Spatial data labeling method for delineating prospecting target area

Abstract

The invention discloses a spatial data labeling method for delineating a prospecting target area, which comprises the following steps: collecting mining area data, preprocessing to obtain spatial data, and dividing the spatial data by using a Thiessen polygon based on the spatial data to obtain a typical ore deposit research to-be-labeled area model area, the method comprises the following steps: obtaining an initial value according to geological structure features and element features of a model area of a typical ore deposit research to-be-labeled area, obtaining a first index and a second index based on the initial value, obtaining a first label and a second label by using the first index through an evidence weight method, and correcting the first index to obtain a correction label; and marking the model area of the typical ore deposit research to-be-marked area to obtain a marked area, expressing the marked area through polygonal vector elements, training the vector elements and the marked area by using a deep learning algorithm to obtain a spatial data marking model, and outputting a visual marking result. The method not only can improve the efficiency of spatial data annotation, but also has good interpretability.

Description

Technical Field

[0001] The present invention relates to the field of data annotation, and particularly to a spatial data annotation method for delineating prospecting target areas. Background Art

[0002] In the field of mineral exploration, accurately delineating prospecting target areas is crucial for improving the efficiency and accuracy of mineral resource exploration. With the development of geological exploration technology, the limitations of traditional methods that solely rely on remote sensing images or geochemical data have become increasingly prominent. They cannot fully explore the geological feature correlations behind the data, resulting in insufficient accuracy and reliability in delineating prospecting target areas.

[0003] When existing spatial data annotation methods process complex geological data, they lack scientific classification and targeted processing of the annotation areas, affecting the efficiency and accuracy of subsequent analysis. In the model training process, traditional methods fail to fully utilize the powerful spatial feature extraction ability of deep learning algorithms, resulting in insufficient learning of the relationship between geological features and metallogenic potential by the annotation model, and the output annotation results are difficult to accurately guide mineral exploration practices. Summary of the Invention

[0004] The purpose of the present invention is to provide a spatial data annotation method for delineating prospecting target areas.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions: The first aspect of the present invention provides a spatial data annotation method for delineating prospecting target areas, including: Collecting mine area data and preprocessing it to obtain spatial data, where the spatial data includes remote sensing images, geochemical data, geological structure data, stratigraphic data, lithologic data, alteration data, geophysical data, and rock mass data; Constructing a three-dimensional Thiessen polygon based on the spatial data points projected onto the surface by the ore body distribution and the depth information, using the three-dimensional Thiessen polygon to divide the spatial data to obtain a model area for studying typical ore deposits in the ore body, and extracting the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical characteristics, geochemical element characteristics, and lithologic characteristics of the model area data for studying typical ore deposits to obtain initial values; Randomly generating multiple groups of initial weight values for the initial values, using the automatic regularization Gauss-Newton method to linearly approximate and optimize the initial weight values, iteratively calculating to obtain a first index, and nonlinearly optimizing the initial values and the initial weight values based on a support vector machine to obtain a second index; Using the first index as an input parameter of the evidence weight method to construct an evidence layer, obtaining a first label and a second label based on the threshold difference of the evidence layer, and correcting the first label based on the second index to obtain a corrected label; Using calibration tags and second tags to label the model area of the target area for typical ore deposit research to obtain a labeled area, representing the labeled area through polygon vector elements, training the polygon vector elements and the labeled area using a deep learning algorithm to obtain a spatial data annotation model, and using the spatial data annotation model to output a visual annotation result.

[0006] As a further method, the method for collecting and preprocessing mine area data to obtain spatial data includes: Collect elevation remote sensing images, collect geochemical exploration data with drilling depth information, collect geological structure data containing underground structure extension depth information, collect stratigraphic data, collect rock mass data, collect geophysical property data, collect lithologic data, collect alteration data, collect mineralization data, collect ore body data, use geographic information system technology to use the elevation remote sensing image as the spatial coordinate base, overlay the geochemical exploration data sampling points on the coordinate space coordinate base in the form of point elements according to the sampling point coordinates, overlay the geological structure data on the spatial coordinate base in the form of line elements or surface elements according to the geographical spatial coordinates, and use the bilinear interpolation method to unify the resolution of the spatial coordinate base to form spatial data.

[0007] As a further method, the method for constructing a Thiessen polygon based on the spatial data points projected onto the surface and depth information of the ore body distribution, using the Thiessen polygon to divide the spatial data to obtain the model area of the target area for typical ore deposit research, and extracting the initial values of the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical exploration characteristics, geochemical element characteristics, and lithologic characteristics of the model area data of the target area for typical ore deposit research, includes: Extract spatial data points and the corresponding depth information of the spatial data points based on the spatial data, import the spatial data points into the geographic information system, construct different spatial data point planes based on the depth information of different spatial data points, calculate the Euclidean distance between any two points in the spatial data point plane, use the Delaunay triangulation algorithm to generate the perpendicular bisectors of the connections between adjacent spatial data points, use the perpendicular bisectors to intersect to form the boundary of the Thiessen polygon, and divide the spatial data point plane into different model areas of the target area for typical ore deposit research based on the boundary of the Thiessen polygon; The geological structure characteristics and element characteristics of the model area data of the target area for typical ore deposit research are extracted to obtain initial values. Based on the geological structure data of the model area of the target area for typical ore deposit research, the fault intersection density is obtained, the outlier values and the proportion of outlier points of the element content in the geochemical exploration data in the model area of the target area for typical ore deposit research are statistically analyzed, the outlier values are normalized to obtain weight values, and the fault intersection density and the proportion of outlier points are weighted and summed to obtain the initial values.

