Artificial intelligence-based mine exploration system

Abstract

The application discloses a mine area exploration system based on artificial intelligence, which comprises a multi-source data acquisition module, a data optimization module, a mine area candidate area potential identification module and a mine area intelligent exploration module. The application relates to the technical field of artificial intelligence data analysis and computer vision, in particular to a mine area exploration system based on artificial intelligence. The mine area candidate area potential identification module and the mine area intelligent exploration module are combined innovatively, the accuracy of the ore-forming area identification and the pertinence of the exploration are improved, the K nearest neighbor algorithm is used to construct a weighted graph structure, an improved graph convolutional neural network with a parallel updating mechanism is introduced, high-precision automatic potential classification of a large range of candidate mine areas is realized, a composite chaotic mapping initialization method and a sine exponential inertia weight improved particle swarm optimization algorithm are introduced, the accuracy and stability of yield prediction are improved, and high-precision yield prediction of each subarea of the ore-forming area is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence data analysis and computer vision, and particularly discloses a mineral area exploration system based on artificial intelligence. BACKGROUND

[0002] The mineral area exploration system based on artificial intelligence is an intelligent information processing and decision support system for mineral resource exploration. It uses artificial intelligence technology to collect, fuse, analyze and model multi-source data, realizes data processing automation, target area intelligent identification and exploration scheme intelligent decision-making in the exploration process through deep feature extraction and pattern recognition, and thus improves the decision-making efficiency of mineral exploration, optimizes resource allocation and reduces the overall operation cost.

[0003] However, the traditional mineral area exploration system has obvious technical problems in determining the exploration range, which leads to waste of exploration resources, prolongation of exploration period, incomplete exploration and insufficient identification of mineralization areas, and thus affects the integrity and accuracy of exploration results. The existing mineral area potential evaluation model fails to fully utilize the spatial adjacency relationship between candidate area features and is difficult to consider multi-order neighborhood information in the feature extraction process, which leads to insufficient classification accuracy of mineral area potential, unclear class boundaries and low differentiation degree of different potential categories. The hyperparameter setting of the existing mineral area yield prediction model is unreasonable, and the fixed initialization method and linear inertia weight adjustment are used in the hyperparameter optimization process, which leads to insufficient search space coverage and easy falling into local optimum, and thus the prediction accuracy of the model is insufficient. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides a mine area exploration system based on artificial intelligence. In view of the obvious technical problems in the determination of the exploration range of the traditional mine area exploration system, which leads to waste of exploration resources, prolongation of the exploration period, and incomplete exploration and insufficient identification of the ore-forming area, thereby affecting the integrity and accuracy of the exploration results, the present scheme innovatively combines the mine area candidate region potential identification module with the mine area intelligent exploration module. First, the mine area candidate region potential identification module is used to automatically classify and evaluate the potential of a large range of candidate regions, and high-potential ore-forming areas are preferentially selected. Then, the mine area intelligent exploration module is used to perform partitioned yield prediction and accurate mining area determination in the high-potential ore-forming area. This can effectively prevent resource and time waste caused by an excessively large exploration range, and avoid the risk of omission caused by an excessively small exploration range. It improves the accuracy of ore-forming area identification and the pertinence of exploration, enhances the utilization efficiency of exploration data, improves the integrity of exploration results, and realizes the transition from blind global exploration to accurate partitioned exploration. In view of the technical problems in the existing mine area potential evaluation model that fail to fully utilize the spatial adjacency relationship between candidate region features and that are difficult to simultaneously consider multi-order neighborhood information during feature extraction, thereby leading to insufficient classification accuracy of mine area potential, unclear class boundaries, and low differentiation degree of different potential categories, the present scheme innovatively proposes to use a K-nearest neighbor algorithm to construct a weighted graph structure, calculate node feature similarity using a Gaussian kernel function, establish adjacency relationships, and generate a graph structure that can reflect spatial adjacency and feature similarity. An improved graph convolutional neural network with a parallel update mechanism is introduced to realize synchronous extraction and fusion of multi-order neighborhood features, and a multi-head attention mechanism is combined to enhance feature expression capability. This can effectively improve the integrity and differentiation degree of feature expression, significantly improve the accuracy and robustness of mine area potential identification, enhance the separability of high-potential and low-potential regions, improve the generalization ability of the model under complex geological backgrounds, and realize high-precision automatic potential classification of a large range of candidate mine areas. In view of the technical problems of the existing mine area yield prediction model, such as unreasonable hyperparameter setting and reliance on fixed initialization methods and linear inertia weight adjustment during hyperparameter optimization, which leads to insufficient search space coverage and easy trapping in local optimum, thereby resulting in insufficient prediction accuracy of the model, the present scheme innovatively introduces a composite chaotic mapping initialization method into the particle swarm optimization algorithm to enhance the diversity of particle initial distribution and global search capability. A sinusoidal exponential inertia weight is used to realize adaptive search intensity adjustment with dynamic fluctuations during iteration, balancing global exploration and local development. This significantly improves the efficiency and accuracy of hyperparameter search of the mine area yield prediction model, enhances the adaptability of the model to complex geological and mineralization conditions, improves the accuracy and stability of yield prediction, and realizes high-precision yield prediction of each sub-area of the ore-forming area and optimization of the mining scheme.

[0005] The technical scheme adopted by the present application is as follows: The present application provides a mine area exploration system based on artificial intelligence, which comprises a multi-source data acquisition module, a data optimization module, a mine area candidate region potential identification module and a mine area intelligent exploration module.

[0006] The multi-source data acquisition module specifically acquires mine area exploration original data through data acquisition operations.

[0007] The data optimization module is used to improve the quality of original data and the adaptability of models, and specifically acquires mine area exploration optimized data through image data preprocessing and structured data preprocessing.

[0008] The mine area candidate region potential identification module is used to identify high-potential ore-forming regions in a large range of mine area candidate regions, and specifically comprises the following steps: a mine area potential evaluation model is established, historical data is used to fully train the model, real-time data is input into the trained model, and the potential evaluation results of each candidate mine area are output; the large range of candidate regions is screened according to the potential evaluation results, and the ore-forming region results are obtained.

[0009] The mine area intelligent exploration module is used to implement zonification yield prediction and accurate mining area determination in the ore-forming region, and specifically comprises the following steps: a mine area yield prediction model is established, historical data is used to train the prediction model, an improved particle swarm optimization algorithm is used to search for the optimal super parameter combination of the model, the super parameters of the prediction model are updated, the mine area yield prediction model with the optimal performance is obtained, real-time data of each sub-region is input into the optimal prediction model, and the yield prediction results of each sub-region are output; finally, a mine area zonification yield distribution map is generated according to the yield prediction results, and intelligent exploration planning and efficient operation management of the target mine area are realized.

