameters are determined, and in the Kriging interpolation, a long-range search radius is set along the construction direction of the ore-bearing structure, and a short-range search radius is set perpendicular to the stratigraphic direction.

[0008] Furthermore, the extraction of curvature and tortuosity of linear structures as geometric features specifically involves: using remote sensing images or geophysical data to identify curvature abrupt change points and tortuosity distribution areas of linear structures, and calculating curvature values ​​and tortuosity density to quantify structural deformation intensity.

[0009] Furthermore, gray-level co-occurrence matrix analysis is performed on multispectral or hyperspectral remote sensing images to calculate the energy, entropy, and contrast parameters within each grid cell. High entropy and high contrast regions correspond to the rough or disordered texture features of mylonite or tectonic breccia, serving as remote sensing proxy information for tectonic alteration zones.

[0010] Furthermore, taking the fracture structure as the center, the fractal box counting method is used to analyze the logarithmic relationship of the number of mineral points within different buffer radii, and the radius interval with the highest linear correlation is determined as the optimal buffer range. The distance from each grid cell to the fracture structure is calculated, and the fractal density function is applied for power-law decay weighting to obtain a continuously changing fractal distance field.

[0011] Furthermore, based on the original geochemical sampling point data, fractal filtering is used to extract the enrichment or depletion intensity values ​​of elements relative to the background field. Then, singular values ​​are gridded using nearest neighbor interpolation or simple interpolation methods to ensure that local anomaly features are not smoothed, serving as input features indicating deep mineralization information.

[0012] Furthermore, before inputting the extracted features into the neural network, principal component analysis is used to screen geometric features, texture features, lithological combination entropy, continuous fractal distance field and multifractal singular values, eliminating redundant features with low correlation to known mineral deposits and retaining the optimal feature subset for model training.

[0013] The present invention has the following beneficial effects: In this invention, a quadtree adaptive grid partitioning strategy is adopted to automatically subdivide areas with drastic geochemical anomalies into smaller grids to capture weak anomalies, while maintaining large grids in areas with uniform background fields to reduce computational redundancy. At the same time, an anisotropic variability function model is introduced for hydrothermal deposits to perform directional gridding, so that the data is correlated and interpolated along the direction of the ore-bearing structure, which truly reflects the distribution law of strata-controlled deposits. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of the intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information proposed in this invention. Detailed Implementation

[0015] 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.

[0016] Please see Figure 1 As shown, this invention is a method for intelligent delineation of gold and silver prospecting target areas based on GIS multi-source information, comprising the following steps: Data preprocessing: Collect geological, geophysical and geochemical, and remote sensing multi-source geoscience information data of the work area, and use a quadtree adaptive gridding strategy to perform multi-scale gridding on the multi-source geoscience information data. At the same time, extract various mineralization information for medium-low temperature hydrothermal deposits and introduce an anisotropic variability function model for directional gridding, so that the data are correlated and interpolated along the mineralized structural construction direction. The quadtree algorithm is applied to the geochemical point data, and the grid is dynamically divided according to the local variance threshold. Two sets of gridded data are output, grid A and grid B. Grid A (detail-preserving version) uses nearest neighbor interpolation to preserve the local high value anomalies of the original geochemical data. Grid B (smooth version) uses Kriging interpolation (including anisotropic variability function) to generate a smooth geochemical background field. Feature extraction: Based on the preprocessed grid data, the curvature and tortuosity of linear structures are automatically extracted as geometric features using high-precision remote sensing images or geophysical data. The energy, entropy, and contrast of the remote sensing images are extracted as texture features through gray-level co-occurrence matrix analysis. The lithological diversity index is calculated within the grid cell as the lithological assemblage entropy (the number of lithological types and the distribution area ratio of each type of lithology are counted in each grid cell, and the lithological diversity index is calculated according to the information entropy formula. The grid with the higher entropy value represents the lithological abrupt interface or tectonic mixed rock zone, indicating the favorable location for gold and silver mineralization). At the same time, the optimal buffer radius is determined by fractal box counting method based on the fracture data and a continuous fractal distance field is generated. Specifically: using the grid cells of grid B as a unified spatial index, for features requiring detail (texture features of subtle texture changes in local altered minerals and multifractal singular values), in each cell of grid B, high-resolution data of grid A at the corresponding location is called for calculation, and then the calculation results are assigned to the cells of grid B. Model Construction: The extracted geometric features, texture features, lithological combination entropy, continuous fractal distance field, and multifractal singular values ​​of geochemical data are used as the input feature set and fed into the neural network for training. A fractal regularization term is introduced into the loss function, which is the difference between the fractal dimension of the mineralization probability map predicted by the model and the standard fractal dimension of the known mineralization area as a penalty term to constrain the model output to conform to the geological fractal law. The construction of the fractal regularization term is as follows: During the neural network training process, CA fractal analysis is performed on each batch of output mineralization probability maps to calculate its fractal dimension. The absolute difference between the fractal dimension and the pre-statistically calculated standard fractal dimension of the known mineralization area is multiplied by the regularization coefficient and added to the total loss function, so that the model is constrained by the geological fractal law while fitting the training data. Target area delineation: The trained neural network is used to generate a mineralization probability map, and the prediction results are verified posteriorly by combining CA fractal theory to delineate mineral exploration target areas with different exploration potentials.

