A three-dimensional metallogenic prediction method based on multi-source spectral data fusion

Abstract

The application discloses a three-dimensional metallogenic prediction method based on multi-source spectral data fusion. The method is characterized in that: historical data errors are corrected, multi-source spectral data are standardized, multi-source spectral data are spatially registered with geological records and gold grade test results to construct an integrated database, and multi-source heterogeneous data fusion storage is realized; geological prior constraints are embedded into machine learning model training to establish a nonlinear mapping between multi-source spectral features and mineralization probability; finally, a three-dimensional ore body shape model is coupled and constructed by using a mineralization probability constraint field to delineate a deep exploration target area, a leap from experience-driven prediction to data and knowledge dual-driven prediction is realized, the coincidence degree of the three-dimensional ore body model and the actual situation is improved, and drilling verification cost is reduced.

Description

Technical Field

[0001] This application relates to the field of mining technology, and in particular to a three-dimensional mineralization prediction method based on multi-source spectral data fusion. Background Technology

[0002] Mineral resource exploration and deep prospecting prediction are fundamental tasks in solid mineral development. With the increasing depletion of easily identifiable shallow ore bodies, three-dimensional metallogenic prediction technology based on multi-source data fusion has become an important development direction in gold exploration. In gold deposit exploration practice, constructing three-dimensional orebody models by integrating multi-source information from geological surveys, geophysics, and geochemistry can effectively reveal ore-controlling structural features and metallogenic regularities, providing target area guidance for deep drilling projects. For gold deposits like the Liaoshang Gold Deposit, with extremely complex orebody morphology, well-developed ore-controlling structures, and multi-scale spectral response characteristics of mineralization information, how to utilize multi-source spectral data such as shortwave infrared spectroscopy, thermal infrared spectroscopy, and portable X-ray fluorescence spectroscopy to achieve accurate inference of the three-dimensional morphology of ore bodies and intelligent prediction of metallogenic potential is a key technical problem that urgently needs to be solved in the deep exploration of complex structural gold deposits.

[0003] However, in existing 3D metallogenic prediction technologies for the Liaoshang gold deposit, orebody spatial modeling still primarily relies on manual ore body connection by geologists based on sparse borehole data. Due to the complex and varied combination of ore bodies in the planar and vertical directions, manual interpolation and orebody connection struggle to simultaneously account for the nonlinear characteristics of tectonic ore-controlling laws and multi-source geochemical constraints. Differences in the understanding of different geologists lead to significantly different modeling results, resulting in a large discrepancy between the constructed 3D orebody model and the actual situation. Furthermore, existing methods lack standardized multi-source spectral data acquisition procedures. Shortwave infrared spectroscopy, thermal infrared spectroscopy, and portable X-ray fluorescence spectroscopy data are acquired independently and stored separately. There is a lack of a unified 3D spatial coordinate framework and data registration mechanism between these data and the mine's existing geological logging database and gold grade testing results. This leads to broken spatial correlations among the heterogeneous multi-source data, resulting in spatial label noise in the machine learning training samples based on this database. Furthermore, most existing machine learning prediction models are black-box architectures, failing to effectively integrate the geological prior knowledge obtained from the analysis of mineralization regularities into model training. Their decision-making process is difficult to verify through interpretability methods, and the prediction results cannot be consistent with known geological regularities such as ore-controlling structures and alteration zoning. Geologists lack confidence in the intelligent prediction target area, which seriously restricts the practical application of prediction results in drilling engineering deployment and increases the amount of ineffective work and exploration costs in drilling verification. Summary of the Invention

[0004] This application proposes a three-dimensional mineralization prediction method that integrates multi-source spectral data. It has the advantage of deeply coupling geological prior knowledge with data-driven models, and can solve the problems of large subjective bias in manual mineralization, spatial correlation fracture of multi-source spectral data, and insufficient geological interpretability of machine learning prediction models in existing technologies, which lead to insufficient reliability of target area delineation.