[0010] Further, the multi-source data acquisition module specifically acquires mine area exploration original data by acquiring mine area related data from an external data platform; the mine area exploration original data comprises historical mine area screening data, historical ore-forming region yield prediction data, real-time mine area screening data and real-time ore-forming region yield prediction data; the historical mine area screening data and the real-time mine area screening data both comprise medium-resolution remote sensing image data and regional structured data; the historical mine area screening data further comprises historical mine area types; the historical ore-forming region yield prediction data and the real-time ore-forming region yield prediction data both comprise high-resolution remote sensing image data, regional environmental field feature data and element distribution feature data; the historical ore-forming region yield prediction data further comprises mine area yield data.

[0011] Further, the data optimization module specifically comprises the following steps:

[0012] The image data preprocessing specifically includes image cloud removal processing, radiation correction processing, image format standardization processing and image quality enhancement, and remote sensing optimized image data is obtained.

[0013] The structured data preprocessing specifically includes data cleaning, data standardization and data encoding processing, and structured optimized data is obtained.

[0014] Further, the mine area candidate region potential identification module specifically includes the following steps:

[0015] The mine area potential evaluation model is established, including the following steps:

[0016] The candidate region identification feature extraction is specifically image feature extraction and structured feature extraction on the image data and the structured data respectively, mine area candidate image features and mine area candidate structured features are generated, and the two types of features are spliced and fused to obtain mine area candidate comprehensive features.

[0017] The discriminant feature matrix acquisition is used to map the mine area candidate comprehensive features to a low-dimensional discriminant subspace, and enhance the class separability between the mine area candidate regions; specifically, the mine area candidate comprehensive features and the corresponding historical mine area type labels are jointly input, a nonlinear kernel mapping relationship between samples is constructed, nonlinear discriminant compression processing is performed on the mine area candidate comprehensive features by kernel Fisher discriminant analysis method, the intra-class and inter-class scatter is calculated, and the generalized eigenvalue problem is solved to construct a projection matrix, and finally a set of candidate low-dimensional discriminant features is obtained, and a candidate mine discriminant feature matrix is obtained.

[0018] The candidate region graph structure establishment is used to model the adjacency relationship and similarity of the discriminant features in space; specifically including the following steps:

[0019] The two-dimensional feature matrix is constructed, specifically linear normalization processing is performed on each dimension feature in the candidate mine discriminant feature matrix, each feature is sorted in descending order based on feature importance, and the sorted feature sequence is mapped in sequence according to column, and a two-dimensional feature matrix is obtained.

[0020] The feature importance is specifically a comprehensive analysis of the inter-class dispersion and the intra-class dispersion of the jth dimension feature of the candidate mine discriminant feature matrix Y, and a single-dimensional Fisher discriminant score is calculated.

[0021] The node set construction is specifically that the two-dimensional feature matrix is cut by sliding according to a set window size and a sliding step, local feature blocks are extracted in sequence, and each feature block is vectorized to form a node feature vector, and all node feature vectors form a node set.

[0022] Edge connection establishment, specifically, the Euclidean distance is calculated for any two node feature vectors in the node set, and the Euclidean distance is mapped into a similarity weight by using a Gaussian kernel function, for each node, the k nodes with the highest similarity are selected to form its neighbor set, a weighted edge is established according to the similarity weight, and the similarity weight is assigned to the adjacent matrix element to form an initial weighted adjacent matrix, then the initial weighted adjacent matrix is symmetrized and a self-loop edge is added to obtain a final weighted adjacent matrix;

[0023] Graph structure output, specifically, first calculate the center coordinates of each node in the two-dimensional feature matrix, and normalize them to After the interval, the spatial position encoding of the node is generated by linear mapping, then the spatial position encoding and the node feature are spliced and fused to obtain a node fusion feature vector, and all node fusion feature vectors are sequentially combined to form an initial node feature matrix, finally, the node set, edge set, final weighted adjacent matrix and initial node feature matrix are combined to obtain a candidate graph structure;

[0024] Mine area potential result output, specifically, the candidate graph structure is input into an improved graph convolutional neural network with a parallel update mechanism to extract mine area potential classification features, the mine area potential classification features are mapped through a fully connected layer and the probability distribution of each potential category of the candidate mine area is calculated through a Softmax function, the category corresponding to the maximum probability is selected as the final prediction result to obtain the potential evaluation result of the candidate mine area;

[0025] The improved graph convolutional neural network includes a parallel multi-order graph convolutional layer and a multi-head attention mechanism; specifically, first set the number of parallel branches to Q, and each branch corresponds to a normalized adjacent matrix of a different order; each branch receives the initial node feature matrix and the normalized adjacent matrix of the corresponding order in the first layer, and simultaneously performs graph convolution operation to generate single-order node features of this layer; in the subsequent lth layer, each branch receives the single-order node features output by the previous layer and the normalized adjacent matrix of the corresponding order and simultaneously performs graph convolution operation to generate single-order node features of this layer, after parallel convolution of each branch, the convolution node features of all branches are spliced according to the feature dimension to generate multi-order neighborhood features, realizing parallel update and layer-by-layer extraction of multi-order neighborhood features; the multi-order neighborhood features are input into the multi-head attention mechanism to obtain mine area potential classification features;

[0026] Mine area potential evaluation model training, specifically, the historical mine area screening data is used as training data for model training, and finally a trained mine area potential evaluation model is obtained;

[0027] The metallogenic region selection process involves inputting real-time mineral area screening data into a trained mineral area potential assessment model, outputting the potential assessment results of each candidate mineral area, filtering based on the potential assessment results of each candidate mineral area, and selecting areas with high potential mineral areas as the final metallogenic regions.

[0028] Furthermore, the intelligent exploration module for the mining area specifically includes the following steps:

[0029] Establishing a mining area production prediction model includes the following steps:

[0030] The extraction of mining area production prediction features involves extracting production prediction image features and production prediction structured features from image data and structured data respectively, generating production prediction image features and production prediction structured features, and then splicing and fusing the two types of features to obtain comprehensive production prediction features.

[0031] The output of the production forecast results is specifically obtained by performing nonlinear regression analysis on the comprehensive characteristics of the production forecast based on a feedforward neural network model to obtain the production forecast results for the mining area.

[0032] The training of the mining area production prediction model involves using historical mining area production prediction data as training data to train the prediction model, and finally obtaining the trained mining area production prediction model.

