Mineral resource prediction method and device based on deep learning, and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a deep learning-based mineral resource prediction method and device, and a storage medium. The method comprises the steps of performing data correction and normalization processing on multi-source geological data to obtain a standardized data set, extracting features by using a deep learning model and fusing the features, constructing a metallogenic knowledge graph to obtain knowledge vectors, constructing a prediction model based on a Transform network and a Bayesian framework to obtain a prediction result, and obtaining a prediction result. And after space optimization, constructing a three-dimensional visual model by using the depth generation model, and performing interpretive analysis to obtain a prediction verification result. According to the method, multi-source geological data can be effectively fused, multiple prediction tasks can be completed at the same time, and the prediction result is made to conform to the geological law and have high accuracy.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a mineral prediction method, device and storage medium based on deep learning. Background Art

[0002] Existing mineral prediction methods mainly include traditional methods based on geostatistics and prediction methods based on machine learning. Traditional geostatistical methods use Kriging interpolation, multi-point geostatistics and other methods to perform spatial analysis on mineralization prediction, but such methods are overly dependent on expert experience and prior knowledge. Machine learning methods such as support vector machines and random forest algorithms have been introduced into the field of mineral prediction. They predict mineralization potential by establishing a mapping relationship between geological features and mineralization. These methods can process multi-source geological data and automatically learn feature representation.

[0003] However, the existing technology has the following problems: First, the current methods often process different types of geological data (such as remote sensing images, geophysical and geochemical data) separately, and lack an effective multi-source data fusion mechanism; second, traditional machine learning methods have limited ability to extract data features, and it is difficult to capture complex geological characteristics and mineralization laws; third, existing prediction models usually only focus on a single prediction task (such as mineralization type prediction) and cannot handle multiple related prediction tasks at the same time; fourth, the prediction results lack interpretability, making it difficult to provide geologists with intuitive prediction basis; finally, the constraints of geological knowledge are not fully considered in the prediction process, resulting in the prediction results may violate basic geological laws. Summary of the invention

[0004] The present application provides a deep learning-based mineral prediction method, device and storage medium for effectively integrating multi-source geological data, completing multiple prediction tasks simultaneously, and making the prediction results conform to geological laws and have high accuracy.

[0005] In the first aspect, the present application provides a mineral prediction method based on deep learning, and the mineral prediction method based on deep learning includes: preprocessing remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain a standardized multi-source data set; based on the standardized multi-source data set, constructing a feature extraction model through an improved ResNet structure, a three-dimensional convolutional neural network and a graph neural network, and using a variational autoencoder to perform feature fusion to obtain deep coding features; based on the deep coding features, constructing a mineralization knowledge graph through a multi-head self-attention mechanism and a message passing mechanism, and obtaining a knowledge vector using a graph embedding algorithm; based on the deep coding features and the knowledge vector, constructing a multi-task prediction model through a Transformer network, and performing uncertainty assessment through a Bayesian neural network framework to obtain a mineralization prediction result; based on the mineralization prediction result, iteratively optimizing through spatial clustering and Bayesian optimization algorithms to obtain an optimized prediction result; based on the optimized prediction result, constructing a three-dimensional visualization model using a deep generative model, and performing explanatory analysis through a Grad-CAM algorithm to obtain a target prediction verification result.

[0006] In a second aspect, the present application provides a mineral deposit prediction device based on deep learning, and the mineral deposit prediction device based on deep learning comprises: The processing module is used to pre-process remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain standardized multi-source data sets; A fusion module is used to construct a feature extraction model based on the standardized multi-source data set through an improved ResNet structure, a three-dimensional convolutional neural network, and a graph neural network, and to perform feature fusion using a variational autoencoder to obtain deep coding features; A construction module is used to construct a mineralization knowledge graph based on the deep coding features through a multi-head self-attention mechanism and a message passing mechanism, and obtain a knowledge vector using a graph embedding algorithm; An evaluation module, used to construct a multi-task prediction model through a Transformer network according to the deep coding features and the knowledge vector, and to perform uncertainty evaluation through a Bayesian neural network framework to obtain a mineralization prediction result; An optimization module, used to perform iterative optimization through spatial clustering and Bayesian optimization algorithm according to the mineralization prediction result to obtain an optimized prediction result; The analysis module is used to construct a three-dimensional visualization model based on the optimized prediction results using a deep generative model, and perform explanatory analysis through a Grad-CAM algorithm to obtain a target prediction verification result.

[0007] The third aspect of the present application provides a computer-readable storage medium, which stores instructions, and when the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned deep learning-based mineral prediction method.

[0008] In the technical solution provided by this application, multi-source geological data is preprocessed through data correction and normalization algorithms to ensure the standardization and consistency of different types of data and improve the accuracy of subsequent feature extraction. The feature extraction model constructed by using improved ResNet structure, three-dimensional convolutional neural network and graph neural network can perform deep feature extraction for the characteristics of remote sensing images, geophysical exploration and geochemical data respectively, while the feature fusion mechanism of variational autoencoder realizes the effective integration of multi-source data. The multi-head self-attention mechanism and message passing mechanism are used to construct the mineralization knowledge map, which fully explores the correlation between geological elements, and the knowledge vector obtained by the graph embedding algorithm effectively encodes the complex geological relationship. The multi-task prediction model constructed based on the Transformer network realizes the joint prediction of mineralization type, scale and grade, and the introduction of the Bayesian neural network framework provides uncertainty assessment of the prediction results. The iterative optimization process of spatial clustering and Bayesian optimization algorithm improves the spatial continuity and geological rationality of the prediction results, making the prediction results more in line with the actual geological laws. Finally, a three-dimensional visualization model was constructed through a deep generative model and the Grad-CAM algorithm was used for explanatory analysis, which not only provided an intuitive display of the prediction results, but also revealed the key basis in the prediction process, enhanced the interpretability and credibility of the prediction results, and improved the accuracy, reliability and practicality of mineral prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1 A schematic diagram of an embodiment of a mineral deposit prediction method based on deep learning in an embodiment of the present application; Figure 2 A schematic diagram of a node-edge relationship structure in an embodiment of the present application; Figure 3 A schematic diagram of the mineralization knowledge map in the embodiment of this application; Figure 4 This is a schematic diagram of an embodiment of a mineral prediction device based on deep learning in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The embodiments of the present application provide a method, device and storage medium for mineral prediction based on deep learning. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the mineral deposit prediction method based on deep learning includes: Step S101, preprocessing remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain a standardized multi-source data set; Step S102: Based on the standardized multi-source data set, a feature extraction model is constructed by improving the ResNet structure, the three-dimensional convolutional neural network and the graph neural network, and a variational autoencoder is used to perform feature fusion to obtain deep coding features; Step S103: Based on the deep coding features, a mineralization knowledge graph is constructed through a multi-head self-attention mechanism and a message passing mechanism, and a knowledge vector is obtained using a graph embedding algorithm; Step S104: construct a multi-task prediction model through a Transformer network based on the deep coding features and knowledge vectors, and perform uncertainty assessment through a Bayesian neural network framework to obtain a mineralization prediction result; Step S105, performing iterative optimization through spatial clustering and Bayesian optimization algorithm according to the mineralization prediction result to obtain an optimized prediction result; Step S106: Based on the optimized prediction results, a three-dimensional visualization model is constructed using a deep generative model, and an explanatory analysis is performed using the Grad-CAM algorithm to obtain a target prediction verification result.