[0005] To achieve the above objectives, this application provides the following technical solution: A three-dimensional mineralization prediction method based on multi-source spectral data fusion, characterized by the following specific steps: S1. Data Correction and Acquisition: Correct errors in historical exploration data, establish standardized acquisition procedures, collect rock and mineral samples, and complete spectral testing preprocessing. S2. Data Fusion and Database Construction: Spatially register multi-source spectral data with geological logging data and gold grade test results to build an integrated comprehensive database, establish related indexes, and achieve fusion storage; S3. Analysis of metallogenic regularity: Based on the database, the spectral-geochemical characteristics of ore bodies are identified, ore-forming rock bodies are identified, the ore-controlling structural features and metallogenic regularity are analyzed, the mapping relationship between spectral response and ore body distribution is established, and geological a priori constraints are formed. S4. Interpretable Model Construction: Using multi-dimensional fusion features as input, geological prior constraints are integrated into the training of machine learning models to construct a mineralization probability prediction model, analyze the contribution of each spectral feature to mineralization indication, and output the mineralization probability prediction results. S5. Three-dimensional target area delineation: Based on the mineralization probability prediction results, a three-dimensional morphological model of the ore body is constructed, and visualization results are generated to delineate the deep exploration target area and guide the verification of drilling projects.

[0006] Preferably, in step S1, the specific method for data correction and collection is as follows: S1.1 Based on the comparison between historical exploration data and actual underground geological conditions, correct systematic errors in coordinate deviation, stratigraphic sequence, structural morphology and ore body positioning, and simultaneously conduct detailed underground geological surveys to clarify the spatial occurrence relationship between rock masses, carbonate veins and ore bodies, assess the degree of impact of engineering disturbance on the original data, and establish geological data benchmarks. S1.2 Establish a standardized multi-source spectral data acquisition process, systematically collect rock and mineral samples from historical borehole cores and underground roadways, complete short-wave infrared spectroscopy, thermal infrared spectroscopy and X-ray fluorescence spectroscopy tests in a controlled indoor environment, and perform noise reduction, correction and normalization processing on the multi-source spectral data to construct a multi-source spectral geochemical dataset.

[0007] Preferably, in step S2, the specific method for data fusion and database construction is as follows; S2.1. Spatial registration and coordinate unification are performed between the standardized multi-source spectral geochemical dataset and the existing geological logging data, gold grade test results and geophysical interpretation data of the mine to eliminate spatial misalignment and coordinate system deviation caused by differences in acquisition benchmarks of multi-source data and establish a unified three-dimensional spatial coordinate framework. S2.2. Based on a unified three-dimensional spatial coordinate framework, construct an integrated comprehensive database, establish a multi-source heterogeneous data association index for spectral features, geochemical elements, geological structures and mineralization information, complete the fusion storage and collaborative management of multi-scale mineralization information under a unified spatial benchmark, and form a multi-source joint constraint mineralization association.

[0008] Preferably, step S2.1 is performed as follows: S2.11. Perform cross-source heterogeneous spatial registration between the spectral dataset and the existing geological logging data, gold grade test results, and geophysical interpretation data of the mine. Identify and eliminate spatial misalignment and coordinate deviation caused by differences in acquisition benchmarks of multi-source data, and realize position alignment and geometric correction of data from different sources under a unified spatial benchmark. S2.12. Based on the registered multi-source data, establish a three-dimensional spatial coordinate framework, incorporate spectral features, geochemical elements, geological structures, and mineralization information into the same coordinate system, eliminate spatial disconnect, establish a spatial correlation benchmark across data types, and realize accurate overlay and topological correlation of multi-source data in a unified space.

[0009] Preferably, step S2.2 is performed as follows: S2.21. Construct an integrated comprehensive database based on a three-dimensional spatial coordinate framework, organize spectral features, geochemical elements, geological structures and mineralization information in a structured manner according to a unified data architecture, perform fusion storage of multi-source heterogeneous data under precise spatial coordinate constraints, and establish a seamless access mechanism for cross-type data. S2.22. Establish a multi-source heterogeneous data association index within the integrated database, perform cross-type association mapping and semantic alignment of spectral features, geochemical elements, geological structures and mineralization information, and form a multi-source joint constraint mineralization association network through the collaborative management of multi-scale mineralization information.