[0033] Prediction model performance optimization specifically involves obtaining the optimal hyperparameter combination of the model through an improved search algorithm, updating the prediction model hyperparameters based on the optimal hyperparameter combination, and obtaining the optimal mine production prediction model; including the following steps:

[0034] Initializing the particle swarm involves encoding the hyperparameters of the trained mine production prediction model into search individual position vectors, and generating N individual particle position vectors using a composite chaotic mapping initialization method. Each individual represents a candidate combination of prediction model hyperparameters, resulting in the initial particle swarm. The formula used is as follows:

[0035] ;

[0036] In the formula, Indicates the first The position of each individual particle. This represents the position of the i-th particle. Represents a random number in the range [0,1].

[0037] Calculating particle fitness values ​​specifically involves calculating the fitness values ​​of particles within the population. The performance of the mining area production prediction model based on the individual particle position is used as the particle fitness value.

[0038] The sine exponential inertia weight is calculated, specifically by adaptively adjusting the search intensity through the iteration number, and a sine term is introduced to realize periodic fluctuation, and the formula is as follows:

[0039] ;

[0040] In the formula, denotes the sine exponential inertia weight, denotes the maximum iteration number, t denotes the current iteration number, denotes a random number in the range of [0, 1];

[0041] Particle updating, specifically particle velocity updating through the sine exponential inertia weight, and particle position updating based on the updated particle velocity;

[0042] Particle optimal position updating, specifically re-evaluating the fitness value of all updated particles, and comparing the fitness value of the current particle with the global optimal position of the current particle, if the fitness value of the current particle is better, updating the global optimal position of the particle;

[0043] Particle search termination, specifically when the particle fitness value is higher than the fitness threshold and the maximum iteration number is reached, terminating the search and obtaining the global optimal position of the particle; the global optimal position of the particle specifically refers to the optimal hyperparameter combination of the model;

[0044] Prediction model hyperparameter updating, specifically adjusting the hyperparameters of the mine area yield prediction model according to the optimal hyperparameter combination of the model, to obtain the optimal mine area yield prediction model;

[0045] Mine area partition yield prediction, specifically dividing the ore-forming area into multiple sub-areas according to the equal-area grid, inputting the real-time ore-forming area yield prediction data of each sub-area into the optimal mine area yield prediction model, and obtaining the mine area yield prediction results of each sub-area;

[0046] Mine exploration area determination, specifically analyzing the mine area yield prediction results of each sub-area, generating a complete mine area partition yield distribution map based on spatial visualization technology, accurately positioning and prioritizing different yield grade areas according to the distribution map, forming a decision support scheme for mine production management and resource allocation, thereby optimizing the mining plan and utilization efficiency of mineral resources, and realizing intelligent exploration planning and efficient operation management of the target mine area.

[0047] The above-mentioned scheme has the following beneficial effects:

[0048] (1) In view of the technical problems that the determination of the exploration range in the traditional mine area exploration system has obvious deficiencies, which leads to waste of exploration resources, prolongation of the exploration period, incomplete exploration and insufficient identification of the mineralization area, thereby affecting the integrity and accuracy of the exploration results, the scheme innovatively combines the mine area candidate region potential identification module with the mine area intelligent exploration module. First, the mine area candidate region potential identification module is used to automatically classify and evaluate the potential of a large range of candidate regions, and high-potential mineralization areas are preferentially selected. Then, the mine area intelligent exploration module is used to conduct partitioned yield prediction and accurate mining area determination in the high-potential mineralization area. This can effectively prevent resource and time waste caused by an excessively large exploration range, and avoid the risk of omission caused by an excessively small exploration range. It improves the accuracy of mineralization area identification and the pertinence of exploration, enhances the utilization efficiency of exploration data, improves the integrity of exploration results, and realizes the transition from blind global exploration to accurate partitioned exploration.

[0049] (2) In view of the technical problems that the existing mine area potential evaluation model fails to fully utilize the spatial adjacency relationship between candidate region features, and it is difficult to consider multi-order neighborhood information during feature extraction, which leads to insufficient classification accuracy of mine area potential, unclear class boundaries, and low differentiation degree of different potential classes, the scheme innovatively proposes to use K-nearest neighbor algorithm to construct a weighted graph structure, use Gaussian kernel function to calculate node feature similarity and establish adjacency relationship, and generate a graph structure that can reflect spatial adjacency and feature similarity. An improved graph convolutional neural network with parallel update mechanism is introduced to realize synchronous extraction and fusion of multi-order neighborhood features, and a multi-head attention mechanism is combined to enhance feature expression capability. This can effectively improve the integrity and differentiation degree of feature expression, significantly improve the accuracy and robustness of mine area potential identification, enhance the separability of high-potential and low-potential regions, improve the generalization ability of the model under complex geological background, and realize high-precision automatic potential classification of large-scale candidate mine areas.

[0050] (3) In view of the technical problems that the hyperparameter setting of the existing mine area yield prediction model is unreasonable, and the hyperparameter optimization process relies on fixed initialization method and linear inertia weight adjustment, which leads to insufficient search space coverage and easy falling into local optimum, thereby leading to insufficient prediction accuracy of the model, the scheme innovatively introduces a compound chaotic mapping initialization method in the particle swarm optimization algorithm to enhance the diversity of particle initial distribution and global search ability. The sine exponential inertia weight is used to realize adaptive search intensity adjustment with dynamic fluctuation of iterations, and to balance global exploration and local development. This significantly improves the efficiency and accuracy of hyperparameter search of the mine area yield prediction model, enhances the adaptability of the model to complex geology and mineralization conditions, improves the accuracy and stability of yield prediction, realizes high-precision yield prediction of each sub-area of the mineralization area and optimization of mining scheme. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A module schematic diagram of the mining area exploration system based on artificial intelligence provided by the present application;

[0052] Figure 2 A flowchart schematic diagram of the mining area candidate region potential identification module;

[0053] Figure 3 A flowchart schematic diagram of the mining area candidate region potential identification module for establishing a mining area potential evaluation model;

[0054] Figure 4 A flowchart schematic diagram of the ore-forming area yield prediction module;

[0055] Figure 5 A flowchart schematic diagram of the ore-forming area yield prediction module for model performance optimization;

[0056] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, are used to explain the present application, and do not constitute a limitation of the present application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0058] In the description of the present application, it should be understood that the terms 'upper', 'lower', 'front', 'back', 'left', 'right', 'top', 'bottom', 'inner', 'outer' and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the systems or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0059] Embodiment one, refer to Figure 1 The mining area exploration system based on artificial intelligence provided by the present application comprises a multi-source data acquisition module, a data optimization module, a mining area candidate region potential identification module, a model optimization module and a mining area intelligent exploration module.

[0060] The multi-source data acquisition module specifically acquires mining area related data from an external data platform, obtains mining area exploration original data, and sends the data to the data optimization module.