[0013] It is understandable that the execution subject of the present application may be a mineral prediction device based on deep learning, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0014] Specifically, it starts with multi-source data preprocessing. For remote sensing image data, the atmospheric correction algorithm is used to remove the influence of atmospheric scattering and absorption. In the specific processing, the original remote sensing data is calibrated for radiation and the atmospheric optical thickness is calculated. The surface reflectance data and atmospheric parameters are jointly inverted to eliminate the interference of the atmosphere on the spectral characteristics of the ground objects. Then geometric correction is performed to establish the correspondence between ground control points and image points. The terrain is corrected through the digital elevation model, and the image data is unified into the standard coordinate system. The processing of geophysical exploration data includes noise elimination of gravity data and magnetic data. The median filtering method is used to remove random noise and mutation signals, and then the data is mapped to the [0,1] interval through the minimum-maximum normalization method. For geochemical data, the box plot method is used to identify outliers, the interquartile range is calculated, the data points outside the normal range are marked as abnormal, and the missing and abnormal data are filled by the interpolation method.

[0015] In the feature extraction and fusion stage, the standardized remote sensing image data is input into the improved ResNet structure, and the spatial features are extracted through the convolution layer and residual connection. In the specific processing, each convolution layer performs a sliding window operation on the input data, extracts the local feature map, and maintains the original feature information through the residual connection. The geophysical exploration data is processed by a three-dimensional convolutional neural network, and the three-dimensional data volume is hierarchically extracted to capture the geophysical anomaly features at different depth levels. The geochemical data is constructed into a node-edge relationship structure, and information is aggregated through a graph neural network. The nodes represent the element content information, and the edges represent the correlation between the elements. All extracted features are fused through a variational autoencoder. The encoder part compresses the multi-source features into the latent space, and the decoder part reconstructs the features to ensure information retention. In the knowledge graph construction stage, multi-head self-attention calculations are performed on the deep encoding features, the feature space is divided into multiple subspaces for parallel processing, and the association weights between different features are calculated. The feature information is transmitted and updated between nodes through a message passing mechanism to construct a mineralization knowledge graph, in which nodes represent geological elements and edges represent the relationship between elements. The graph embedding algorithm generates node sequences through random walk sampling and converts graph structure data into vector representation.

[0016] In the multi-task prediction model construction phase, the deep coding features and knowledge vectors are spliced ​​and input into the Transformer network, and the sequence features are captured through position encoding and self-attention mechanisms. The model contains a shared coding layer and dedicated prediction branches for mineralization type, scale, and grade. The Bayesian neural network framework estimates the uncertainty of the prediction results through sampling, generating a prediction distribution rather than a single value. In the iterative optimization phase, high-probability prediction areas are identified through spatial clustering, and the spatial distance matrix between regions is calculated for grouping. The Bayesian optimization algorithm optimizes the prediction parameters through sampling point selection and objective function evaluation, and continuously updates and supplements the exploration positions.

[0017] In the visualization verification phase, the optimized prediction results are 3D reconstructed and voxelized, and a 3D visualization representation is constructed through a deep generative model. The Grad-CAM algorithm performs an interpretive analysis on the model, calculates the gradient and activation map of the feature layer, and identifies the areas that play a decisive role in the prediction results.

[0018] For example, the surface reflectivity of remote sensing image data collected in a certain study area increased by 15% after removing the atmospheric influence, and the signal-to-noise ratio of gravity anomaly data increased from the original 3:1 to 8:1 after filtering. The fused depth features capture the spatial correlation between geological elements such as faults, rock masses, and alteration zones. The constructed knowledge graph contains more than 500 geological element nodes and more than 1,000 relationship edges. The multi-task prediction model simultaneously outputs the prediction results of mineralization type, scale and grade, and the prediction uncertainty interval estimated by the Bayesian framework is ±10%. Three high-potential target areas were identified during the iterative optimization process. After additional exploration and verification, the prediction results were 85% consistent with the actual situation. The three-dimensional visualization model clearly shows the spatial distribution of mineralization prediction. The Grad-CAM analysis results show that the fault intersection zone and alteration halo contribute the most to the prediction results.

[0019] In the embodiment of the present application, multi-source geological data is preprocessed by data correction and normalization algorithms to ensure the standardization and consistency of different types of data and improve the accuracy of subsequent feature extraction. The feature extraction model constructed by using improved ResNet structure, three-dimensional convolutional neural network and graph neural network can perform deep feature extraction for the characteristics of remote sensing images, geophysical exploration and geochemical data respectively, and the feature fusion mechanism of variational autoencoder realizes the effective integration of multi-source data. The multi-head self-attention mechanism and message passing mechanism are used to construct the mineralization knowledge map, which fully explores the correlation between geological elements, and the knowledge vector obtained by the graph embedding algorithm effectively encodes the complex geological relationship. The multi-task prediction model constructed based on the Transformer network realizes the joint prediction of mineralization type, scale and grade, and the introduction of the Bayesian neural network framework provides uncertainty assessment of the prediction results. The iterative optimization process of spatial clustering and Bayesian optimization algorithm improves the spatial continuity and geological rationality of the prediction results, making the prediction results more in line with the actual geological laws. Finally, a three-dimensional visualization model was constructed through a deep generative model and the Grad-CAM algorithm was used for explanatory analysis, which not only provided an intuitive display of the prediction results, but also revealed the key basis in the prediction process, enhanced the interpretability and credibility of the prediction results, and improved the accuracy, reliability and practicality of mineral prediction.

[0020] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Through atmospheric correction, atmospheric scattering removal and radiation correction are performed on remote sensing image data to obtain preliminary corrected remote sensing image data; (2) Based on the preliminary corrected remote sensing image data, geometric distortion correction and geographic coordinate registration are performed through the digital elevation model to obtain corrected remote sensing image data; (3) The geophysical exploration data are divided into gravity data and magnetic data according to the data type, and the noise of the gravity data and magnetic data are eliminated by median filtering to obtain the de-noised geophysical exploration data; (4) Based on the de-noised geophysical data, numerical normalization is performed using the minimum-maximum method to obtain normalized geophysical data; (5) Based on the element content data in the geochemical data, the outliers are identified using the box plot method and the data are filled in using the interpolation method to obtain the processed geochemical data; (6) The corrected remote sensing image data, normalized geophysical data, and processed geochemical data are fused to obtain a standardized multi-source data set.