[0017] The specific workflow of the intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information is as follows: Raw data layer, input: Geological data: geological map of the work area, lithological distribution, occurrence of tectonic alteration zones, and coordinates of known ore deposits; Geochemical data: Element content values ​​of primary / secondary halos (point data, including coordinates and element concentrations); Remote sensing data: Multispectral / hyperspectral imagery (raster data, including band reflectance); Geophysical data: Magnetic and electrical geophysical data (raster or lattice data); Construction data: linear features of fracture and shear zones (vector line data); Data cleaning and coordinate unification: Project all data to the same coordinate system, remove outliers, unify data format, and output a standardized dataset after spatial registration; Adaptive gridding: A quadtree algorithm is applied to the geochemical point data to dynamically divide the grid based on the local variance threshold. High-variability areas (such as porphyry mineralization centers) are recursively subdivided into 10m × 10m grids, while low-variability areas (background field) maintain a 100m × 100m grid. Two sets of gridded data are output: Grid A: Nearest neighbor interpolation is used to preserve local high-value anomalies in the original geochemical data for subsequent multifractal singular value calculations; Grid B: Kriging interpolation (including anisotropic variation function) is used to generate a smooth geochemical background field for extracting conventional features such as combined anomaly maps; Directional gridding: If sedimentary / stratabound deposits exist in the working area, extract the occurrence (strike, dip, dip angle) of the structural alteration zone, set anisotropic search ellipsoids in the Kriging interpolation, with long range along the strike direction (e.g., 500m) and short range perpendicular to the strike direction (e.g., 50m), and output anisotropic gridded data. The grid cells are stretched along the bedding direction to reflect the distribution law of stratabound deposits. Geometric feature extraction: Input high-resolution remote sensing images and geophysical linear structure interpretation results, automatically identify linear structures, calculate the structural curvature and tortuosity density of each grid cell, and output a set of geometric feature layers (curvature map, tortuosity density map). Texture feature extraction: Input multispectral / hyperspectral remote sensing image, perform gray-level co-occurrence matrix analysis on each grid cell, calculate energy, entropy, and contrast parameters, and output texture feature layer set (energy map, entropy map, contrast map). High entropy / high contrast areas indicate mylonite or tectonic breccia. Lithological combination entropy calculation: Input lithological polygon data from the geological map, count the lithological types and area ratios in each grid cell, calculate the lithological diversity index based on the information entropy formula, and output the lithological combination entropy layer. Grids with high entropy values ​​represent lithological abrupt interfaces or tectonic mélange zones. Fractal buffer modeling: Input fracture structure vector line data and known mineral point distribution. Use box counting method, set an increasing radius r with the fracture as the center, count the number of mineral points N(r) within each radius, and fit the model. The curve is used to determine the radius interval with the highest linear correlation as the optimal buffer radius. The distance d from each grid cell to the break is calculated, and the fractal density function is applied. ( Power-law decay weighting is applied to the fractal dimension to output a continuous fractal distance field layer (the raster value decays with the power law of distance, replacing the traditional 0 / 1 binary buffer). Multifractal singular value calculation: Input grid A, use fractal filtering methods (such as SA method) to separate anomalies from background in geochemical data, calculate the element singular values ​​of each grid cell (reflecting enrichment / depletion intensity), and output multifractal singular value layer set (such as singular value map of elements such as Au, Ag, Cu). Feature selection and dimensionality reduction: Input all the above feature layers (geometry, texture, entropy, fractal buffer, singular values, etc.), use principal component analysis to select the feature subset with the highest correlation to known mineral deposits, remove redundant features, and output the optimized feature set (reduced dimensionality and maximized information content). Dataset partitioning: The feature set is divided into a training set (including known ore deposit grids and randomly selected non-ore deposit grids) and a validation set, and the training sample set (including feature vectors and labels) is output. The neural network is constructed using fractal regularization. The input training sample set undergoes forward propagation. The neural network receives feature vectors and outputs the mineralization probability value for each grid cell. A fractal regularization term is introduced into the loss function: Total Loss = Mean Squared Error (Error between Predicted Probability and True Label) + Fractal Regularization Term. The fractal regularization term is calculated by performing CA fractal analysis on the mineralization probability map output in the current batch and calculating its fractal dimension. The standard fractal dimension of the known mineralized area absolute difference Multiply by regularization coefficient Backpropagation is performed to update the network weights, so that the model output not only fits the training data, but is also constrained by fractal laws, and outputs the trained machine learning model (including the optimal weight parameters). Full-area prediction: Input the feature layer set of the entire working area, input the grid features of the entire area into the trained model, calculate the mineralization probability of each grid, and output the mineralization probability distribution map (raster data, value range 0-1). Post-hoc verification and target area delineation: CA fractal analysis is performed again on the mineralization probability map to determine the probability threshold classification (such as high, medium and low potential areas), and mineral exploration target areas are delineated in combination with the experience of geological experts.