[0061] The data optimization module receives the data sent by the multi-source data acquisition module, and is used to improve the quality of the original data and the model adaptability, specifically, through image data preprocessing and structured data preprocessing, the mine area exploration optimization data is obtained, and the data is sent to the mine area candidate area potential identification module and the mine area intelligent exploration module;

[0062] The mine area candidate area potential identification module receives the data sent by the data optimization module, and is used to identify high-potential mineralization areas in a large range of mine area candidate areas, specifically, by establishing a mine area potential evaluation model, and using historical data to fully train the model, real-time data is input into the trained model, and the potential evaluation results of each candidate mine area are output, the large range of candidate areas is screened according to the potential evaluation results, and the mineralization area results are obtained, and the data is sent to the mine area intelligent exploration module;

[0063] The mine area intelligent exploration module receives the data sent by the data optimization module and the mine area candidate area potential identification module, and is used for implementing partitioned yield prediction and accurate mining area determination in the mineralization area; Specifically, first, a mine area yield prediction model is established, and historical data is used to train the prediction model, then an improved particle swarm optimization algorithm is used to search for the optimal super parameter combination of the model, the super parameters of the prediction model are updated, and the optimal mine area yield prediction model is obtained, then the real-time data of each sub-area is input into the optimal prediction model, and the yield prediction results of each sub-area are output, finally, the mine area partitioned yield distribution map is generated according to the yield prediction results, and the intelligent exploration planning and efficient operation management of the target mine area are realized.

[0064] By performing the above operation, in view of the obvious technical problems in the determination of the exploration range in the traditional mine area exploration system, which leads to waste of exploration resources, prolongation of exploration period, and incomplete exploration and insufficient mineralization area identification, thereby affecting the integrity and accuracy of the exploration results, the present scheme innovatively combines the mine area candidate area potential identification module and the mine area intelligent exploration module, first, the mine area candidate area potential identification module is used to automatically classify and evaluate the potential of the large range of candidate areas, and the high-potential mineralization area is preferentially screened out; then the mine area intelligent exploration module is used to perform partitioned yield prediction and accurate mining area determination in the high-potential mineralization area; which can effectively prevent the waste of resources and time caused by too large exploration range, and avoid the omission risk caused by too small exploration range, improve the accuracy of mineralization area identification and the pertinence of exploration, enhance the utilization efficiency of exploration data, improve the integrity of exploration results, and realize the change from blind global exploration to accurate partitioned exploration.

[0065] Embodiment two, refer to Figure 1The embodiment is based on the above-mentioned embodiment, and the multi-source data acquisition module specifically acquires the mine area exploration related data on an external platform to obtain mine area exploration original data; the mine area exploration original data includes historical mine area screening data, historical ore-forming area yield prediction data, real-time mine area screening data, and real-time ore-forming area yield prediction data; the historical mine area screening data and the real-time mine area screening data both include medium-resolution remote sensing image data and regional structured data; the historical mine area screening data further includes historical mine area types; the historical ore-forming area yield prediction data and the real-time ore-forming area yield prediction data both include high-resolution remote sensing image data, regional environmental field feature data, and element distribution feature data; the historical ore-forming area yield prediction data further includes mine area yield data; the medium-resolution remote sensing image data is specifically multispectral remote sensing image data, has a spatial resolution of 20-30 meters, has a wide coverage range and a high update frequency, and is suitable for preliminary screening of mine areas in a large area range and potential identification of mine area candidate regions; the regional structured data is used for identifying the ore-forming geological background of a target region; includes lithology distribution map quantitative data, structural line quantitative data, stratum map quantitative data, and contact zone and rock mass boundary quantitative data; the historical mine area types include high-potential mine areas, medium-potential mine areas, low-potential mine areas, and no-potential mine areas; the high-resolution remote sensing image data is specifically multispectral remote sensing image data with a spatial resolution better than 5 meters, and is suitable for mineralization patch identification, alteration feature extraction, and mine body boundary refinement in a small range; the element distribution feature data includes soil composition analysis data, rock sample composition analysis data, and water body sample composition analysis data.

[0066] Embodiment three, refer to Figure 1 The embodiment is based on the above-mentioned embodiment, and the data optimization module specifically includes the following steps:

[0067] The image data preprocessing is used for uniformly optimizing the image data, specifically includes image cloud removal processing, radiation correction processing, image format standardization processing, and image quality enhancement, and remote sensing optimized image data is obtained;

[0068] The image data includes medium-resolution remote sensing image data and high-resolution remote sensing image data;

[0069] The image cloud removal processing is used for removing the ground object shielding interference caused by clouds, cloud shadows, etc. in the remote sensing image, improving the effective information coverage range and analysis accuracy of the image, and specifically adopts a cloud detection algorithm based on a mask to automatically identify the high-light area in the image and generate a cloud mask to shield the shielded area;

[0070] The radiation correction processing is used for eliminating the spectral inconsistency problems of remote sensing images generated under different time, different satellite sensors and atmospheric conditions, and ensuring the comparability of image brightness and reflectivity. Specifically, the digital values of the original image are converted into apparent and ground object reflectivity by a radiation calibration and atmospheric correction method, and the spectral characteristics are unified.

[0071] The image format standardization processing is used for unifying the size and pixel value range of image data. Specifically, the remote sensing image is scaled to a unified size, and the image pixel value is normalized to map it to a standard numerical interval of 0 to 1, thereby eliminating the numerical differences caused by different image sources and imaging conditions.

[0072] The image enhancement is used for expanding sample diversity. Specifically, random geometric transformation of the image and random brightness and contrast adjustment of the image are performed.

[0073] The structured data preprocessing is used for unified optimization processing of structured data, specifically including data cleaning, data standardization and data encoding processing, to obtain structured optimized data.

[0074] The structured data includes regional structured data, regional environmental field feature data and element distribution feature data.

[0075] The data cleaning is used to improve the quality and consistency of the structured data. Specifically, the structured data is subjected to missing value filling, outlier removal and field standardization processing. The missing value filling specifically fills the missing values in the geological field, geophysical detection points and geochemical samples by a mean filling method. The outlier removal specifically detects and removes extreme values and logical outliers in the original data by a Z-Score algorithm. The field standardization processing specifically standardizes the data in different formats by a unified unit conversion rule to ensure the consistency of the model input.

[0076] The data standardization processing is used to unify the dimension of the structured numerical data. Specifically, all continuous variables are mapped to the [0, 1] interval by a min-max normalization method.

[0077] The data encoding processing is used to convert the classification fields or discrete label fields in the original data into recognizable numerical vectors. Specifically, the category fields in the original data are encoded by a one-hot encoding method to convert discrete text and label variables into sparse numerical vectors.