[0008] As a further method, the method for randomly generating multiple groups of initial weight values for the initial values, linearly approximating and optimizing the initial weight values using the automatic regularization Gauss-Newton method, and iteratively calculating to obtain the first exponent, includes: Randomly generate multiple groups of initial weight values for the initial values according to the data categories of the model areas of the areas to be marked in the study of typical ore deposits. Use the initial values of the model areas of the areas to be marked in the study of typical ore deposits and the data of the model areas of the areas to be marked in the study of typical ore deposits to construct a first function through the ridge regression algorithm. Take each group of initial weight values as the starting point of iteration, and iteratively optimize the parameter weights of the first function through the automatic regularization Gauss-Newton method. Use the Jacobian matrix to adjust the parameter weights during the iteration until the first function converges to obtain the optimized parameter weights. The formula of the automatic regularization Gauss-Newton method is: where is the weight vector of the th iteration, is the weight vector of the th iteration, is the Jacobian matrix, is the transpose matrix, which multiplies with the Jacobian matrix to form an approximate Hessian matrix, is the regularization parameter of the th iteration, is the th iteration, the th residual between the initial value of the model area of the area to be marked in the study of typical ore deposits and the predicted value of the first function, is the th iteration, the th residual between the initial value of the model area of the area to be marked in the study of typical ore deposits and the predicted value of the first function, is a very small positive number to prevent the denominator from being zero, is the identity matrix, is the th iteration's residual vector, represents the total number of model areas of the areas to be marked in the study of typical ore deposits; Multiply the optimized parameter weights by the parameter values of the first function and sum them to obtain the first index.

[0009] As a further method, the method for non-linearly optimizing the initial values and initial weight values based on the support vector machine to obtain the second index includes: Take the initial values as the target variables, use the initial weight values corresponding to the initial values as the feature variables, and obtain the training data set based on the initial values of all model areas of the areas to be marked in the study of typical ore deposits in the spatial data; Based on the kernel function selection support vector machine model, if the Pearson coefficient between the initial values is greater than 0.7, the linear kernel is selected; if the global Moran index between the initial values is greater than 0.5, the polynomial kernel function is selected; if it is less than 0.3, the radial basis kernel function is selected. Use the training data set to train the support vector machine, and optimize the parameters of the support vector machine using the sequential minimal optimization algorithm to reduce the structural risk, obtaining a trained support vector machine model. Input the initial weight values of the model area of the typical deposit research target area into the trained support vector machine model to obtain the output value, and perform normalization processing on the output value to obtain the second index.

[0010] As a further method, the method of using the first index as an input parameter of the evidence weight method to construct an evidence layer and obtaining the first label and the second label according to the difference of the evidence layer threshold includes: Based on the model area of the typical deposit research target area, use the first index as an input parameter of the evidence weight method. Use the remote sensing image, geochemical data, geological structure data, stratigraphic data, lithology data, alteration data, geophysical exploration data, and rock mass data of the model area of the typical deposit research target area as evidence factors. Statistically calculate the conditional probabilities of the evidence factors in the mineralization and non-mineralization states, calculate the evidence factor weight values by taking the natural logarithm of the ratio of the occurrence probabilities of the evidence factors under the mineralization and non-mineralization conditions, and use the geographic information system to perform spatial superposition calculation on the evidence factor weight value layers to form an evidence layer; Perform frequency statistics on the evidence layer values, set the potential threshold using the natural breaks classification method based on the data distribution, and divide the labels according to the comparison results of the evidence layer values and the potential threshold. If the evidence layer value is greater than the potential threshold, it is divided into the first label; if it is less than the potential threshold, it is divided into the second label.

[0011] As a further method, the method of correcting the first label based on the second index to obtain the corrected label includes: Interpolate the element content anomaly data of the model area of the typical deposit research target area using the IDW and global Kriging methods to generate an element content raster map, calculate the mean and standard deviation of the raster map, use the element mean plus twice the standard deviation to obtain the lower limit of the anomaly value, and use the contrast value method to divide the single element content by the element mean to obtain the first correction parameter; Calculate the fault density by extracting the unit area fault length from the geological data of the model area of the typical deposit research target area, perform line density analysis on the fold axis data to statistically extract the fold frequency of the unit distance fold times, generate a slope raster using the DEM to calculate the slope standard deviation, and normalize and sum the fault density, fold frequency, and slope standard deviation to obtain the second correction parameter; Use the second index and the first and second correction parameters to correct the first label through the correction formula, and the correction formula is: Where For the correction label, is the first exponent, is the second exponent, is the second correction parameter, is the maximum value of the second correction parameter, is the first correction parameter, is the maximum value of the first correction parameter, is the first label, , , , is the difference coefficient, is the subscript of the difference coefficient.

[0012] As a further method, the method of using the correction label and the second label to label the model area of the area to be labeled in the study of typical ore deposits to obtain the labeled area, and representing the labeled area by polygon vector elements, includes: Using the correction label and the second label to label the model area of the area to be labeled in the study of typical ore deposits to obtain the labeled area, calculating the variance of the boundary data of the labeled area, setting the complexity threshold as the sum of the 75% quantile of the variance and 1.5 times the interquartile range based on the variance, using the complexity threshold to divide the low-complexity labeled area and the high-complexity labeled area, and decomposing the low-complexity labeled area using the row-column convex decomposition formula. The formula is: where is the set composed of all convex polygons after decomposition, represents any one of the convex polygons in, which is an independent convex region unit after decomposition, represents the convex polygon any vertex in, which is used to define the object range of the interior angle calculation, vertex the interior angle at, is the vector pointing from vertex to the vector, is the vector pointing from vertex to the vector, is the low-complexity labeled area, is the k-th cutting operation on, is the union operation; Making the low-complexity labeled area into a regular polygon vector element; for the high-complexity labeled area, extracting the boundary coordinate points and connecting the coordinate points in sequence to form a closed figure to obtain the polygon vector element.

[0013] As a further method, the method of using a deep learning algorithm to train polygon vector features and annotation areas to obtain a spatial data annotation model and using the spatial data annotation model to output a visual annotation result includes: Normalize the geometric coordinates of the polygon vector features, perform one-hot encoding on the annotation area categories, generate a learning dataset based on the polygon vector features and the one-hot encoding, and divide the learning dataset into a training set, a validation set, and a test set; Input the training set into the deep learning model. The model extracts the spatial features of the polygon vector features through convolutional layers and pooling layers, makes a preliminary prediction based on the spatial features, calculates the difference between the prediction result and the one-hot encoded label using the cross-entropy loss function. During training, input the validation set into the model to verify the overfitting state of the model through the change of the loss function value and the accuracy. Feed the difference back to each layer of the model through the backpropagation algorithm to adjust the model parameters, continuously narrow the difference until the model training ends. Input the test set into the trained deep learning model, evaluate the model performance using accuracy and recall to obtain the spatial data annotation model, and use this spatial data annotation model to output a visual annotation result.