[0018] In one embodiment, the quadtree adaptive mesh partitioning specifically includes: A threshold is set based on the local variance or coefficient of variation of the geochemical data. In areas exceeding the threshold, the data is recursively subdivided into small grids of 10m×10m to capture the weak, disseminated signals of veinlets in porphyry gold and silver deposits. In background field areas below the threshold, a large grid of 100m×100m is maintained. Simultaneously, during the gridding process, two sets of grid data, grid A and grid B, are generated for the geochemical data. Grid A uses nearest neighbor interpolation to retain local high-value anomalies in the original geochemical data, while grid B uses kriging interpolation to generate a smooth geochemical background field.

[0019] In one embodiment, the method for constructing the anisotropic variation function model is as follows: based on the occurrence information of the tectonic alteration zone, the strike, dip and dip angle parameters are determined, and in the Kriging interpolation, a long-range search radius is set along the construction direction of the ore-bearing structure, and a short-range search radius is set perpendicular to the stratigraphic direction.

[0020] In one embodiment, extracting the curvature and tortuosity of the linear structure as geometric features specifically involves: using remote sensing images or geophysical data to identify curvature abrupt change points and tortuosity distribution areas of the linear structure, and calculating curvature values ​​and tortuosity density to quantify the structural deformation intensity.

[0021] In one embodiment, gray-level co-occurrence matrix analysis is performed on multispectral or hyperspectral remote sensing images to calculate the energy, entropy, and contrast parameters within each grid cell. High entropy and high contrast regions correspond to the rough or disordered texture features of mylonite or tectonic breccia, serving as remote sensing proxy information for tectonic alteration zones.

[0022] In one embodiment, taking the fracture structure as the center, the fractal box counting method is used to analyze the logarithmic relationship of the number of mineral points within different buffer radii, and the radius interval with the highest linear correlation is determined as the optimal buffer range. The distance from each grid cell to the fracture structure is calculated, and the fractal density function is applied for power-law decay weighting to obtain a continuously changing fractal distance field.

[0023] In one embodiment, based on the original geochemical sampling point data, a fractal filtering method is used to extract the enrichment or depletion intensity values ​​of elements relative to the background field. Then, the singular values ​​are gridded using nearest neighbor interpolation or simple interpolation methods to ensure that local anomaly features are not smoothed, serving as input features indicating deep mineralization information.