[0078] Embodiment four, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above-mentioned embodiments. The mine area candidate region potential identification module specifically includes the following steps:

[0079] The mine area potential evaluation model is established, including the following steps:

[0080] The candidate area identification feature extraction is used to extract the fusion features reflecting the regional metallogenic potential, specifically, image feature extraction and structured feature extraction are performed on the image data and structured data respectively, mine area candidate image features and mine area candidate structured features are generated, and the two types of features are spliced and fused to obtain mine area candidate comprehensive features;

[0081] The image feature extraction specifically uses a convolutional neural network to learn the spatial structure features of the medium-resolution remote sensing image data in the mine exploration optimization data, to obtain the mine area candidate image features;

[0082] The structured feature extraction specifically uses a multi-layer perception network to extract deep features from the regional structured data in the mine exploration optimization data, to obtain the mine area candidate structured features;

[0083] The discriminant feature matrix acquisition is used to map the mine area candidate comprehensive features to a low-dimensional discriminant subspace, to enhance the class separability between the mine area candidate regions; Specifically, the mine area candidate comprehensive features and the corresponding historical mine area type labels are jointly input to construct a nonlinear kernel mapping relationship between samples, the mine area candidate comprehensive features are processed by nonlinear discriminant compression through kernel Fisher discriminant analysis method, the intra-class and inter-class scatter is calculated, and the generalized eigenvalue problem is solved to construct a projection matrix, finally a set of candidate low-dimensional discriminant features is obtained, and a candidate mine discriminant feature matrix is obtained;

[0084] The candidate area graph structure establishment is used to model the adjacency relationship and similarity of discriminant features in space; Specifically, the following steps are included:

[0085] The two-dimensional feature matrix is constructed, specifically, the features in the candidate mine discriminant feature matrix are linearly normalized, and the features are sorted in descending order based on feature importance, and the sorted feature sequence is mapped in sequence by column to obtain a two-dimensional feature matrix;

[0086] The feature importance is used to measure the discriminant ability of each dimension feature in the candidate mine discriminant feature matrix in distinguishing different potential mine area categories; Specifically, the inter-class dispersion and intra-class dispersion of the jth dimension feature of the candidate mine discriminant feature matrix Y are comprehensively analyzed, and its single-dimensional Fisher discriminant score is calculated; The formula is as follows:

[0087] ;

[0088] In the formula, represents the number of class c samples, represents the mean value of the jth dimension feature of the cth class sample, μj represents the overall mean of all samples in the jth dimension, σjc represents the variance of the jth feature of the cth sample, min represents the minimum integer, preventing the denominator from being zero, and C represents the total number of potential mine area categories, Fj represents the single-dimensional Fisher discriminant score of the jth feature, used to evaluate the importance of the feature;

[0089] Node set construction, which is used to convert a two-dimensional feature matrix into a node set required for a graph structure, specifically, the two-dimensional feature matrix is cut and slid according to a set window size and sliding step, local feature blocks are extracted in turn, and each feature block is vectorized to form a node feature vector, and all node feature vectors form a node set; the formula used is as follows:

[0090] ;

[0091] In the formula, N represents the total number of nodes, and H represents the height of the two-dimensional feature matrix, W represents the width of the two-dimensional feature matrix, h represents the height of the sliding window, and w represents the width of the sliding window, h represents the sliding step of the window in the height direction, w represents the sliding step of the window in the width direction, floor represents the floor operator;

[0092] Edge connection establishment, which is used to build a weighted edge connection relationship based on feature similarity in the node set to form the adjacency information of the graph structure; specifically, the Euclidean distance is calculated for any two node feature vectors in the node set, and the Euclidean distance is mapped to a similarity weight using a Gaussian kernel function, for each node, the k nodes with the highest similarity are selected to form its neighbor set, a weighted edge is established according to the similarity weight, and the similarity weight is assigned to the adjacency matrix element to form an initial weighted adjacency matrix, then the initial weighted adjacency matrix is symmetrized and a self-loop edge is added to obtain a final weighted adjacency matrix; the formula used is as follows:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] In the formula, sim(i, j) represents the similarity weight of node i and node j, dis(i, j) represents the Euclidean distance of node i and node j, The Gaussian kernel bandwidth parameter is represented by k, which represents the number of neighbors in the KNN and is used to determine the connection density of each node. This represents the initial weighted adjacency matrix. This represents the transpose of the initial weighted adjacency matrix. This represents the identity matrix, used to add self-loop edges. This represents the final weighted adjacency matrix. This represents information about the edge between node i and node j. Let i represent the set of neighbors of node i;

[0098] The graph structure output is used to incorporate the spatial location information of nodes and fuse it with node content features during the graph structure construction process to generate a complete graph structure. Specifically, it first calculates the center coordinates of each node in the two-dimensional feature matrix and normalizes them to... After the interval is defined, a spatial location code for the node is generated through linear mapping. This spatial location code is then concatenated and fused with the node features to obtain a fused node feature vector. All fused node feature vectors are then sequentially combined to form an initial node feature matrix. Finally, the node set, edge set, final weighted adjacency matrix, and initial node feature matrix are combined to obtain the candidate region map structure. The formula used is as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, This represents the spatial location encoding vector of the i-th node. The weight matrix represents the spatial location encoding mapping. Represents the spatial location encoding bias vector. This represents the row index of node i in the two-dimensional feature matrix. This represents the column index of node i in the two-dimensional feature matrix. This indicates a splicing and merging operation. This represents the feature transformation weight matrix after splicing and fusion. This represents the fused feature vector of the i-th node. Representing node characteristics, Represents the initial node feature matrix. This represents the fused feature vector of the first node. Indicates the first Each node fuses its feature vector. Indicates the candidate region map structure. a set of nodes, a set of edges, constructed by connecting each node with the nodes in its neighborhood set into edges;

[0104] a mine potential result output, used for automatic potential classification evaluation, specifically, a candidate area graph structure is input into an improved graph convolutional neural network adopting a parallel update mechanism, mine potential classification features are extracted, the mine potential classification features are mapped through a full connection layer and the probability distribution of each potential category of the candidate mine area is calculated through a Softmax function, the category corresponding to the maximum probability is selected as the final prediction result, and the potential evaluation result of the candidate mine area is obtained;