[0014] The second aspect of the present invention provides a spatial data annotation system for delineating an ore prospecting target area, including: A data acquisition module for collecting mining area data and performing preprocessing to obtain spatial data, where the spatial data includes remote sensing images, geochemical exploration data, geological structure data, stratigraphic data, rock mass data, geophysical exploration data, and lithologic data; A typical ore deposit research target area model area generation module for constructing Thiessen polygons based on spatial data points and depth information, using the Thiessen polygons to divide the spatial data to obtain a typical ore deposit research target area model area, and extracting the geological structure features and element features of the typical ore deposit research target area model area data to obtain initial values; An index calculation module for randomly generating multiple groups of initial weight values for the initial values, linearly approximating and optimizing the initial weight values using the automatic regularization Gauss-Newton method, iteratively calculating to obtain a first index, and nonlinearly optimizing the initial values and the initial weight values based on a support vector machine to obtain a second index; A label generation module for using the first index as an input parameter of the evidence weight method to construct an evidence layer, obtaining a first label and a second label based on the evidence layer threshold difference, and correcting the first label based on the second index to obtain a corrected label; A label model generation module for using the corrected label and the second label to annotate the typical ore deposit research target area model area to obtain an annotation area, representing the annotation area through polygon vector features, using a deep learning algorithm to train the polygon vector features and the annotation area to obtain a spatial data annotation model, and using the spatial data annotation model to output a visual annotation result.

[0015] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) By collecting multi-source mine area data such as remote sensing images, geochemical exploration data, geological structure data, stratigraphic data, rock mass data, geophysical exploration data, and lithological data, and using geographic information system technology and bilinear interpolation method for preprocessing, the present invention realizes the organic integration of multi-source data, more comprehensively reflects the geological characteristics of the mine area, and significantly improves the accuracy and reliability of prospecting target area delineation.

[0016] (2) By using the automatic regularization Gauss-Newton method and support vector machine to perform linear approximation optimization and non-linear optimization respectively, combining the weight-of-evidence method to construct an evidence layer and correcting labels through multiple parameters, the present invention realizes the judgment of metallogenic potential.

[0017] (3) By dividing the complexity of the annotation area, using the row-column convex decomposition formula to obtain regular vector elements for low-complexity annotation areas, retaining the original contour for high-complexity annotation areas, combining deep learning algorithms to extract the spatial features of vector elements and training the model, this differential processing not only improves the data processing efficiency, but also ensures that the key geological features in complex areas are not lost. The finally output visual annotation results can intuitively present the distribution of prospecting potential and provide a clear decision-making basis for geological exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a step flow chart of a spatial data annotation method for delineating prospecting target areas in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Refer to Figure 1 As shown, the present invention provides a spatial data annotation method for delineating prospecting target areas, including: Collecting typical ore deposit data in the mine area for preprocessing to obtain spatial data, where the spatial data includes remote sensing images, geochemical exploration data, geological structure data, stratigraphic data, rock mass data, geophysical exploration data, and lithological data; In the actual assessment, remote sensing images are collected to obtain multi-spectral images of the mining area, including four bands of blue, green, red, and near-infrared. Geochemical data are collected, and data of 1200 soil sampling points covering an area of 80 km² are collected. Eight elements such as Cu, Pb, Zn, Au, and Ag are detected. The sampling depth is 0 - 20 cm, and the drilling verification depth is 300 m. Geological structure data are collected. Twenty-eight main faults are interpreted by geophysical seismic reflection method, including strike, dip angle, and underground extension depth of 50 - 800 m. Twelve fold axes are extracted based on geological mapping, and the positions of the axes and the attitudes of the two wings are recorded. The remote sensing images are spatially registered with geochemical points and geological structure lines in the WGS84 coordinate system using ArcGIS Pro. Radiometric calibration and atmospheric correction are performed on the remote sensing images to enhance mineralization alteration information. All data are resampled to a resolution of 5 m using bilinear interpolation to generate spatial data, including remote sensing bands, geochemical elements, and geological structure attributes.

[0021] Based on the spatial data points projected onto the surface and depth information of the ore body distribution, a three-dimensional Thiessen polygon is constructed. The spatial data is divided using the three-dimensional Thiessen polygon to obtain a model area of the target area for typical ore deposit research. The geological structure characteristics and element characteristics of the data in the model area of the target area for typical ore deposit research are extracted to obtain initial values. In the actual assessment, the coordinates and depth information of geochemical points are extracted, and depth layers of 0 - 100 m, 100 - 200 m, and 200 - 300 m are constructed. In each depth layer, Thiessen polygons are generated using the CreateThiessenPolygons tool. A total of 896 model areas of the target area for typical ore deposit research are divided in the spatial data. The number of fault intersections within 1 km² of each model area of the target area for typical ore deposit research is counted. For example, the number of fault intersections in a certain model area of the target area for typical ore deposit research is 12 / km². The length of the fold axis per unit area in the model area of the target area for typical ore deposit research is calculated in m / km². The element characteristics of the model area of the target area for typical ore deposit research are extracted. Taking the Cu element as an example, the mean value is 80 ppm, the standard deviation is 25 ppm, and the outlier is defined as the mean plus twice the standard deviation (>130 ppm). The proportion of outliers in a certain model area of the target area for typical ore deposit research is calculated. For example, there are 15 geochemical points, of which 3 are outliers, and the proportion is 20%. After normalization, the weight value is 0.2. The fault intersection density and the proportion of outliers are weighted and summed to obtain an initial value of 7.28.