[0010] Preferably, in step S3, the specific method for analyzing the mineralization regularity is as follows; S3.1. Call upon multi-source fusion data from the integrated database to determine the spectral-geochemical characteristics of ore bodies and ores, identify ore-forming rock bodies and accurately date them in conjunction with the genetic studies of typical regional ore deposits, analyze the control and destructive effects of ore-controlling structures on mineralization, summarize the mineralization regularity and condense the geological prior knowledge. S3.2 Construct a quantitative mapping relationship between the spatial distribution of ore bodies, alteration zoning and multi-source spectral response characteristics, and synergistically correlate the planar and vertical combination forms of ore bodies, pinch-out and recurrence patterns and the spatial distribution of lateral veins with spectral characteristics, so as to transform the mineralization regularity into the geological a priori constraints of the machine learning model.

[0011] Preferably, in step S4, the specific method for constructing the interpretable model is as follows; S4.1. Geological prior constraints are embedded into the machine learning model training process. The tectonic ore-controlling law and alteration zoning characteristics are introduced as constraints into the model optimization objective function. A mineralization probability prediction model is constructed by combining mineralization label samples and the parameters are iteratively optimized. A nonlinear mapping relationship between multi-source spectral features and mineralization probability is established. S4.2. Verify the interpretability of the prediction model. Quantitatively analyze the independent contribution and synergistic gain of each spectral feature to mineralization indication through feature attribution analysis. Evaluate the reliability of the model based on geological prior constraints. Ensure that the prediction results are consistent with the ore-controlling structures and alteration zoning distribution characteristics. Output the mineralization probability of each spatial location in the mining area.

[0012] Preferably, step S4.1 is performed as follows: S4.11. Transform the geological prior knowledge obtained from the analysis of mineralization regularity into computable constraints and embed it into the training process of machine learning model. Use the multi-dimensional fusion features in the integrated database as input, combine mineralization label samples to construct a mineralization probability prediction model, and establish an initial mapping between multi-source spectral features and mineralization response under geological prior constraints. S4.12. Perform iterative optimization on the mineralization probability prediction model, adjust the model parameters and hyperparameters under geological prior constraints, so that the spatial distribution of the prediction results conforms to the ore-controlling structure and alteration zoning law, and establish a stable nonlinear mapping between multi-source spectral characteristics and mineralization probability by cross-validation to constrain the model's generalization ability.

[0013] Preferably, step S4.2 is implemented in the following way: S4.21. Conduct interpretability verification of the mineralization probability prediction model, use the characteristic attribution analysis method to analyze the independent contribution and synergistic gain of short-wave infrared, thermal infrared and X-ray fluorescence spectral characteristics to mineralization indication, reveal the mechanism of multi-source spectroscopy driving mineralization prediction, and establish a quantitative correlation between characteristic importance and mineralization response. S4.22. Evaluate the reliability of model predictions under geological prior constraints, verify the consistency between the mineralization probability distribution and the ore-controlling structures and alteration zoning characteristics, ensure that the prediction results conform to the spatial constraints of metallogenic regularity, output the mineralization probability of each spatial location in the mining area, and provide data input for three-dimensional ore body morphology modeling.

[0014] Preferably, in step S5, the specific method for delineating the three-dimensional target area is as follows; S5.1 Receive the mineralization probability prediction results and couple them to the three-dimensional geological model of the mining area as a mineralization constraint field. Under the constraint of the spatial distribution of mineralization probability, infer the three-dimensional geometric shape of the ore body in the plane and vertical combination form, construct the three-dimensional spatial morphology model of the ore body, and minimize the subjective intervention of manual ore connection. S5.2. Generate three-dimensional visualized mineralization prediction results based on the three-dimensional spatial morphology model of the ore body, delineate deep exploration target areas with high mineralization potential according to the mineralization probability distribution characteristics, and output the spatial coordinates of the target area and the mineralization potential classification information to provide direct and clear spatial positioning guidance for drilling engineering verification.