[0105] The improved graph convolutional neural network includes a parallel multi-order graph convolutional layer and a multi-head attention mechanism; specifically, the number of parallel branches is first set to Q, and each branch corresponds to a normalized adjacency matrix of a different order; each branch receives an initial node feature matrix and a normalized adjacency matrix of a corresponding order in the first layer, and simultaneously performs graph convolution operation to generate single-order node features of the layer; in the subsequent lth layer, each branch receives the single-order node features output by the previous layer and the normalized adjacency matrix of the corresponding order and simultaneously performs graph convolution operation to generate single-order node features of the layer, after parallel convolution of each branch, the convolution node features of all branches are spliced according to the feature dimension to generate multi-order neighborhood features, realizing parallel update and layer-by-layer extraction of multi-order neighborhood features; the multi-order neighborhood features are input into the multi-head attention mechanism to obtain mine potential classification features;

[0106] The normalized adjacency matrix of different orders is specifically first-order neighborhood information based on the final weighted adjacency matrix for the first-order branch, the second-order branch is expanded to the second-order neighborhood, and so on to the Qth-order neighborhood;

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] wherein, the normalized adjacency matrix of the qth order, an adjacency matrix of the q-th order, a degree matrix of a single-order node feature of the q-th branch in the first layer, a single-order node feature of the q-th branch in the l+1-th layer, a weight matrix of the q-th branch in the first layer, a weight matrix of the q-th branch in the l+1-th layer, a single-order node feature of the q-th branch in the l-th layer, a single-order node feature of the first branch in the L-th layer, a single-order node feature of the second branch in the L-th layer, a single-order node feature of the Q-th branch in the L-th layer, a multi-order neighborhood feature of the L-th layer, , and represent the query, key and value projection matrices of the k-th head, respectively, , and represent the query, key and value matrices of the k-th head, respectively, represent the output feature of the h-th attention head, represent the query feature vector of node i on the h-th head, represent the key feature vector of neighbor node j on the h-th head, represent the feature dimension of a single attention head, represent the value feature vector of neighbor node j on the h-th head, represent the vector concatenation operation, represent the output feature of the first attention head, represent the output feature of the H-th attention head, H represents the total number of attention heads, represent the multi-head attention fusion weight matrix, represent the weight matrix of the fully connected layer, represent the bias parameter of the fully connected layer, represent the mining area potential classification feature, represent the predicted probability that the candidate mining area belongs to the class c, L represents the total number of convolutional layers;

[0116] The mining area potential evaluation model is trained, specifically, historical class mining area screening data is used as training data, and the model is trained, and finally a trained mining area potential evaluation model is obtained.

[0117] ​The metallogenic area regional screening is specifically screening the candidate metallogenic areas according to the potential evaluation results of the candidate metallogenic areas, and selecting the area with a high potential evaluation result as the final metallogenic area.

[0118] By performing the above operation, the technical problem that the candidate area characteristics cannot be fully utilized in the existing metallogenic area potential evaluation model, and the multi-order neighborhood information is difficult to be considered simultaneously in the feature extraction process, thereby resulting in insufficient classification accuracy of the metallogenic area potential, unclear category boundary, and low distinguishability of different potential categories, is solved. The scheme innovatively proposes to construct a weighted graph structure by using a K-nearest neighbor algorithm, calculate the node feature similarity by using a Gaussian kernel function, and establish an adjacency relationship to generate a graph structure that can reflect the spatial adjacency and feature similarity. An improved graph convolutional neural network with a parallel updating mechanism is introduced to realize the synchronous extraction and fusion of multi-order neighborhood features, and a multi-head attention mechanism is combined to enhance the feature expression capability. The integrity and distinguishability of the feature expression can be effectively improved, the accuracy and robustness of the metallogenic area potential recognition can be significantly improved, the separability of the high-potential and low-potential areas can be enhanced, the generalization ability of the model under a complex geological background can be improved, and high-precision automatic potential classification of a large range of candidate metallogenic areas can be realized.

[0119] Embodiment five, refer to Figure 1 , Figure 4 and Figure 5 This embodiment is based on the above-mentioned embodiments, and the metallogenic area intelligent exploration module specifically includes the following steps:

[0120] The metallogenic area yield prediction model is established, specifically including the following steps:

[0121] The metallogenic area yield prediction feature extraction is used to extract the fusion features reflecting the regional metallogenic area yield prediction. Specifically, the image data and the structured data are subjected to yield prediction image feature extraction and yield prediction structured feature extraction, respectively, to generate yield prediction image features and yield prediction structured features, and the two types of features are spliced and fused to obtain yield prediction comprehensive features.

[0122] The yield prediction image feature extraction is specifically to use a convolutional neural network to learn the spatial structure features of the high-resolution remote sensing image data in the metallogenic area exploration optimization data to obtain the yield prediction image features.

[0123] The yield prediction structured feature extraction is specifically to use a multi-layer perception network to extract deep features of the regional environmental field feature data and the element distribution feature data in the metallogenic area exploration optimization data to obtain the yield prediction structured features.

[0124] The output of the production forecast results is specifically obtained by performing nonlinear regression analysis on the comprehensive characteristics of the production forecast based on a feedforward neural network model to obtain the production forecast results for the mining area.

[0125] The training of the mining area production prediction model involves using historical mining area production prediction data as training data to train the prediction model, and finally obtaining the trained mining area production prediction model.

[0126] Prediction model performance optimization specifically involves obtaining the optimal hyperparameter combination of the model through an improved search algorithm, updating the prediction model hyperparameters based on the optimal hyperparameter combination, and obtaining the optimal mine production prediction model; including the following steps:

[0127] Initializing the particle swarm involves encoding the hyperparameters of the trained mine production prediction model into search individual position vectors, and generating N individual particle position vectors using a composite chaotic mapping initialization method. Each individual represents a candidate combination of prediction model hyperparameters, resulting in the initial particle swarm. The formula used is as follows:

[0128] ;

[0129] In the formula, Indicates the first The position of each individual particle. This represents the position of the i-th particle. Represents a random number in the range [0,1].

[0130] Calculating particle fitness values ​​specifically involves calculating the fitness values ​​of particles within the population. The performance of the mining area production prediction model based on the individual particle position is used as the particle fitness value.

[0131] The sinusoidal exponential inertia weight is calculated by adaptively adjusting the search intensity through the number of iterations, while introducing a sine term to achieve periodic fluctuations; the formula used is as follows:

[0132] ;

[0133] In the formula, Indicates the sinusoidal exponential inertia weight. This represents the maximum number of iterations, and t represents the current number of iterations. Represents a random number in the range [0,1].

[0134] Particle update specifically involves updating particle velocity and position; the formulas used are as follows:

[0135] ;

[0136] In the formula, It indicates that the i-th particle is in the... Speed ​​during iteration It indicates that the i-th particle is in the... Speed ​​during iteration This represents the position of the i-th particle in the t-th iteration. This represents the local optimal position of an individual particle. This represents the global optimal position of the particle. and Represents a random number in the range [0,1]. It indicates that the i-th particle is in the... Position during iteration and These represent individual learning factors and group learning factors, respectively.