[0022] Multiple groups of initial weight values are randomly generated for the initial values. The automatic regularization Gauss-Newton method is used to linearly approximate and optimize the initial weight values, and the first index is obtained through iterative calculation. Nonlinear optimization of the initial values and initial weight values is performed based on the support vector machine to obtain the second index. It needs to be explained that there are some linear relationships in spatial data. For example, some geological variables show an approximate linear correlation with mineralization potential within a certain range. The automatic regularized Gauss-Newton method can efficiently optimize these linear relationships. It uses the Jacobian matrix to construct an approximate Hessian matrix through iterative calculations, and gradually adjusts the weights so that the model can quickly and accurately fit the linear part of the spatial data, thereby obtaining a more accurate first index for characterizing the linear characteristics of mineralization potential. There are often complex nonlinear relationships between the characteristics of spatial data, such as the combined impact of geological structure and element distribution on mineralization potential. With the help of the basis kernel function, the support vector machine can map data to a high-dimensional feature space, and transform the data that was originally nonlinearly separable in the low-dimensional space into linearly separable data, thereby effectively processing these nonlinear relationships and obtaining the second index for mining the nonlinear characteristics of mineralization potential.

[0023] In the actual evaluation, the regularization parameter λ is set to 0.3 and the first function is constructed: ,in is the fault intersection density, For the proportion of outliers, the automatic regularized Gauss-Newton method is used to perform linear approximation optimization on the initial weight values, which are randomly generated, for example is 0.5, is 0.3, the maximum number of iterations is set to 100, the convergence threshold is 1e-6, the regularization parameter is adjusted dynamically, the initial α is 0.1, and the weight parameter is obtained after optimization is 0.65, is 0.43, is 0.2, and the first exponent is calculated to be 8.176.

[0024] In the actual evaluation, the initial value is used as the target variable, and the initial weight value corresponding to the initial value is used as the characteristic variable. The training data set is obtained based on the initial values of the model area of all typical mineral deposits in the spatial data. The global Moran index between the initial values is calculated to be 0.2 less than 0.3, then the radial basis kernel function RBF is selected, and C is obtained as 1 and γ is 0.1 through grid search. The support vector machine is trained using the training data set. For example, if the initial weight value is input as [0.5, 0.3], the second index after output normalization is 0.82.

[0025] The first index is used as an input parameter of the evidence weight method to construct an evidence layer, a first label and a second label are obtained according to the difference in the threshold of the evidence layer, and the first label is corrected based on the second index to obtain a corrected label; In the actual evaluation, based on the study of typical ore deposits, the first index is used as the input parameter of the evidence weight method for the model area of the target area to be marked. The remote sensing images, geochemical data, and geological structure data of the model area of the target area to be marked by the study of typical ore deposits are used as evidence factors. The conditional probabilities of fault density in ore-forming or non-ore-forming areas are statistically calculated. For example, the probability that the fault density in the ore-forming area is greater than 10 faults / km² is 0.75, and that in the non-ore-forming area is 0.3. The weight value of the evidence factor 0.916 is obtained by taking the natural logarithm of the ratio of the occurrence probabilities of the evidence factors under ore-forming and non-ore-forming conditions. The evidence layers are calculated using ArcGIS Pro spatial overlay to generate continuous rasters. The evidence layers are classified into 5 categories using the natural breaks classification method. The potential threshold is set as the upper limit of the 4th category, which is 0.75. The first label greater than 0.75 and the second label less than or equal to 0.75 are divided.

[0026] In the actual evaluation, for the abnormal data of element content in the model area of the target area to be marked by the study of typical ore deposits, methods such as IDW and global Kriging are used to interpolate and generate element content raster maps, and the mean value and standard deviation of the raster maps are calculated. For example, the lower limit of the abnormal value is obtained by adding twice the standard deviation of 50 ppm to the mean value of 80 ppm of the Cu element, which is 130 ppm. The first correction parameter 1.625 is obtained by dividing the single element content by the element mean value using the contrast value method. For the geological data of the model area of the target area to be marked by the study of typical ore deposits, the fault density is extracted by calculating the fault length per unit area, and the line density analysis is performed on the fold axis data to statistically extract the fold frequency per unit distance. The slope standard deviation is calculated using the DEM to generate slope rasters. After normalizing the fault density of 12 faults / km², the fold frequency of 500 m / km², and the slope standard deviation of 15°, the weighted sum is obtained as 0.9, 0.7, and 0.8, and the second correction parameter is 0.81. The first label is corrected using the second index and the first and second correction parameters through the correction formula, and the corrected label is 0.13. The model area of the target area to be marked by the study of typical ore deposits is labeled using the corrected label and the second label to obtain the labeled area. The labeled area is represented by polygon vector elements. The polygon vector elements and the labeled area are trained using a deep learning algorithm to obtain a spatial data labeling model, and the spatial data labeling model is used to output a visual labeling result.

[0027] In the actual evaluation, calibration labels and second labels are used to label the model areas of the areas to be labeled in the study of typical ore deposits to obtain labeled areas. The variance of the boundary data of the labeled areas is calculated. For example, the variance of a model area of an area to be labeled in the study of a typical ore deposit is 95. Based on the variance, the quartile method is used to set the complexity threshold as the sum of the 75th percentile of the variance and 1.5 times the interquartile range, and the complexity threshold of 210 is obtained. The low-complexity labeled areas and high-complexity labeled areas are divided using the complexity threshold. The low-complexity areas with a variance lower than 210 are decomposed into 4 convex polygons using the row-column convex decomposition formula. For example, the vertex coordinate sequence of a convex polygon is [(100,200),(150,250),(200,220),(120,180),(180,190),(140,230)]. The high-complexity areas with a variance higher than 210 are directly extracted with 12 coordinate points to form a closed curve, and the coordinate points are connected in sequence to form a closed figure to obtain polygon vector elements.

[0028] In the actual evaluation, the geometric coordinates of the polygon vector elements are normalized, the category of the labeled area is one-hot encoded, a learning dataset is generated based on the polygon vector elements and the one-hot encoding, and the learning dataset is divided into a training set, a validation set, and a test set according to 7:2:1.