[0021] Specifically, atmospheric scattering mainly includes two types: Rayleigh scattering and Mie scattering. Rayleigh scattering is caused by atmospheric molecules, and the scattering intensity is inversely proportional to the fourth power of the wavelength, mainly affecting the short-wave band; Mie scattering is caused by aerosols, and the scattering intensity is approximately inversely proportional to the wavelength. The atmospheric correction process calculates the relationship between the radiance of the top atmosphere and the surface reflectance by establishing the radiation transfer equation. Radiation correction converts the image grayscale value into physical radiation, including sensor calibration and solar irradiance correction. Sensor calibration converts the digital number (DN value) into a radiance value, and the solar irradiance correction considers the influence of the solar zenith angle and the distance between the sun and the earth. The digital elevation model (DEM) is used for geometric distortion correction, mainly to correct the image deformation caused by terrain undulations. During processing, a ground control point network is established, and each control point contains image coordinates and field coordinate information. The terrain elevation of each pixel is calculated using DEM data, and the corresponding relationship between pixels in image space and geographic space is established in combination with the sensor imaging geometric model. The corrected image is reprojected into the standard geographic coordinate system through the resampling method, and the grayscale value of the resampled pixel is determined by bilinear interpolation or cubic convolution.

[0022] The processing of geophysical exploration data first divides the data into gravity data and magnetic data by type. Gravity data reflects the underground density distribution, and magnetic data reflects the distribution of magnetic minerals. The two types of data are processed by median filtering. Median filtering is a nonlinear filtering method. For each data point, the middle value of all points in its neighborhood after sorting by size is taken as the output. This method effectively removes impulse noise and can better maintain the edge characteristics of the data. For one-dimensional time series data, the sliding window size is set to an odd number to ensure the uniqueness of the median. Data normalization uses the minimum-maximum method to map data of different dimensions to a unified interval. For each data point x, calculate its normalized value y: y = (x-min) / (max-min), where min and max are the minimum and maximum values ​​of the data series, respectively. This processing method maintains the relative distribution characteristics of the data and facilitates comprehensive analysis of data of different physical quantities.

[0023] The outlier identification of geochemical data uses the box plot method to calculate the quartiles Q1 (25% position) and Q3 (75% position) of the data, and the interquartile range IQR=Q3-Q1. Data points outside the interval [Q1-1.5IQR, Q3+1.5IQR] are marked as outliers. The marked outliers and missing values ​​are filled using the distance-weighted inverse distance interpolation method. When interpolating, valid data points with similar spatial positions are considered, and the weight is inversely proportional to the distance.

[0024] In the data fusion stage, the three types of processed data are integrated to unify the spatial resolution and coordinate system of the data. A regular grid is established to resample the various types of data to the same grid. The weighted average method is used for data synthesis, and the weight coefficient is determined according to the reliability of each type of data and its contribution to mineralization prediction.

[0025] For example, during the atmospheric correction process of remote sensing data in a certain area, the atmospheric optical thickness is calculated through the radiation transfer equation, and the DN value is converted into surface reflectivity by combining the sensor calibration parameters. During terrain correction, a control point network is established using DEM data with a resolution of 20 meters for geometric correction. There is random noise in the gravity data. After 5-point median filtering, the noise is significantly suppressed and the geological anomaly features are clearer. The magnetic data is normalized and the numerical range is unified to the [0,1] interval. The abnormal points in the geochemical data where the copper content exceeds the normal range are reasonably estimated through spatial interpolation. The standardized multi-source data set obtained by fusion has good spatial consistency, and various types of geological information are fully integrated.

[0026] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Input the remote sensing image data in the standardized multi-source data set into the improved ResNet structure, extract the spatial features through the convolution layer, and transfer the gradient through the residual connection to obtain the surface feature data; (2) Reconstruct the geophysical exploration data in the standardized multi-source data set into three-dimensional data, extract hierarchical features through three-dimensional convolutional neural network, and obtain geophysical feature data; (3) The geochemical data in the standardized multi-source dataset is constructed into a node-edge relationship structure, and the information of the nodes is aggregated through the graph neural network to obtain the chemical characteristic data; (4) Construct a joint feature matrix based on surface feature data, geophysical feature data, and chemical feature data, and use attention weights to fuse different features to obtain fused feature data; (5) Input the fused feature data into the encoder part of the variational autoencoder, perform dimensionality reduction compression through a multi-layer perceptron, and obtain a potential feature vector; (6) Perform reconstruction error calculation and distribution constraints on the latent feature vector, and re-encode the features through nonlinear mapping to obtain deep encoding features.

[0027] Specifically, the improved ResNet structure extracts surface features through repeated convolution operations and residual connections. The standardized remote sensing image data is input to the first convolution layer of the improved ResNet structure. The convolution kernel size is set to 3×3 and the step size is 1. The input data is subjected to a sliding window operation to extract local features. The output of the convolution layer is processed by batch normalization and ReLU activation function. The residual connection directly passes the input data to the next layer and performs element-wise addition with the output of the convolution layer to avoid the gradient vanishing problem when the network is deepened. The stacking of multiple convolution blocks extracts surface features from low-level textures to high-level semantics layer by layer, and obtains surface feature data containing information such as geological structure and lithology distribution. In the three-dimensional reconstruction process of geophysical exploration data, gravity and magnetic data are reorganized into a three-dimensional data volume according to spatial position information. x and y represent plane coordinates, and z represents depth information. The three-dimensional convolutional neural network uses a 3D convolution kernel to extract features simultaneously in three dimensions. The network structure contains multiple 3D convolution layers, each with a convolution kernel size of 3×3×3. It downsamples layer by layer in the depth direction to extract geophysical anomaly features at different depths. The pooling layer uses the maximum pooling operation to reduce the spatial dimension of the feature map and retain significant features. Through the combination of multiple layers of 3D convolution and pooling, geophysical feature data that characterizes the three-dimensional distribution characteristics of underground geological bodies are obtained.

[0028] The graph structure of geochemical data is constructed based on the spatial distribution of element content and the correlation between elements. Each sampling point is a node of the graph, and the node attributes contain the element content information of the point. The edges between nodes are determined based on the spatial distance and the correlation of element content, and connections are established between points with a distance less than a threshold and high element correlation. Figure 2 , which is a schematic diagram of the node-edge relationship structure in an embodiment of the present application, Figure 2 The circular nodes in the middle represent sampling points, and the copper (Cu) and gold (Au) content values ​​of each sampling point are displayed in the nodes; the lines between the nodes represent correlation connections, where w represents the connection weight, and the weight value ranges from 0 to 1. The larger the value, the stronger the correlation. This figure shows how to convert discrete geochemical sampling data into a graph structure with spatial correlation, providing the input data format for subsequent graph neural network feature extraction. The graph neural network aggregates the features of nodes through the message passing mechanism. Each node collects information from its neighboring nodes and updates its own state. After multiple rounds of message passing, the state of each node contains the geochemical characteristics of the local area, and the chemical feature data is aggregated. The construction of the joint feature matrix splices the surface feature data, geophysical feature data, and chemical feature data in the feature dimension. The attention mechanism calculates the weight relationship between different features, maps the features to the query, key, and value spaces, calculates the attention score through the dot product of the query and the key, and obtains the weight coefficient after normalization by the softmax function. The weighted sum with the value vector obtains the attention output. This mechanism dynamically adjusts the importance of different types of features to achieve adaptive fusion of features.