[0137] The optimal position of a particle is updated by re-evaluating the fitness value of all updated particles and comparing it with the current global optimal position of the particle based on the current fitness value. If the current fitness value of the particle is better, the global optimal position of the particle is updated.

[0138] Particle search terminates when the particle fitness value is... When the fitness threshold is exceeded or the maximum number of iterations is reached, the search is terminated and the global optimal position of the particle is obtained; the global optimal position of the particle specifically refers to the optimal combination of hyperparameters of the model.

[0139] The prediction model hyperparameter update specifically involves adjusting the hyperparameters of the mining area production prediction model based on the optimal combination of hyperparameters in the model to obtain the optimal mining area production prediction model.

[0140] Mineral production prediction by zone involves dividing the mineralized area into multiple sub-zones according to an equal-area grid, inputting the real-time mineral production prediction data of each sub-zone into the optimal mineral production prediction model, and obtaining the mineral production prediction results of each sub-zone.

[0141] The determination of the mining exploration area involves summarizing and analyzing the mining output forecast results of each sub-area, and generating a complete mining area output distribution map based on spatial visualization technology. According to this distribution map, areas with different output levels are accurately located and prioritized to form a decision support scheme for mining area production management and resource allocation, thereby optimizing the mining plan and utilization efficiency of mineral resources and realizing intelligent exploration planning and efficient operation management of the target mining area.

[0142] By performing the above operation, in view of the technical problem that the existing mine yield prediction model is not reasonable for the hyperparameter setting, and in the hyperparameter optimization process, it depends on the fixed initialization method and the linear inertia weight adjustment, which leads to insufficient search space coverage and easy to fall into local optimum, resulting in insufficient model prediction accuracy, the scheme innovatively introduces a composite chaotic mapping initialization method in the particle swarm optimization algorithm, which is used to enhance the diversity of particle initial distribution and global search ability; The sine exponential inertia weight realizes the adaptive search intensity adjustment of dynamic fluctuation with iteration, balances global exploration and local development; Significantly improve the efficiency and accuracy of the search of the hyperparameters of the mine yield prediction model, enhance the adaptability of the model to complex geological and mineralization conditions, improve the accuracy and stability of the yield prediction, realize the high-precision yield prediction of each sub-area of the ore-forming area and the optimization of the mining scheme.

[0143] It should be noted that, in this paper, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed or inherent to such a process, method, article or apparatus.

[0144] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, alternatives and variations can be made thereto without departing from the principles and spirit of the application.

[0145] The above describes the present application and its embodiments, which is not restrictive, and the drawings shown are only one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which shall belong to the protection scope of the present application.

Claims

1. An artificial intelligence-based mineral exploration system, characterized in that: It includes a multi-source data acquisition module, a data optimization module, a potential identification module for candidate mining areas, and a smart mining area exploration module; The multi-source data acquisition module specifically obtains raw exploration data of the mining area through data acquisition operations; The data optimization module is used to improve the quality of the original data and the adaptability of the model. Specifically, it obtains optimized exploration data for the mining area through image data preprocessing and structured data preprocessing. The mining area candidate region potential identification module is used to identify high-potential mineralized areas in a large range of mining area candidate regions. Specifically, it establishes a mining area potential assessment model, trains the model using historical data, inputs real-time data into the trained model, outputs the potential assessment results of each candidate mining area, and filters the large range of candidate regions based on the potential assessment results to obtain the mineralized area results. The establishment of the mining area potential assessment model includes candidate area identification feature extraction, discriminant feature matrix acquisition, candidate area map structure establishment, and mining area potential result output. The output of the mining area potential results specifically involves inputting the candidate area map structure into an improved graph convolutional neural network with a parallel update mechanism, extracting the mining area potential classification features, mapping the mining area potential classification features through a fully connected layer, calculating the probability distribution of each potential category of the candidate mining area through the Softmax function, selecting the category corresponding to the highest probability as the final prediction result, and obtaining the potential evaluation result of the candidate mining area. The improved graph convolutional neural network includes parallel multi-order graph convolutional layers and a multi-head attention mechanism; Specifically, the number of parallel branches is first set to Q, and each branch corresponds to a normalized adjacency matrix of a different order. Each branch receives the initial node feature matrix and the normalized adjacency matrix of the corresponding order in the first layer, and performs graph convolution operation synchronously to generate the single-order node features of that layer. In the l-th layer, each branch receives the single-order node features output from the previous layer and the normalized adjacency matrix of the corresponding order, and performs graph convolution operations synchronously to generate the single-order node features of this layer. After the parallel convolution of each branch is completed, the convolution node features of all branches are concatenated according to the feature dimension to generate multi-order neighborhood features, realizing the parallel updating and layer-by-layer extraction of multi-order neighborhood features. The multi-order neighborhood features are input into the multi-head attention mechanism to obtain the mining area potential classification features. The intelligent exploration module for mining areas is used to implement zoned production prediction and precise mining area determination in ore-forming regions. Specifically, it establishes a mining area production prediction model and trains it using historical data. It employs a composite chaotic mapping initialization method and a sinusoidal exponential inertial weighted particle swarm optimization algorithm to search for the optimal hyperparameter combination of the model. The prediction model is then updated based on the optimal hyperparameter combination to obtain the optimal mining area production prediction model. Real-time data from each sub-region is then input into the prediction model, and the production prediction results for each sub-region are output. Based on the production prediction results, a mining area zoning production distribution map is generated, realizing intelligent exploration planning and efficient operation management of the target mining area.

2. The artificial intelligence-based mining exploration system according to claim 1, characterized in that: The potential identification module for candidate mining areas specifically includes the following steps: Establishing a mining area potential assessment model includes the following steps: The candidate area identification feature extraction specifically involves extracting image features and extracting structured features from image data and structured data respectively, generating candidate image features and candidate structured features for the mining area, and then splicing and fusing the two types of features to obtain comprehensive candidate features for the mining area. The discriminant feature matrix is ​​obtained by mapping the candidate comprehensive features of mining areas to a low-dimensional discriminant subspace, thereby enhancing the class separability between candidate mining areas. Specifically, the candidate comprehensive features of mining areas are input together with the corresponding historical mining area type labels to construct a nonlinear kernel mapping relationship between samples. The candidate comprehensive features of mining areas are then subjected to nonlinear discriminant compression processing using the kernel Fisher discriminant analysis method. The intra-class and inter-class divergences are calculated, and the generalized eigenvalue problem is solved to construct the projection matrix. Finally, a set of candidate low-dimensional discriminant features is obtained, resulting in the candidate mining discriminant feature matrix. Candidate region map structure established; Output of mining area potential results; The training of the mining area potential assessment model involves using historical mining area screening data as training data to train the model, and finally obtaining the trained mining area potential assessment model. The metallogenic region selection process involves inputting real-time mineral area screening data into a trained mineral area potential assessment model, outputting the potential assessment results of each candidate mineral area, filtering based on the potential assessment results of each candidate mineral area, and selecting areas with high potential mineral areas as the final metallogenic regions.