[0015] The beneficial effects of this invention are as follows: This application provides a three-dimensional mineralization prediction method based on multi-source spectral data fusion. By correcting historical data errors and standardizing the acquisition of multi-source spectral data, it constructs an integrated database by spatially registering multi-source spectra with geological logging and gold grade analysis results, achieving multi-source heterogeneous data fusion and storage. By embedding geological prior constraints into machine learning model training, a nonlinear mapping between multi-source spectral features and mineralization probability is established. Finally, a three-dimensional orebody morphology model is constructed through mineralization probability constraint field coupling to delineate deep exploration target areas. This method achieves a leap from experience-driven to data- and knowledge-driven prediction, improving the consistency between the three-dimensional orebody model and the actual situation, and reducing drilling verification costs. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the method described in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] Example 1: Please refer to Figure 1 A three-dimensional mineralization prediction method based on multi-source spectral data fusion, characterized by the following specific steps: S1. Data Correction and Acquisition: Correct errors in historical exploration data, establish standardized acquisition procedures, collect rock and mineral samples, and complete spectral testing preprocessing. S2. Data Fusion and Database Construction: Spatially register multi-source spectral data with geological logging data and gold grade test results to build an integrated comprehensive database, establish related indexes, and achieve fusion storage; S3. Analysis of metallogenic regularity: Based on the database, the spectral-geochemical characteristics of ore bodies are identified, ore-forming rock bodies are identified, the ore-controlling structural features and metallogenic regularity are analyzed, the mapping relationship between spectral response and ore body distribution is established, and geological a priori constraints are formed. S4. Interpretable Model Construction: Using multi-dimensional fusion features as input, geological prior constraints are integrated into the training of machine learning models to construct a mineralization probability prediction model, analyze the contribution of each spectral feature to mineralization indication, and output the mineralization probability prediction results. S5. Three-dimensional target area delineation: Based on the mineralization probability prediction results, a three-dimensional morphological model of the ore body is constructed, and visualization results are generated to delineate the deep exploration target area and guide the verification of drilling projects.

[0021] In this embodiment: Step S1 corrects errors in historical exploration data and conducts detailed downhole geological surveys. At the same time, a standardized multi-source spectral acquisition process is established, enabling shortwave infrared spectroscopy, thermal infrared spectroscopy, and X-ray fluorescence spectroscopy data to be acquired under unified quality control standards. This significantly improves the spatial consistency and reliability of basic data, effectively eliminates the impact of original engineering disturbances and coordinate deviations on data credibility, and lays a high-quality data foundation for subsequent multi-source fusion and intelligent prediction.

[0022] Step S2 involves spatially registering and unifying the coordinates of multi-source spectral data with geological logging information and gold grade test results to construct an integrated database with accurate three-dimensional spatial coordinates. This enables the fusion storage and collaborative management of multi-scale mineralization information under a unified spatial benchmark, breaking down the barriers of long-term dispersed storage of spectral, geochemical, and geological data. It effectively solves the problems of data spatial misalignment and correlation distortion, providing multi-dimensional training samples with accurate spatial labels for machine learning.

[0023] Step S3 establishes a quantitative mapping relationship between orebody distribution and spectral response by defining the spectral and geochemical characteristics of the orebody and analyzing the ore-controlling structures and metallogenic regularities. This realizes the transformation of geological mechanism cognition into calculable geological a priori constraints, ensuring that subsequent model training is no longer divorced from geological reality, ensuring a high degree of consistency between the prediction process and known metallogenic regularities, and improving the geological rationality of mineralization prediction.

[0024] Step S4 integrates geological prior constraints into the training of the machine learning model and introduces an interpretability verification mechanism to quantitatively analyze the contribution and synergistic relationship of each spectral feature to mineralization indication. This achieves a high degree of consistency between the spatial distribution of the prediction results and the ore-controlling structures and alteration zoning characteristics. It breaks through the limitation of traditional black-box prediction models that cannot be verified with geological laws, and significantly improves the geologists' trust in the intelligent prediction results and the model's generalization ability in different mining sections.

[0025] Step S5 couples the mineralization probability prediction results to a three-dimensional geological model to construct the spatial morphology of the ore body, generating intuitive three-dimensional visualized mineralization prediction results. This effectively replaces the traditional manual subjective inference method of mineralization, minimizes the deviation caused by human intervention, accurately delineates deep exploration target areas and outputs mineralization potential classification information, effectively reduces the ineffective workload of drilling verification, and improves the scientific nature and economic benefits of exploration deployment.