[0029] The encoder part of the variational autoencoder is composed of a multi-layer perceptron, which performs dimensionality reduction compression on the fused feature data. The encoder maps the input features to the mean vector and variance vector through multiple fully connected layers, and samples the latent feature vector. During the training process, the KL divergence loss term is introduced to constrain the distribution of the latent space to be close to the standard normal distribution, while minimizing the reconstruction error. In the feature recoding stage, the latent feature vector is reconstructed and the features are converted to a new representation space through nonlinear mapping. The reconstruction process uses an inverse transposed convolution layer to gradually restore the spatial structure of the features. At the same time, the error between the reconstructed result and the original feature is calculated, and combined with the KL divergence constraint term, the model is guided to learn more effective feature representations.

[0030] Taking the data processing of a mining area as an example, the multispectral remote sensing image covering an area of ​​20 square kilometers is input into the improved ResNet structure, and the surface features are extracted through 5 residual blocks, each of which contains two convolutional layers and a short-circuit connection. The gravity and magnetic data cover a depth of 500 meters. A 4-layer 3D convolutional network is used for feature extraction, and the size of the feature map is reduced by half layer by layer. The geochemical data contains 10 element content indicators. The graph structure is constructed according to the spatial distribution of the sampling points, and the edge weights between adjacent nodes are determined based on the correlation coefficient of the element content. When the three types of features are fused, the attention mechanism calculates the weights of the surface features, geophysical features, and chemical features to be 0.4, 0.35, and 0.25, respectively. The variational autoencoder compresses the 2048-dimensional fused features into a 128-dimensional latent space, and the reconstruction error converges to a stable value on the validation set, indicating that the encoded features retain the key information of the original data.

[0031] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) The depth coding features are divided into fault features, rock mass features and alteration zone features according to the geological element type, and the initial geological element data are obtained through feature matrix reconstruction; (2) Perform multi-head mapping decomposition on the initial geological element data, calculate the attention weights between elements through the multi-head self-attention mechanism, and obtain the correlation feature data; (3) Establish a set of geological element nodes based on the associated feature data, transfer and update the features between nodes through the message passing mechanism, and obtain the node status data; (4) Convert node status data into an entity-relationship-attribute structure, represent knowledge through semantic coding, and obtain a mineralization knowledge graph; (5) Perform random walk sampling on the nodes in the metallogenic knowledge graph, generate node sequences through the graph embedding algorithm, and obtain graph sequence data; (6) The graph sequence data is vectorized and dimensionally unified, and the knowledge vector is obtained through nonlinear transformation.

[0032] Specifically, the classification processing of deep coding features is divided according to the characteristic attributes of geological elements, and the feature vectors are separated according to the characteristic expressions of different geological element types. Fault features mainly contain linear structural information, which is manifested as linear textures and geophysical anomalies in remote sensing images; rock mass features contain lithology combinations and spatial distribution information; and alteration zone features reflect mineral alteration information. Feature sub-matrices are constructed for these three types of features, and the initial geological element data are reconstructed through matrix splicing. In the multi-head mapping decomposition process, the initial geological element data is mapped to multiple subspaces. The specific operation is to set multiple feature mapping heads, each of which independently performs linear transformation on the input features to generate query vectors, key vectors, and value vectors. The multi-head self-attention mechanism calculates the attention weights in each subspace, and each geological element calculates the similarity score with other elements, and the attention coefficient is obtained after softmax normalization. These coefficients represent the strength of association between different geological elements, and then the correlation feature data is obtained.

[0033] The establishment of node sets is based on associative feature data. Each geological element is regarded as a node, and the node attributes contain the feature information of the element. The message passing mechanism transmits information between nodes. In each round of transmission, the node collects information from its neighbors and updates its own status. The association weights between nodes are considered during information transmission. The neighbor nodes with larger weights contribute more information. After multiple rounds of transmission, node status data containing local structural information is obtained. The construction of the entity-relationship-attribute structure converts the node status data into a knowledge graph representation. Entities correspond to geological element nodes, relationships represent the interactions between entities, and attributes describe the characteristics of entities. Semantic coding maps entities, relationships, and attributes to a unified semantic space to construct a mineralization knowledge graph. Figure 3 As shown, it is a schematic diagram of the mineralization knowledge map in the embodiment of the present application. The circular nodes in the figure represent different types of geological element entities, including fault characteristics (main faults, secondary faults), rock mass characteristics (intrusive rock mass) and alteration zone characteristics (alteration center, alteration halo); the key attribute information of the entity is displayed in the node; the lines between the nodes represent the relationship between geological elements, such as "derivation", "cutting", "control of mineralization", "contact zone alteration" and "peripheral development". The figure shows the complex correlation between geological elements and constitutes a complete mineralization knowledge representation system. This structured representation captures the complex relationship network between geological elements.

[0034] Random walk sampling is performed on the mineralization knowledge graph. Starting from each node, random walks are performed according to the weight probability of the edge to generate node access sequences. These sequences retain the neighborhood structure information of the nodes in the graph. The graph embedding algorithm processes these sequences and maps each node to a vector representation of a fixed dimension. Similar nodes are closer in the embedding space. The vectorized mapping converts the graph sequence data into a numerical vector, and adjusts the vector dimension through nonlinear transformation to ensure that all knowledge vectors have a unified dimensional representation. This process retains the structural information and semantic relationships in the knowledge graph.

[0035] For example, by classifying the deep coding features, we can extract fault features (including regional main faults and secondary faults), rock mass features (including intrusive rocks and surrounding rocks), and alteration zone features (including alteration halos and alteration centers). In the multi-head self-attention calculation, it is found that there is a strong correlation between faults and alteration zones, which is reflected in the higher attention weight. In the process of node information transmission, fault nodes transmit structural information to the surrounding area, alteration zone nodes transmit mineral combination information, and rock mass nodes transmit lithology information. This information forms a directed relationship in the knowledge graph, such as "faults control alteration" and "alteration occurs in the rock contact zone". When random walk sampling, starting from the main fault node and walking along the high-weight edge, it often reaches the alteration center node, indicating that the two are closely related. The obtained knowledge vector encodes these complex geological relationships.

[0036] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Concatenate the deep coding features and knowledge vectors, perform sequence conversion through position encoding, and obtain input sequence data; (2) Decompose the input sequence data into tasks according to mineralization type, ore body size and grade prediction, and obtain multi-task input data through task identifier encoding; (3) Building a multi-task prediction model based on multi-task input data, where the multi-task prediction model includes a shared encoding layer, a task-specific decoding layer, and a prediction head layer. Feature transformation is performed through the multi-layer self-attention calculation of the Transformer network to obtain task feature data; (4) Input the task characteristic data into the mineralization type prediction branch, the ore body size prediction branch and the grade prediction branch respectively, and perform decoding processing through the task specific decoding layer to obtain branch prediction data; (5) Construct probability distribution parameters based on branch prediction data, perform sampling and uncertainty evaluation through the Bayesian neural network framework, and obtain prediction distribution data; (6) The predicted distribution data is subjected to multi-task weighted fusion, and the mineralization prediction results are obtained through confidence threshold screening and normalization processing.