3. The artificial intelligence-based mining exploration system according to claim 1, characterized in that: The establishment of the candidate region map structure specifically includes the following steps: To construct a two-dimensional feature matrix, the features in each dimension of the candidate ore discrimination feature matrix are linearly normalized, and the features are sorted in descending order based on their importance. The sorted feature sequence is then mapped column by column to obtain the two-dimensional feature matrix. The importance of the features is specifically determined by performing a comprehensive analysis of the inter-class and intra-class dispersion of the j-th dimension of the candidate ore discrimination feature matrix Y, and calculating its single-dimensional Fisher discrimination score. The node set construction involves sliding and cropping the two-dimensional feature matrix according to the set window size and sliding step size, extracting local feature blocks in sequence, and vectorizing each feature block to form a node feature vector. All node feature vectors are then used to construct a node set. Edge connection establishment involves calculating the Euclidean distance between the feature vectors of any two nodes in the node set, and mapping the Euclidean distance to similarity weights using a Gaussian kernel function. For each node, the k nodes with the highest similarity are selected to form its neighbor set. Weighted edges are established based on the similarity weights, and the similarity weights are assigned to the elements of the adjacency matrix to form an initial weighted adjacency matrix. Subsequently, the initial weighted adjacency matrix is ​​symmetricized and self-loop edges are added to obtain the final weighted adjacency matrix. The graph structure output specifically involves first calculating the center coordinates of each node in the two-dimensional feature matrix, and then normalizing them to... After the interval, the spatial location code of the node is generated by linear mapping. Then, the spatial location code and the node feature are concatenated and fused to obtain the node fusion feature vector. All the node fusion feature vectors are arranged in order to form the initial node feature matrix. Finally, the node set, edge set, final weighted adjacency matrix and initial node feature matrix are combined to obtain the candidate region map structure.

4. The artificial intelligence-based mining exploration system according to claim 1, characterized in that: The intelligent exploration module for the mining area specifically includes the following steps: Establishing a mining area production prediction model includes the following steps: The extraction of mining area production prediction features involves extracting production prediction image features and production prediction structured features from image data and structured data respectively, generating production prediction image features and production prediction structured features, and then splicing and fusing the two types of features to obtain comprehensive production prediction features. The output of the production forecast results is specifically obtained by performing nonlinear regression analysis on the comprehensive characteristics of the production forecast based on a feedforward neural network model to obtain the production forecast results for the mining area. The training of the mining area production prediction model involves using historical mining area production prediction data as training data to train the prediction model, and finally obtaining the trained mining area production prediction model. Predictive model performance optimization; Mineral production prediction by zone involves dividing the mineralized area into multiple sub-zones according to an equal-area grid, inputting the real-time mineral production prediction data of each sub-zone into the optimal mineral production prediction model, and obtaining the mineral production prediction results of each sub-zone. The determination of the mining exploration area involves summarizing and analyzing the mining output forecast results of each sub-area, and generating a complete mining area output distribution map based on spatial visualization technology. According to this distribution map, areas with different output levels are accurately located and prioritized to form a decision support scheme for mining area production management and resource allocation, thereby optimizing the mining plan and utilization efficiency of mineral resources and realizing intelligent exploration planning and efficient operation management of the target mining area.

5. The artificial intelligence-based mining exploration system according to claim 4, characterized in that: The performance optimization of the prediction model specifically includes the following steps: Initializing the particle swarm involves encoding the hyperparameters of the trained mine production prediction model into search individual position vectors, and generating N individual particle position vectors using a composite chaotic mapping initialization method. Each individual represents a candidate combination of prediction model hyperparameters, resulting in the initial particle swarm. The formula used is as follows: ； In the formula, Indicates the first The position of each individual particle. This represents the position of the i-th particle. Represents a random number in the range [0,1]. Calculating particle fitness values ​​specifically involves calculating the fitness values ​​of particles within the population. The performance of the mining area production prediction model based on the individual particle position is used as the particle fitness value. The sinusoidal exponential inertia weight is calculated by adaptively adjusting the search intensity through the number of iterations, while introducing a sine term to achieve periodic fluctuations; the formula used is as follows: ； In the formula, Indicates the sinusoidal exponential inertia weight. This represents the maximum number of iterations, and t represents the current number of iterations. Represents a random number in the range [0,1]. Particle update specifically involves updating particle velocity using a sinusoidal exponential inertia weight, and then updating particle position based on the updated particle velocity. The optimal position of a particle is updated by re-evaluating the fitness value of all updated particles and comparing it with the current global optimal position of the particle based on the current fitness value. If the current fitness value of the particle is better, the global optimal position of the particle is updated. Particle search terminates when the particle fitness value is... When the fitness threshold is exceeded or the maximum number of iterations is reached, the search is terminated and the global optimal position of the particle is obtained; the global optimal position of the particle specifically refers to the optimal combination of hyperparameters of the model. The prediction model hyperparameter update specifically involves adjusting the hyperparameters of the mining area production prediction model based on the optimal combination of hyperparameters to obtain the optimal mining area production prediction model.

6. The artificial intelligence-based mining exploration system according to claim 1, characterized in that: The multi-source data acquisition module specifically obtains raw exploration data for the mining area by acquiring data related to the mining area from an external data platform. This raw exploration data includes historical mining area screening data, historical metallogenic area production prediction data, real-time mining area screening data, and real-time metallogenic area production prediction data. Both the historical and real-time mining area screening data include medium-resolution remote sensing image data and regional structured data. The historical mining area screening data also includes historical mining area types. Both the historical and real-time metallogenic area production prediction data include high-resolution remote sensing image data, regional environmental field characteristic data, and elemental distribution characteristic data. The historical metallogenic area production prediction data also includes mining area production data.

7. The artificial intelligence-based mining exploration system according to claim 1, characterized in that: The data optimization module specifically includes the following steps: Image data preprocessing specifically includes image cloud removal, radiometric correction, image format standardization, and image quality enhancement to obtain optimized remote sensing image data; Structured data preprocessing specifically includes data cleaning, data standardization, and data encoding to obtain structured and optimized data.

Images