[0029] The U-Net model is selected. The input layer is set as the normalized coordinates (x_norm, y_norm), the encoding path is 4 convolutional layers of 3×3, the max pooling is 2×2, the decoding path is the skip connection between the upsampling layer and the encoding layer, and the output layer is 3 neurons. Among them, [1,0,0] in the one-hot encoding represents high potential, [0,1,0] represents medium potential, and [0,0,1] represents low potential. The training parameters are set. The batch size is 16, the learning rate is 0.0001, and the iteration is 200 rounds. The loss function is set as cross entropy, and the optimizer is selected as Adam. The training stops when the accuracy of the validation set is greater than or equal to 95%. Finally, the test set indicators are calculated, the accuracy is 96.2%, the recall rate is 0.94, and the F1 is 0.95. A spatial data labeling model is obtained. The visualization labeling result is output using this spatial data labeling model, and the labeling result is rendered as a thematic map using the visualization module of the prospecting prediction spatial data labeling system. High-potential areas (red): the area accounts for 12%, labeled as type I target areas; medium-potential areas (yellow): the area accounts for 25%, labeled as type II target areas; low-potential areas (blue): the area accounts for 63%, labeled as type III areas. The remote sensing image and geological structure lines are superimposed to generate a three-dimensional visualization scene with contour lines.

[0030] In this embodiment, the method for collecting mining area data and preprocessing it to obtain spatial data includes: Collect elevation remote sensing images, collect geochemical exploration data with drilling depth information, collect geological structure data containing underground structure extension depth information, collect stratigraphic data, collect rock mass data, collect geophysical property data, collect lithologic data, collect alteration data, collect mineralization data, collect ore body data, use geographic information system technology to take the elevation remote sensing image as the spatial coordinate base, overlay the geochemical exploration data sampling point coordinates on the coordinate space coordinate base in the form of point elements, overlay the geological structure and other data on the spatial coordinate base in the form of line elements or surface elements according to geographical coordinates, and use the bilinear interpolation method to unify the resolution of the spatial coordinate base to form spatial data.

[0031] In this embodiment, the method for constructing a Thiessen polygon based on spatial data points and depth information, using the Thiessen polygon to divide the spatial data to obtain a model area of the target area for typical ore deposit research, and extracting the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical characteristics, geochemical element characteristics, lithologic characteristics, etc. of the model area data of the target area for typical ore deposit research to obtain initial values includes: Extract spatial data points and the corresponding depth information of the spatial data points from the spatial data, import the spatial data points into the geographic information system, construct different spatial data point planes based on the depth information of different spatial data points, calculate the Euclidean distance between any two points in the spatial data point plane, use the Delaunay triangulation algorithm to generate the perpendicular bisectors of the connections between adjacent spatial data points, use the perpendicular bisectors to intersect to form the boundaries of the Thiessen polygon, and divide the spatial data point plane into different model areas of the target area for typical ore deposit research based on the boundaries of the Thiessen polygon; Extract the geological structure characteristics and element characteristics of the model area data of the target area for typical ore deposit research to obtain initial values. Obtain the fault intersection density based on the geological structure data of the model area of the target area for typical ore deposit research, count the outlier values and the proportion of outlier points of the element content in the geochemical exploration data in the model area of the target area for typical ore deposit research, normalize the outlier values as weight values, and perform weighted summation on the fault intersection density and the proportion of outlier points to obtain the initial values.

[0032] In this embodiment, the method for randomly generating multiple groups of initial weight values for the initial values, using the automatic regularization Gauss-Newton method to linearly approximate and optimize the initial weight values, and iteratively calculating to obtain the first exponent includes: Randomly generate multiple groups of initial weight values for the initial values according to the data categories of the model area of the target area for typical ore deposit research. Use the initial values of the model area of the target area for typical ore deposit research and the data of the model area of the target area for typical ore deposit research to construct a first function through the ridge regression algorithm. Take each group of initial weight values as the iteration starting point, iteratively optimize the parameter weights of the first function through the automatic regularization Gauss-Newton method, and adjust the parameter weights using the Jacobian matrix during the iteration until the first function converges to obtain the optimized parameter weights. The formula of the automatic regularization Gauss-Newton method is: where is the weight vector of the -th iteration, is the weight vector of the -th iteration, is the Jacobian matrix, is the transpose matrix, which multiplies with the Jacobian matrix to form an approximate Hessian matrix, is the regularization parameter of the -th iteration, is the -th iteration, and is the residual between the initial value of the model area of the -th typical deposit research target area to be marked and the predicted value of the first function, is the -th iteration, and is the residual between the initial value of the model area of the -th typical deposit research target area to be marked and the predicted value of the first function, is a very small positive number to prevent the denominator from being zero, is the identity matrix, is the residual vector of the -th iteration, represents the total number of model areas of the typical deposit research target area to be marked; Multiply the optimized parameter weights with the first function parameter values and sum them to obtain the first index.

[0033] In this embodiment, the method for non-linearly optimizing the initial value and the initial weight value based on the support vector machine to obtain the second index includes: Taking the initial value as the target variable, using the initial weight value corresponding to the initial value as the feature variable, and obtaining a training data set based on the initial values of all typical deposit research target area model areas in the spatial data; Selecting a support vector machine model based on the kernel function. If the Pearson coefficient between the initial values is greater than 0.7, select the linear kernel. If the global Moran index between the initial values is greater than 0.5, select the polynomial kernel function. If it is less than 0.3, select the radial basis kernel function. Use the training data set to train the support vector machine, optimize the support vector machine parameters using the sequential minimal optimization algorithm to reduce the structural risk, obtain the trained support vector machine model, input the initial weight value of the typical deposit research target area model area into the trained support vector machine model to obtain the output value, and normalize the output value to obtain the second index.