[0024] In one embodiment, before inputting the extracted features into the neural network, principal component analysis is used to screen geometric features, texture features, lithological combination entropy, continuous fractal distance field and multifractal singular values, eliminating redundant features with low correlation to known mineral deposits and retaining the optimal feature subset for model training.

[0025] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent delineation of gold and silver mineral exploration target areas based on GIS multi-source information, characterized in that, Includes the following steps: Data preprocessing: Collect geological, geophysical and geochemical, and remote sensing multi-source geoscience information data of the work area, and use a quadtree adaptive gridding strategy to perform multi-scale gridding on the multi-source geoscience information data. At the same time, extract various mineralization information for medium-low temperature hydrothermal deposits and introduce an anisotropic variability function model for directional gridding, so that the data are correlated and interpolated along the mineralized structural construction direction. Feature extraction: Based on the preprocessed grid data, the curvature and tortuosity of linear structures are extracted as geometric features. The energy, entropy and contrast of remote sensing images are extracted as texture features through gray-level co-occurrence matrix analysis. The lithological diversity index is calculated within the grid cell as the lithological assemblage entropy. At the same time, based on the fracture data, the fractal box counting method is used to determine the optimal buffer radius and generate a continuous fractal distance field. Model Construction: The extracted geometric features, texture features, lithological combination entropy, continuous fractal distance field, and multifractal singular values ​​of geochemical data are used as input feature sets and fed into the neural network for training. A fractal regularization term is introduced into the loss function, which is the difference between the fractal dimension of the mineralization probability map predicted by the model and the standard fractal dimension of the known mineralization area as a penalty term to constrain the model output to conform to the geological fractal law. Target area delineation: The trained neural network is used to generate a mineralization probability map, and the prediction results are verified posteriorly by combining CA fractal theory to delineate mineral exploration target areas with different exploration potentials.

2. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, The quadtree adaptive mesh partitioning specifically includes: A threshold is set, and areas exceeding the threshold are recursively subdivided into small grids of 10m×10m. In background field areas below the threshold, a large grid of 100m×100m is maintained. At the same time, during the gridding process, two sets of grid data, grid A and grid B, are generated for geophysical and geochemical data respectively. Grid A uses nearest neighbor interpolation to retain local high-value anomalies in the original geochemical data, while grid B uses Kriging interpolation to generate a smooth geochemical background field.

3. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, The method for constructing the anisotropic variation function model is as follows: based on the occurrence information of the tectonic alteration zone, the strike, dip and dip angle parameters are determined, and in the Kriging interpolation, a long-range search radius is set along the construction direction of the ore-bearing structure, and a short-range search radius is set perpendicular to the stratigraphic direction.

4. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, The extraction of curvature and tortuosity zones of linear structures as geometric features specifically involves: using remote sensing images or geophysical data to identify curvature abrupt change points and tortuosity zone distribution areas of linear structures, and calculating curvature values ​​and tortuosity zone densities to quantify structural deformation intensity.

5. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, Gray-level co-occurrence matrix analysis is performed on multispectral or hyperspectral remote sensing images to calculate the energy, entropy, and contrast parameters within each grid cell. High entropy and high contrast regions correspond to the rough or disordered texture features of mylonite or tectonic breccia, serving as remote sensing proxy information for tectonic alteration zones.

6. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, Centered on the fracture structure, the fractal box counting method is used to analyze the logarithmic relationship of the number of mineral points within different buffer radii. The radius interval with the highest linear correlation is determined as the optimal buffer range. The distance from each grid cell to the fracture structure is calculated, and the fractal density function is applied for power-law decay weighting to obtain a continuously changing fractal distance field.

7. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, Based on the original geochemical sampling point data, a fractal filtering method is used to extract the enrichment or depletion intensity values ​​of elements relative to the background field, and the singular values ​​are gridded.

8. The intelligent delineation method for gold and silver mineral exploration target areas based on GIS multi-source information as described in claim 1, characterized in that, Before inputting the extracted features into the neural network, principal component analysis is used to screen geometric features, texture features, lithological combination entropy, continuous fractal distance field and multifractal singular values, eliminating redundant features with low correlation to known mineral deposits and retaining the optimal feature subset for model training.

Images