[0037] Specifically, feature splicing is performed on the deep coding features and knowledge vectors. The splicing operation is performed on the feature dimension. The deep coding features contain geological feature information, and the knowledge vector contains geological element relationship information. The two are connected on the same dimension to form a joint feature vector. Position coding processes the position dependency in the sequence by adding position information to the feature vector. Sine and cosine functions are used to generate position coding vectors of different frequencies, which are added to the feature vector to obtain the input sequence data. The multi-task decomposition process targets three prediction targets: mineralization type (including different genetic types such as porphyry type and hydrothermal type), ore body scale (grading from small to super large) and grade prediction (numerical prediction of ore grade). Each task is assigned a unique task identifier, which is converted into a vector form through one-hot encoding. The task identifier is combined with the input sequence data to construct multi-task input data containing task information.

[0038] The core architecture of the multi-task prediction model consists of three layers: the shared encoding layer uses the Transformer structure to process the common features of all tasks, the task-specific decoding layer decodes features for different tasks, and the prediction head layer generates prediction results. The self-attention mechanism in the Transformer network captures long-distance dependencies by calculating the correlation between each position in the sequence and all positions. The stacking of multiple layers of self-attention allows the model to learn feature representations at different levels and obtain task feature data.

[0039] The processing process for the three prediction branches can be expressed as the following mathematical formula: 1. Decoding calculation of mineralization type prediction branch: , in: predict probability distributions for mineralization types; Predict weight matrix for mineralization type; It is a hidden layer state of mineralization characteristics; is the bias term; is the softmax activation function.

[0040] 2. Decoding calculation of ore body size prediction branch: , in: is the predicted value of ore body size; is the scale prediction transformation matrix; It is the scale feature representation; Bias for scale prediction; is the ReLU activation function.

[0041] 3. Decoding calculation of grade prediction branch: , in: To predict the grade results; is the grade prediction coefficient matrix; is the grade feature tensor; is the bias term for grade prediction; is a linear activation function.

[0042] The Bayesian neural network framework characterizes the uncertainty of prediction through probability distribution. Each prediction branch outputs a probability distribution rather than a single value. Monte Carlo sampling is used to extract multiple samples from the prediction distribution, and the mean is calculated as the prediction value, and the variance is used as the uncertainty measure. This method takes into account the uncertainty of model parameters and prediction results to generate prediction distribution data. Multi-task weighted fusion takes into account the results of the three prediction branches and assigns weights according to the importance of each task and the reliability of the prediction. By setting a confidence threshold to screen high-confidence predictions, the prediction values ​​are normalized to obtain the mineralization prediction results.

[0043] For example, the input data of a certain study area is processed, and the deep coding features include surface alteration, structural linearity and lithology distribution characteristics. The knowledge vector encodes the mineralization laws such as fault control and rock contact zone control. Through the analysis of the multi-task prediction model, the mineralization type prediction branch identifies the characteristics of porphyry copper mineralization, which is consistent with the known regional geological background; the ore body scale prediction branch estimates the ore body magnitude based on the scale of the ore-bearing rock body and the intensity of alteration, and predicts it to be medium-sized; the grade prediction branch predicts the grade distribution based on the intensity of alteration and the intensity of element anomalies. The Bayesian framework evaluation shows that the prediction at the intersection of faults has a higher confidence, while the prediction of the marginal area has a larger uncertainty. The three prediction results are integrated to identify the most favorable mineralization target area.

[0044] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Sort the mineralization prediction results according to the prediction probability value, and obtain the high probability prediction area data through probability threshold segmentation; (2) Construct a spatial distance matrix based on the high-probability prediction area data, group the areas by spatial clustering, and obtain potential target area data; (3) Extract boundaries and perform spatial overlay analysis on potential target area data, optimize through geological parameter constraints, and obtain preliminary optimized data; (4) Input the preliminary optimization data into the Bayesian optimization algorithm, optimize the parameters through sampling point selection and objective function evaluation, and obtain supplementary survey location data; (5) Update geological parameters based on supplementary exploration location data, optimize data through iterative calculation and parameter adjustment, and obtain iterative optimization data; (6) The iterative optimization data is comprehensively evaluated and weighted, and the optimized prediction results are obtained through multiple rounds of iterative calculations.

[0045] Specifically, in the process of optimizing the mineralization prediction results, the prediction probability values ​​are sorted from large to small, and a probability threshold (usually determined based on the statistical distribution of known mineral deposits) is selected for data segmentation. The sorted data is reorganized according to the spatial location information, and the prediction areas above the threshold are extracted to form high-probability prediction area data. These data contain the spatial coordinates and prediction probability values ​​of each prediction point. The construction of the spatial distance matrix is ​​based on the spatial coordinate information in the high-probability prediction area data, and the Euclidean distance between the prediction points is calculated to form a distance matrix. Spatial clustering uses the DBSCAN (density-based spatial clustering algorithm) method. By setting the two parameters of neighborhood radius and minimum number of points, spatially close high-probability prediction points are combined into potential target areas. This clustering method can identify target areas of any shape and automatically filter out noise points.

[0046] The boundary extraction process uses the convex hull algorithm to calculate the outer boundary of each target area, while taking into account the complexity of the internal structure. The spatial overlay analysis overlays the target area boundary with geological parameters (including lithology boundary, fault distribution, alteration zone range, etc.), and adjusts the target area boundary through geological parameter constraints. This process takes into account the integrity of the geological unit and the continuity of the geological boundary.

[0047] The Bayesian optimization algorithm further optimizes the preliminary optimization data, and its core formula is: , in: It is the best supplementary survey location; is the expected improvement function; is the probability improvement function; is the confidence upper bound function; is the weight coefficient of different acquisition functions; X is the feasible sampling space.

[0048] During the updating process of geological parameters, the supplementary exploration location data is integrated with the original geological parameters, and the regional geological parameter distribution is updated through spatial interpolation methods such as Kriging interpolation. During the iterative calculation process, the prediction results are re-evaluated after each round of update, and the model parameters are adjusted until the parameter change is less than the preset threshold. This process continuously optimizes the spatial distribution characteristics of geological parameters. The comprehensive evaluation stage considers multiple evaluation indicators, including target area prediction probability, geological rationality, economic feasibility, etc. By setting weight coefficients for different indicators, multiple evaluation indicators are combined, and optimized prediction results are obtained after multiple rounds of iterative calculations.

[0049] For example: Sort the mineralization prediction results by prediction probability, set 0.75 as the probability threshold, and extract the high probability prediction area. Through spatial clustering analysis, three potential target areas were identified at the intersection of faults, each of which contains multiple high probability prediction points. After boundary extraction, it was found that target area one and the known copper mineralization belt showed a belt-like distribution feature in space, target area two was located in the core of the anticline structure, and target area three extended along the fault. The target area boundary was optimized in combination with fault, lithology and alteration information to determine the best supplementary exploration position. Two drill holes were arranged in target area one. After obtaining new core data, the geological parameters were updated. It was found that the mineralization intensity of target area one showed regular changes with increasing depth. After multiple rounds of iterative optimization, the priority exploration target area with the greatest exploration potential was identified.