[0034] In this embodiment, the method for constructing an evidence layer using the first index as an input parameter of the evidence weight method and obtaining the first label and the second label according to the evidence layer threshold difference includes: Based on the study of typical ore deposits, for the model area of the target area to be marked, the first index is used as the input parameter of the evidence weight method. The remote sensing images, geochemical data, and geological structure data of the model area of the target area to be marked by typical ore deposit research are used as evidence factors. The conditional probabilities of the evidence factors in the ore-forming and non-ore-forming states are statistically calculated. The weight value of the evidence factor is obtained by taking the natural logarithm of the ratio of the occurrence probabilities of the evidence factor under the ore-forming and non-ore-forming conditions. The geographical information system is used to perform spatial superposition calculation on the evidence factor weight value layer to form an evidence layer; Frequency statistics are performed on the evidence layer values. Based on the data distribution, the natural break classification method is used to set the potential threshold. Labels are divided according to the comparison result of the evidence layer value and the potential threshold. If the evidence layer value is greater than the potential threshold, it is divided into the first label; if it is less than the potential threshold, it is divided into the second label.

[0035] In this embodiment, the method for correcting the first label based on the second index to obtain a corrected label includes: For the element content anomaly data of the model area of the target area to be marked by typical ore deposit research, methods such as IDW and global Kriging are used to interpolate and generate an element content raster map. The mean and standard deviation of the raster map are calculated. The lower limit of the anomaly value is obtained by adding twice the standard deviation to the element mean. The contrast value method is used to divide the single element content by the element mean to obtain the first correction parameter; For the geological data of the model area of the target area to be marked by typical ore deposit research, the fault density is extracted by calculating the fault length per unit area, the fold frequency is extracted by analyzing the line density of the fold axis data and counting the number of folds per unit distance, and the slope standard deviation is calculated by generating a slope raster using DEM. After normalizing the fault density, fold frequency, and slope standard deviation, they are weighted and summed to obtain the second correction parameter; The first label is corrected using the second index and the first and second correction parameters through a correction formula, and the correction formula is: Where is the corrected label, is the first index, is the second index, is the second correction parameter, is the maximum value of the second correction parameter, is the first correction parameter, is the maximum value of the first correction parameter, is the first label, 、 、 、 is the coefficient of variation, is the subscript of the coefficient of variation.

[0036] In this embodiment, the method of using a calibration label and a second label to label the model area of the area to be marked in the study of typical ore deposits to obtain a marked area and representing the marked area as a polygon vector element includes: Using a calibration label and a second label to label the model area of the area to be marked in the study of typical ore deposits to obtain a marked area, calculating the variance of the boundary data of the marked area, setting a complexity threshold based on the variance as the sum of the 75th percentile of the variance and 1.5 times the interquartile range using the interquartile method, using the complexity threshold to divide the low-complexity marked area and the high-complexity marked area, and decomposing the low-complexity marked area using the row-column convex decomposition formula. The formula is: Where is the set composed of all convex polygons after decomposition, represents any one of the convex polygons in, which is an independent convex region unit after decomposition, represents the convex polygon any vertex in, which is used to define the object range for calculating the interior angle, Vertex at the interior angle, is the vector pointing from vertex to the vector, is the vector pointing from vertex to the vector, is the low-complexity marked area, is for the k-th cutting operation performed on, is the union operation; Making the low-complexity marked area into a regular polygon vector element; for the high-complexity marked area, extracting the boundary coordinate points and connecting the coordinate points in sequence to form a closed figure to obtain the polygon vector element.

[0037] In this embodiment, the method of using a deep learning algorithm to train the polygon vector element and the marked area to obtain a spatial data annotation model and using the spatial data annotation model to output a visual annotation result includes: Normalizing the geometric coordinates of the polygon vector element, performing one-hot encoding on the marked area category, generating a learning data set based on the polygon vector element and the one-hot encoding, and dividing the learning data set into a training set, a validation set, and a test set; Input the training set into the deep learning model. The model extracts the spatial features of polygon vector features through convolutional layers and pooling layers, makes a preliminary prediction based on the spatial features, calculates the difference between the prediction result and the one-hot encoded label using the cross-entropy loss function. During training, input the validation set into the model to verify the overfitting state of the model through the loss function value and accuracy change. Feed the difference back to each layer of the model through the backpropagation algorithm to adjust the model parameters, continuously narrow the difference until the model training ends. Input the test set into the trained deep learning model, evaluate the model performance using accuracy and recall rate, obtain the spatial data annotation model, and use this spatial data annotation model to output the visual annotation result.

[0038] The second aspect of the present invention also provides a spatial data annotation system for delineating prospecting target areas, including: A data acquisition module, which is used to collect mine area data and preprocess it to obtain spatial data. The spatial data includes spatial data such as structures, strata, rock masses, geochemical exploration, geophysical exploration, remote sensing, heavy minerals, ore bodies, mineralization, alteration, and lithology. A typical deposit research target area model area generation module, which is used to construct a three-dimensional Thiessen polygon based on the spatial data points and depth information projected onto the surface by the ore body distribution, divide the spatial data using the three-dimensional Thiessen polygon to obtain the ore body model typical deposit research target area model area, and extract the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical exploration characteristics, geochemical element characteristics, lithology characteristics, etc. of the typical deposit research target area model area data to obtain initial values. An index calculation module, which is used to randomly generate multiple groups of initial weight values for the initial values, linearly approximate and optimize the initial weight values using the automatic regularization Gauss-Newton method, iteratively calculate to obtain the first index, and perform non-linear optimization on the initial values and initial weight values based on the support vector machine to obtain the second index. A label generation module, which is used to construct an evidence layer by using the first index as the input parameter of the evidence weight method, obtain the first label and the second label according to the evidence layer threshold difference, and correct the first label based on the second index to obtain the corrected label. A label model generation module, which is used to label the typical deposit research target area model area with the corrected label and the second label to obtain the labeled area, represent the labeled area through polygon vector features, train the polygon vector features and the labeled area using the deep learning algorithm to obtain the spatial data annotation model, and use the spatial data annotation model to output the visual annotation result.

[0039] The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of this technology can make various modifications or supplements to the described specific embodiments or use similar methods to replace them. As long as they do not deviate from the structure of the invention or exceed the scope defined by this claims, they should fall within the protection scope of the present invention.