[0050] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Reconstruct the optimized prediction results according to the spatial coordinates and obtain voxelized prediction data through three-dimensional interpolation processing; (2) Input the voxelized prediction data into the deep generative model and obtain the three-dimensional feature representation data through multi-layer encoding and decoding processing; (3) Construct geometric topological relationships based on 3D feature representation data, optimize boundaries through spatial constraints, and obtain a 3D visualization model; (4) Decompose the three-dimensional visualization model hierarchically, calculate the feature activation map through the Grad-CAM algorithm, and obtain the prediction basis data; (5) Compare and analyze the prediction basis data with geological prior knowledge, conduct rationality assessment through rule verification, and obtain verification evaluation data; (6) Perform multi-dimensional comprehensive analysis and weight calculation on the verification and evaluation data, and obtain the target prediction verification results through confidence evaluation.

[0051] Specifically, the three-dimensional visualization processing of the optimized prediction results starts with spatial data reconstruction, converting the prediction results on the two-dimensional plane into a three-dimensional data structure according to the spatial coordinate information. In the specific process, a three-dimensional grid system is established, and the grid size is determined according to the prediction accuracy requirements. The discrete prediction points are spatially interpolated by the Kriging interpolation method to generate a continuous three-dimensional prediction field. The voxelization process discretizes the continuous field into regular voxel units, each of which contains the prediction value and spatial position information.

[0052] The deep generative model adopts a three-dimensional convolutional autoencoder structure. The encoder part compresses the voxelized prediction data into a low-dimensional feature space through a multi-layer three-dimensional convolutional network. Batch normalization and ReLU activation functions are used after each convolution. The decoder part gradually restores the spatial structure through the transposed convolution layer and recovers the three-dimensional feature representation. In the encoding and decoding process, the jump connection retains the feature information of different scales, and obtains the three-dimensional feature representation data containing multi-scale features. The construction of geometric topological relationships is based on the three-dimensional feature representation data. The isosurface is extracted to construct the initial geometric boundary, and then the connectivity relationship between voxels is established. The spatial constraints include geological constraints such as stratum occurrence and fault strike. The minimum surface energy method is used to optimize the boundary to ensure compliance with geological laws. The three-dimensional visualization model constructs a three-dimensional structure with geological significance on the basis of satisfying the topological relationship and spatial constraints.

[0053] The calculation process of the Grad-CAM algorithm slices the three-dimensional visualization model in layers to obtain two-dimensional feature maps of different depths. The class activation map is calculated for each layer of feature map. The specific steps include: calculating the gradient of the predicted category to the feature map, performing global average pooling on the gradient to obtain the weight of each channel, weighted superposition of the weight and the feature map, and obtaining a heat map after ReLU activation. This process reveals the key areas of concern of the model and generates prediction basis data. The comparative analysis of the prediction basis data and geological prior knowledge includes multiple aspects: rationality of spatial distribution (whether it conforms to the mineralization law), structural control relationship (whether it is consistent with the main mineralization-controlling structure), and lithology correlation (whether it corresponds to favorable mineralization lithology). By establishing an evaluation rule library, each rule is quantified into a scoring indicator, and the verification evaluation data is obtained through comprehensive calculation.

[0054] The multi-dimensional comprehensive analysis is evaluated from three dimensions: prediction accuracy, geological rationality and economic feasibility. Prediction accuracy is evaluated by cross-validation to evaluate the prediction ability of the model, geological rationality is judged based on geological laws, and economic feasibility considers resource endowment and mining conditions. Different dimensional indicators are combined in a weighted manner, and the weight coefficient is determined based on expert experience to obtain the target prediction verification result.

[0055] For example, the original prediction results include the predicted probability values ​​of the plane distribution. Through the reconstruction of three-dimensional data, a voxel model within a depth range of 500 meters is established, and the voxel size is 10 meters × 10 meters × 10 meters. The three-dimensional feature representation obtained after the depth generation model processing clearly depicts the spatial distribution characteristics of the ore body, showing the regularity of extension along the fault strike. Through the analysis of geometric topological relationships, the contact relationship between the ore body and the surrounding rock, as well as the spatial connectivity between different mineralized zones, are determined. Grad-CAM analysis shows that the model pays great attention to the fault intersection zone and the rock contact zone, which is consistent with the regional mineralization law. These characteristics are compared with the characteristics of known deposits to verify the rationality of the prediction results. Through multi-dimensional comprehensive analysis, the target area with the most exploration value is determined.

[0056] In a specific embodiment, the process of performing the step of hierarchical decomposition of the three-dimensional visualization model may specifically include the following steps: (1) Slice and decompose the 3D visualization model and obtain layered data through gridding; (2) Sort and group the hierarchical data according to the depth information, and obtain multi-level structured data through hierarchical index encoding; (3) Extract features from each level of the multi-level structure data and obtain feature weight data through target category gradient calculation; (4) Perform weighted superposition of feature weight data and feature maps of the corresponding level, and obtain heat map data through the Grad-CAM algorithm; (5) Perform threshold segmentation and regional connectivity analysis based on heat map data, and obtain key area data through spatial correlation calculation; (6) The key regional data are sorted by feature importance and regionally integrated, and the prediction basis data is obtained through spatial combination relationship analysis.

[0057] Specifically, the analysis and processing of the three-dimensional visualization model is sliced ​​and decomposed, and sliced ​​along the three main planes of XY, YZ, and XZ to form a series of two-dimensional sections. Each section is gridded according to a fixed grid spacing, and the attribute values ​​and spatial position information of the grid nodes are recorded. The gridding process takes into account the continuity of the geological body, and adopts an adaptive grid division method to encrypt the grid in areas where the geological characteristics change dramatically, so as to obtain hierarchical data reflecting the spatial distribution characteristics. The sorting and grouping of hierarchical data are organized based on depth information, and the slice data of different depths are arranged in order from shallow to deep. The hierarchical index encoding adopts a tree structure, where the parent node represents the larger scale feature and the child node represents the local detail feature. Each level is given a unique coding identifier, and an inheritance relationship between levels is established to form a multi-level structure data containing multiple hierarchical features.

[0058] During the feature extraction process, convolution operations are performed on the data at each level to extract feature maps of different scales. The target category gradient calculation reflects the sensitivity of the prediction result to the feature map, and the contribution of each feature map to the target prediction is calculated through back propagation. The global average of the gradient is calculated for each feature channel to obtain the feature weight data that represents the importance of the feature. The weighted superposition of the feature weight and the feature map uses a pixel-by-pixel multiplication operation to multiply the weight coefficient with the corresponding feature map. The Grad-CAM algorithm processes the weighted results through the ReLU activation function, highlights the areas of positive contribution, and generates heat map data that reflects the basis for the prediction. The high-value areas on the heat map represent the spatial locations that contribute more to the prediction results.