Claims

1. A spatial data annotation method for delineating ore prospecting target areas, characterized in that, Including the following steps: Collect typical deposit data in the mining area and preprocess it to obtain spatial data, where the spatial data includes structural, stratigraphic, rock mass, geochemical exploration, geophysical exploration, remote sensing, heavy minerals, ore body, mineralization, alteration, and lithology spatial data; Construct a three-dimensional Thiessen polygon based on the spatial data points projected onto the surface by the ore body distribution and the depth information, use the three-dimensional Thiessen polygon to divide the spatial data to obtain the target area for the study of the typical deposit of the ore body model, and extract the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical exploration characteristics, geochemical exploration element characteristics, and lithology characteristics of the data in the sub-target model area to obtain initial values; Randomly generate multiple groups of initial weight values for the initial values, use the automatic regularization Gauss-Newton method to linearly approximate and optimize the initial weight values, perform iterative calculations to obtain the first index, and perform non-linear optimization on the initial values and initial weight values based on the support vector machine to obtain the second index; Use the first index as the input parameter of the evidence weight method to construct an evidence layer, obtain the first label and the second label according to the threshold difference of the evidence layer, and correct the first label based on the second index to obtain a corrected label; Use the corrected label and the second label to label the sub-target model area to obtain a labeled area, represent the labeled area by polygon or volume vector elements, use a deep learning algorithm to train the polygon or volume vector elements and the labeled area to obtain a spatial data labeling model, and use the spatial data labeling model to output a visual labeling result.

2. The spatial data annotation method for delineating ore prospecting target areas according to claim 1, wherein The method for collecting typical deposit data in the mining area and preprocessing it to obtain spatial data includes: Collect elevation remote sensing images, collect geochemical exploration data with drilling depth information, collect geological structure data containing underground structure extension depth information, collect stratigraphic data, collect rock mass data, collect geophysical property data, collect lithology data, collect alteration data, collect mineralization data, collect ore body data, use geographic information system technology to use the elevation remote sensing image as the spatial coordinate base, overlay the geochemical exploration data sampling point coordinates on the coordinate space coordinate base in the form of point elements, overlay the geological structure data on the spatial coordinate base in the form of line elements or surface elements according to the geographical spatial coordinates, and use the bilinear interpolation method to unify the resolution of the spatial coordinate base to form spatial data.

3. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that, The method for constructing a three-dimensional Thiessen polygon based on the spatial data points projected onto the surface by the ore body distribution and the depth information, using the three-dimensional Thiessen polygon to divide the spatial data to obtain the sub-target ore body model area, and extracting the geological structure characteristics, stratigraphic characteristics, rock mass characteristics, geophysical exploration characteristics, geochemical exploration element characteristics, and lithology characteristics of the data in the sub-target model area to obtain initial values includes: Extract spatial data points and the corresponding depth information of the spatial data points based on the spatial data, import the spatial data points into the geographic information system, construct different spatial data point planes based on the depth information of different spatial data points, calculate the Euclidean distance between any two points in the spatial data point plane, use the Delaunay triangulation algorithm to generate the perpendicular bisectors of the connections between adjacent spatial data points, use the perpendicular bisectors to intersect to form the boundary of the Thiessen polygon, and divide the spatial data point plane into different target areas for the study of typical deposits based on the boundary of the Thiessen polygon; Obtain the initial values based on the geological structure characteristics and element characteristics of the data in the sub-target model area. Obtain the fault intersection density based on the geological structure data of the model area to be marked in the study of typical ore deposits. Statistically calculate the proportion of outlier values and outlier points in the geochemical data in the sub-target model area. Normalize the outlier values as weight values, and perform weighted summation on the fault intersection density and the proportion of outlier points to obtain the initial values.

4. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that The method of randomly generating multiple groups of initial weight values for the initial values and using the automatic regularization Gauss-Newton method to perform linear approximation optimization on the initial weight values and obtaining the first exponent through iterative calculation includes: Randomly generate multiple groups of initial weight values for the initial values according to the data categories in the sub-target model area. Use the initial values in the sub-target model area and the data in the sub-target model area to construct a first function through the ridge regression algorithm. Use each group of initial weight values as the starting point for iteration, and iteratively optimize the parameter weights of the first function through the automatic regularization Gauss-Newton method. Adjust the parameter weights using the Jacobian matrix during the iteration until the first function converges to obtain the optimized parameter weights. The formula for the automatic regularization Gauss-Newton method is: where is the weight vector for the -th iteration, is the weight vector for the -th iteration, is the Jacobian matrix, is the transpose matrix, which multiplies with the Jacobian matrix to form an approximate Hessian matrix, is the regularization parameter for the -th iteration, is the residual between the initial value and the predicted value of the first function for the -th iteration in the -th model area of the study target area for a typical deposit, is the residual between the initial value and the predicted value of the first function for the -th iteration in the -th model area of the study target area for a typical deposit, is a very small positive number to prevent the denominator from being zero, is the identity matrix, is the residual vector for the -th iteration, represents the total number of model areas of the study target area for a typical deposit; Multiply the optimized parameter weights by the parameter values of the first function and sum them to obtain the first exponent.

5. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that, The method of performing non-linear optimization on the initial values and the initial weight values based on the support vector machine to obtain the second exponent includes: Use the initial values as the target variables, and use the corresponding initial weight values of the initial values as the feature variables to obtain the training data set based on the initial values of all typical ore deposit study areas to be marked in the spatial data; Select the support vector machine model based on the kernel function. If the Pearson coefficient between the initial values is greater than 0.7, select the linear kernel. If the global Moran index between the initial values is greater than 0.5, select the polynomial kernel function. If it is less than 0.3, select the radial basis kernel function. Use the training data set to train the support vector machine, and optimize the parameters of the support vector machine through the sequential minimal optimization algorithm to reduce the structural risk to obtain the trained support vector machine model. Input the initial weight values of the typical ore deposit study area to be marked into the trained support vector machine model to obtain the output values, and perform normalization processing on the output values to obtain the second exponent.

6. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that The method of using the first exponent as the input parameter of the evidence weight method to construct the evidence layer and obtaining the first label and the second label according to the threshold difference of the evidence layer includes: Based on the sub-target model area, use the first exponent as the input parameter of the evidence weight method. Use the remote sensing images, geochemical data, geological structure data, stratigraphic data, lithology data, alteration data, geophysical exploration data, and rock mass data in the sub-target model area as evidence factors. Statistically calculate the conditional probabilities of the evidence factors in the ore-forming and non-ore-forming states. Calculate the evidence factor weight values by taking the natural logarithm of the ratio of the occurrence probabilities of the evidence factors under the ore-forming and non-ore-forming conditions. Use the geographic information system to perform spatial overlay calculations on the evidence factor weight value layers to form the evidence layer; Perform frequency statistics on the evidence layer values. Set the potential threshold using the natural break classification method based on the data distribution. Divide the labels according to the comparison results of the evidence layer values and the potential threshold. If the evidence layer value is greater than the potential threshold, it is classified as the first label. If it is less than the potential threshold, it is classified as the second label.

7. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that, The method for correcting the first label based on the second exponent to obtain a corrected label includes: Interpolating the abnormal data of the element content in the sub-target model area using the IDW and global Kriging methods to generate an element content raster map, calculating the mean and standard deviation of the raster map, obtaining the lower limit of the abnormal value by adding twice the standard deviation to the element mean, and obtaining the first correction parameter by dividing the single-element content by the element mean using the contrast value method; Extracting the fault density by calculating the fault length per unit area of the geological data in the sub-target model area, analyzing the line density of the fold axis data to statistically extract the fold frequency per unit distance, calculating the standard deviation of the slope using the DEM to generate a slope raster, normalizing the fault density, fold frequency, and slope standard deviation and then weighted summing them to obtain the second correction parameter; Using the second exponent and the first and second correction parameters to correct the first label through a correction formula, and the correction formula is: Among them is a correction label is the first exponent is the second exponent is the second correction parameter is the maximum value of the second correction parameter is the first correction parameter is the maximum value of the first correction parameter is the first label 、 、 、 are difference coefficients is the subscript of the difference coefficient 8. A spatial data annotation method for delineating prospecting target areas according to claim 1, characterized in that The method for obtaining a labeled area by using the corrected label and the second label to label the model area of the typical ore deposit research target area, and representing the labeled area as a polygon vector element includes: Labeling the sub-target model area with the corrected label and the second label to obtain a labeled area, calculating the variance of the boundary data of the labeled area, setting the complexity threshold as the sum of the 75% quantile of the variance and 1.5 times the interquartile range based on the variance, using the complexity threshold to divide the low-complexity labeled area and the high-complexity labeled area, and decomposing the low-complexity labeled area using the row-column convex decomposition formula, and the formula is: Among them is the set composed of all convex polygons after decomposition, represents any convex polygon in, which is an independent convex region unit after decomposition, represents the convex polygon any vertex in, which is used to limit the object range for calculating the interior angle, vertex the interior angle at the vertex, is the vector pointing from vertex to ; is the vector pointing from vertex to ; is the low-complexity annotation area, is the k-th cutting operation on, is the union operation; Making the low-complexity labeled area into a regular polygon vector element; for the high-complexity labeled area, extracting the boundary coordinate points and connecting the coordinate points in sequence to form a closed figure to obtain a polygon vector element.

9. A spatial data annotation method for delineating ore prospecting target areas according to claim 1, characterized in that, The method for training the polygon vector element and the labeled area using a deep learning algorithm to obtain a spatial data labeling model and outputting a visual labeling result using the spatial data labeling model includes: Normalizing the geometric coordinates of the polygon vector element, performing one-hot encoding on the labeled area category, generating a learning data set based on the polygon vector element and the one-hot encoding, and dividing the learning data set into a training set, a validation set, and a test set; Inputting the training set into a deep learning model, the model extracts the spatial features of the polygon vector element through convolutional layers and pooling layers, makes a preliminary prediction based on the spatial features, calculates the difference between the prediction result and the one-hot encoded label using the cross-entropy loss function, validates the overfitting state of the model by inputting the validation set into the model during training through the change of the loss function value and the accuracy, feeds back the difference to each layer of the model through the backpropagation algorithm to adjust the model parameters, continuously reducing the difference until the model training ends, inputting the test set into the trained deep learning model, evaluating the model performance using accuracy and recall to obtain a spatial data labeling model, and using this spatial data labeling model to output a visual labeling result.

10. A spatial data annotation system for delineating ore prospecting target areas, which is used to execute a spatial data annotation method for delineating ore prospecting target areas according to any one of claims 1 to 9, characterized in that, The system includes: A data acquisition module for acquiring mining area data and preprocessing it to obtain spatial data, and the spatial data includes remote sensing images, geochemical data, geological structure data, stratigraphic data, rock mass data, geophysical exploration data, and lithological data; The sub-target model area generation module is used to construct Thiessen polygons based on spatial data points and depth information, divide the space data using the Thiessen polygons to obtain sub-target model areas, and extract the geological structure features and element features of the sub-target model area data to obtain initial values; The index calculation module is used to randomly generate multiple groups of initial weight values for the initial values, linearly approximate and optimize the initial weight values using the automatic regularization Gauss-Newton method, iteratively calculate to obtain the first index, and perform non-linear optimization on the initial values and initial weight values based on the support vector machine to obtain the second index; The label generation module is used to construct an evidence layer with the first index as the input parameter of the weight-of-evidence method, obtain the first label and the second label according to the evidence layer threshold difference, and correct the first label based on the second index to obtain a corrected label; The label model generation module is used to use the corrected label and the second label to label the model area to be labeled in the study of typical ore deposits to obtain a labeled area, represent the labeled area through polygon vector elements, train the polygon vector elements and the labeled area using a deep learning algorithm to obtain a spatial data labeling model, and use the spatial data labeling model to output a visual labeling result.

Images