[0059] The threshold segmentation process uses the OTSU algorithm to adaptively determine the threshold and divide the thermal map data into high-contribution areas and low-contribution areas. The regional connectivity analysis uses the connected component labeling algorithm to identify spatially continuous high-contribution areas. By calculating the spatial distance and feature similarity between regions, the degree of association between regions is evaluated to obtain key regional data with spatial correlation. The feature importance ranking is based on the average thermal value and area size of the region to rank the importance of key regions. In the process of regional integration, spatial proximity and geological similarity are considered to merge spatially close and feature-similar regions. By analyzing the spatial combination relationship of the merged region, including spatial position relationship, geometric morphological characteristics and geological attribute correlation, the prediction basis data is obtained.

[0060] For example: slice the three-dimensional visualization model with a coverage depth of 500 meters, take a horizontal section every 10 meters in the vertical direction, and take vertical sections along the strike and dip directions. The layered data obtained after gridding clearly shows the spatial distribution characteristics of the mineralized belt. Hierarchical coding organizes the data into four levels, corresponding to geological features of different scales. Feature extraction found that deep fault zones and rock contact zones have higher feature weights. Thermal map analysis shows that the favorable areas for mineralization are mainly concentrated at the intersection of faults and lithology transition zones. Regional connectivity analysis identifies three main key areas, corresponding to different mineralization environments. Through spatial combination relationship analysis, it is found that these key areas are distributed in a belt-like manner in space, and are consistent with the direction of the regional tectonic line, providing a reliable spatial basis for mineralization prediction.

[0061] The above describes the mineral deposit prediction method based on deep learning in the embodiment of the present application. The following describes the mineral deposit prediction device based on deep learning in the embodiment of the present application. Figure 4 In the embodiments of the present application, an embodiment of a mineral deposit prediction device based on deep learning includes: The processing module 201 is used to pre-process the remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain a standardized multi-source data set; A fusion module 202 is used to construct a feature extraction model based on the standardized multi-source data set through an improved ResNet structure, a three-dimensional convolutional neural network and a graph neural network, and perform feature fusion using a variational autoencoder to obtain deep coding features; A construction module 203 is used to construct a mineralization knowledge graph based on the deep coding features through a multi-head self-attention mechanism and a message passing mechanism, and obtain a knowledge vector using a graph embedding algorithm; An evaluation module 204 is used to construct a multi-task prediction model through a Transformer network according to the deep coding features and the knowledge vector, and to perform uncertainty evaluation through a Bayesian neural network framework to obtain a mineralization prediction result; An optimization module 205 is used to perform iterative optimization based on the mineralization prediction result through spatial clustering and Bayesian optimization algorithm to obtain an optimized prediction result; The analysis module 206 is used to construct a three-dimensional visualization model based on the optimized prediction results using a deep generative model, and perform explanatory analysis through a Grad-CAM algorithm to obtain a target prediction verification result.

[0062] Through the collaboration of the above components, multi-source geological data are preprocessed through data correction and normalization algorithms, ensuring the standardization and consistency of different types of data and improving the accuracy of subsequent feature extraction. The feature extraction model constructed by improved ResNet structure, three-dimensional convolutional neural network and graph neural network can perform deep feature extraction according to the characteristics of remote sensing images, geophysical exploration and geochemical data, while the feature fusion mechanism of variational autoencoder realizes the effective integration of multi-source data. The multi-head self-attention mechanism and message passing mechanism are used to construct the mineralization knowledge graph, which fully explores the correlation between geological elements. The knowledge vector obtained by the graph embedding algorithm effectively encodes the complex geological relationship. The multi-task prediction model constructed based on the Transformer network realizes the joint prediction of mineralization type, scale and grade, and the introduction of the Bayesian neural network framework provides uncertainty assessment of the prediction results. The iterative optimization process of spatial clustering and Bayesian optimization algorithm improves the spatial continuity and geological rationality of the prediction results, making the prediction results more in line with the actual geological laws. Finally, a three-dimensional visualization model was constructed through a deep generative model and the Grad-CAM algorithm was used for explanatory analysis, which not only provided an intuitive display of the prediction results, but also revealed the key basis in the prediction process, enhanced the interpretability and credibility of the prediction results, and improved the accuracy, reliability and practicality of mineral prediction.

[0063] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the deep learning-based mineral prediction method.

[0064] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0065] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0066] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A mineral deposit prediction method based on deep learning, characterized in that: The mineral deposit prediction method based on deep learning includes: Preprocess remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain standardized multi-source data sets; Based on the standardized multi-source data set, a feature extraction model is constructed by improving the ResNet structure, a three-dimensional convolutional neural network and a graph neural network, and a variational autoencoder is used to perform feature fusion to obtain deep coding features; Based on the deep coding features, a mineralization knowledge graph is constructed through a multi-head self-attention mechanism and a message passing mechanism, and a knowledge vector is obtained using a graph embedding algorithm; Based on the deep coding features and the knowledge vector, a multi-task prediction model is constructed through a Transformer network, and uncertainty assessment is performed through a Bayesian neural network framework to obtain a mineralization prediction result; According to the mineralization prediction results, iterative optimization is performed through spatial clustering and Bayesian optimization algorithm to obtain optimized prediction results; Based on the optimized prediction results, a three-dimensional visualization model is constructed using a deep generative model, and an explanatory analysis is performed using the Grad-CAM algorithm to obtain target prediction verification results.

2. The mineral deposit prediction method based on deep learning according to claim 1, characterized in that: The remote sensing image data, geophysical exploration data and geochemical data are preprocessed by data correction and normalization algorithms to obtain a standardized multi-source data set, including: Performing atmospheric scattering removal and radiation correction on the remote sensing image data through atmospheric correction to obtain preliminary corrected remote sensing image data; According to the preliminary corrected remote sensing image data, geometric distortion correction and geographic coordinate registration are performed through a digital elevation model to obtain corrected remote sensing image data; The geophysical exploration data are divided into gravity data and magnetic data according to data types, and noise is eliminated on the gravity data and the magnetic data respectively by median filtering to obtain de-noised geophysical exploration data; According to the de-noised geophysical data, numerical normalization is performed by a minimum-maximum method to obtain normalized geophysical data; Based on the element content data in the geochemical data, outliers are identified using a box plot method and data are filled using an interpolation method to obtain processed geochemical data; The corrected remote sensing image data, the normalized geophysical data and the processed geochemical data are fused to obtain the standardized multi-source data set.

3. The mineral deposit prediction method based on deep learning according to claim 1, characterized in that: According to the standardized multi-source data set, a feature extraction model is constructed by improving the ResNet structure, a three-dimensional convolutional neural network and a graph neural network, and a variational autoencoder is used for feature fusion to obtain deep coding features, including: Inputting the remote sensing image data in the standardized multi-source data set into the improved ResNet structure, extracting spatial features through the convolution layer, and performing gradient transfer through the residual connection to obtain surface feature data; Reconstructing three-dimensional data of the geophysical exploration data in the standardized multi-source data set, and extracting hierarchical features through the three-dimensional convolutional neural network to obtain geophysical exploration feature data; The geochemical data in the standardized multi-source data set is constructed into a node-edge relationship structure, and information aggregation of the nodes is performed through the graph neural network to obtain chemical characteristic data; Building a joint feature matrix based on the surface feature data, the geophysical feature data and the chemical feature data, and fusing different features using attention weights to obtain fused feature data; Inputting the fused feature data into the encoder part of the variational autoencoder, performing dimensionality reduction compression through a multi-layer perceptron to obtain a potential feature vector; Reconstruction error calculation and distribution constraint are performed on the potential feature vector, and feature recoding is performed through nonlinear mapping to obtain the deep coding feature.

4. The mineral deposit prediction method based on deep learning according to claim 1, characterized in that: Based on the deep coding features, the mineralization knowledge graph is constructed through a multi-head self-attention mechanism and a message passing mechanism, and a knowledge vector is obtained by using a graph embedding algorithm, including: The depth coding features are divided into fault features, rock mass features and alteration zone features according to the geological element types, and initial geological element data are obtained through feature matrix reconstruction; Performing multi-head mapping decomposition on the initial geological element data, calculating the attention weights between elements through the multi-head self-attention mechanism, and obtaining correlation feature data; Establishing a geological element node set based on the correlation feature data, transmitting and updating features between nodes through the message transmission mechanism, and obtaining node status data; The node state data is converted into an entity-relationship-attribute structure, and knowledge is represented by semantic coding to obtain the metallogenic knowledge graph; Performing random walk sampling on the nodes in the metallogenic knowledge graph, generating a node sequence through the graph embedding algorithm, and obtaining graph sequence data; The graph sequence data is vectorized and dimensionally unified, and the knowledge vector is obtained through nonlinear transformation.

5. The mineral deposit prediction method based on deep learning according to claim 1, characterized in that: The multi-task prediction model is constructed through the Transformer network according to the deep coding features and the knowledge vector, and the uncertainty assessment is performed through the Bayesian neural network framework to obtain the mineralization prediction result, including: The deep coding feature and the knowledge vector are concatenated and sequence converted by position coding to obtain input sequence data; Decomposing the input sequence data into tasks according to mineralization type, ore body size and grade prediction, and obtaining multi-task input data through task identifier encoding; Building the multi-task prediction model based on the multi-task input data, wherein the multi-task prediction model comprises a shared encoding layer, a task-specific decoding layer, and a prediction head layer, performing feature transformation through multi-layer self-attention calculation of the Transformer network to obtain task feature data; The task characteristic data are respectively input into the mineralization type prediction branch, the ore body size prediction branch and the grade prediction branch, and decoded through the task specific decoding layer to obtain branch prediction data; Constructing probability distribution parameters based on the branch prediction data, performing sampling and uncertainty evaluation through the Bayesian neural network framework, and obtaining prediction distribution data; The predicted distribution data is subjected to multi-task weighted fusion, and the mineralization prediction result is obtained through confidence threshold screening and normalization processing.

6. The method for mineral prediction based on deep learning according to claim 1, characterized in that: According to the mineralization prediction results, iterative optimization is performed through spatial clustering and Bayesian optimization algorithm to obtain optimized prediction results, including: The mineralization prediction results are sorted according to the prediction probability values, and high probability prediction area data is obtained by probability threshold segmentation; Constructing a spatial distance matrix based on the high probability prediction area data, grouping the areas through the spatial clustering, and obtaining potential target area data; Perform boundary extraction and spatial overlay analysis on the potential target area data, optimize through geological parameter constraints, and obtain preliminary optimized data; Inputting the preliminary optimization data into the Bayesian optimization algorithm, performing parameter optimization through sampling point selection and objective function evaluation, and obtaining supplementary survey location data; Based on the supplementary exploration location data, geological parameters are updated, and data optimization processing is performed through iterative calculation and parameter adjustment to obtain iterative optimization data; The iterative optimization data is comprehensively evaluated and weighted, and the optimization prediction result is obtained through multiple rounds of iterative calculations.

7. The mineral deposit prediction method based on deep learning according to claim 1, characterized in that: Based on the optimization prediction results, a three-dimensional visualization model is constructed using a deep generative model, and an explanatory analysis is performed using a Grad-CAM algorithm to obtain a target prediction verification result, including: Reconstructing the optimized prediction results according to spatial coordinates, and obtaining voxelized prediction data through three-dimensional interpolation processing; Inputting the voxelized prediction data into the deep generative model, and obtaining three-dimensional feature representation data through multi-layer encoding and decoding processing; Constructing a geometric topological relationship based on the three-dimensional feature representation data, optimizing the boundary through spatial constraints, and obtaining the three-dimensional visualization model; Decomposing the three-dimensional visualization model hierarchically, calculating the feature activation map by the Grad-CAM algorithm, and obtaining prediction basis data; Comparing and analyzing the prediction basis data with geological prior knowledge, conducting rationality evaluation through rule verification, and obtaining verification evaluation data; The verification and evaluation data are subjected to a multi-dimensional comprehensive analysis and weight calculation, and the target prediction verification result is obtained through confidence evaluation.

8. The mineral deposit prediction method based on deep learning according to claim 7, characterized in that: The step of performing hierarchical decomposition on the three-dimensional visualization model and calculating a feature activation map by using the Grad-CAM algorithm to obtain prediction basis data includes: Slice and decompose the three-dimensional visualization model to obtain layered data through gridding; Sorting and grouping the hierarchical data according to the depth information, and obtaining multi-level structure data through hierarchical index encoding; Extracting features from each level of the multi-level structure data, and obtaining feature weight data by target category gradient calculation; The feature weight data is weighted and superimposed with the feature map of the corresponding level, and the thermal map data is obtained by the Grad-CAM algorithm; Based on the heat map data, threshold segmentation and regional connectivity analysis are performed, and key area data are obtained by spatial correlation calculation; The key area data are sorted by feature importance and regionally integrated, and the prediction basis data are obtained through spatial combination relationship analysis.

9. A mineral deposit prediction device based on deep learning, used to implement the mineral deposit prediction method based on deep learning as described in any one of claims 1 to 8, characterized in that: The mineral deposit prediction device based on deep learning includes: The processing module is used to pre-process remote sensing image data, geophysical exploration data and geochemical data through data correction and normalization algorithms to obtain standardized multi-source data sets; A fusion module is used to construct a feature extraction model based on the standardized multi-source data set through an improved ResNet structure, a three-dimensional convolutional neural network, and a graph neural network, and to perform feature fusion using a variational autoencoder to obtain deep coding features; A construction module is used to construct a mineralization knowledge graph based on the deep coding features through a multi-head self-attention mechanism and a message passing mechanism, and obtain a knowledge vector using a graph embedding algorithm; An evaluation module, used to construct a multi-task prediction model through a Transformer network according to the deep coding features and the knowledge vector, and to perform uncertainty evaluation through a Bayesian neural network framework to obtain a mineralization prediction result; An optimization module, used to perform iterative optimization through spatial clustering and Bayesian optimization algorithm according to the mineralization prediction result to obtain an optimized prediction result; The analysis module is used to construct a three-dimensional visualization model based on the optimized prediction results using a deep generative model, and perform explanatory analysis through a Grad-CAM algorithm to obtain a target prediction verification result.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the deep learning-based mineral prediction method as described in any one of claims 1-8 is implemented.

Images