[0026] Existing 3D mineralization prediction technologies have long relied on manual inference of ore body morphology through ore-connection, which is heavily constrained by subjective experience. Furthermore, multi-source spectral data is acquired independently and managed in a decentralized manner, lacking a unified spatial benchmark and fusion mechanism. This results in fragmented spatial correlations in the data and label noise in the model training samples. Simultaneously, traditional machine learning models are often black-box architectures, making it difficult to verify the consistency of prediction results with geological laws such as ore-controlling structures and alteration zoning. Geologists lack sufficient confidence in the intelligent prediction target areas, thus limiting the practical application of prediction results in drilling projects. This method establishes a standardized data acquisition process and an integrated database to achieve the fusion, storage, and collaborative management of multi-source heterogeneous data in precise three-dimensional spatial coordinates, eliminating data silos and spatial misalignment. By embedding geological prior constraints obtained from the analysis of mineralization regularities into model training and introducing interpretability verification, the prediction process is deeply coupled with geological mechanisms, ensuring a high degree of consistency between the spatial distribution of mineralization probability and known mineralization regularities. Finally, through three-dimensional orebody morphology modeling and visualized target area delineation, it replaces experience-driven subjective inference, achieving progressive optimization from the data layer to the model layer and then to the application layer. This significantly improves the consistency between the three-dimensional orebody model and the actual situation, reduces the ineffective cost of drilling verification, and provides reliable technical support for the deep exploration of complex structural gold deposits.

[0027] Example 2: Please refer to Figure 1 In step S1, the specific method for data correction and collection is as follows: S1.1 Based on the comparison between historical exploration data and actual underground geological conditions, correct systematic errors in coordinate deviation, stratigraphic sequence, structural morphology and ore body positioning, and simultaneously conduct detailed underground geological surveys to clarify the spatial occurrence relationship between rock masses, carbonate veins and ore bodies, assess the degree of impact of engineering disturbance on the original data, and establish geological data benchmarks. S1.2 Establish a standardized multi-source spectral data acquisition process, systematically collect rock and mineral samples from historical borehole cores and underground roadways, complete short-wave infrared spectroscopy, thermal infrared spectroscopy and X-ray fluorescence spectroscopy tests in a controlled indoor environment, and perform noise reduction, correction and normalization processing on the multi-source spectral data to construct a multi-source spectral geochemical dataset.

[0028] In this embodiment, the specific implementation process of step S1.1 is as follows: First, a set of comparison indicators for the differences between historical exploration data and actual underground geological conditions is established. This set of indicators includes four core indicators: coordinate system parameters, stratigraphic sequence coding, spatial trajectory of structural lines, and orebody boundary morphology. The borehole coordinates, geological profiles, and orebody projection maps in the historical exploration reports are spatially overlaid with the lithological boundaries, fault traces, and orebody outcrops actually exposed in the underground tunnels. The spatial deviation vector of the same geological element in the two types of data is calculated, and the systematic error distribution patterns of coordinate deviation, stratigraphic sequence displacement, structural morphology distortion, and orebody positioning offset are identified. For coordinate deviation, the least squares adjustment method is used to solve the transformation parameters from the historical coordinate system to the mining area's independent coordinate system. The historical borehole opening coordinates and inclination data are systematically corrected. For stratigraphic sequence and structural morphology errors, the rock... Based on the attitude of strata, fault attitude, and fold axes, the stratigraphic boundaries and structural parameters in historical data were redefined to eliminate systematic biases caused by insufficient accuracy in early exploration. Simultaneously, detailed underground geological surveys were conducted, with geological observation points arranged along the main exploration lines and mining roadway system to clarify the spatial relationships between rock masses, carbonate veins, and ore bodies. The three-dimensional spatial coordinates, attitude elements, and contact relationships of each lithological unit, structural element, and mineralized body were recorded. Based on this, an engineering disturbance assessment model was established, using the influence radius of mining operations as a spatial constraint. The degree of change in geological elements at the same location before and after mining was compared and analyzed. The degree of influence of engineering disturbance on the original data was divided into three levels: strong disturbance, moderate disturbance, and weak disturbance. Data from areas of strong disturbance were assigned low confidence weights, while data from areas of weak disturbance were assigned high confidence weights. A geological data benchmark was established by comprehensively correcting the spatial data and the confidence assessment results.

[0029] Step S1.1 involves a systematic comparison between historical exploration data and actual underground geological conditions in the mining area. This includes identifying and correcting coordinate deviations, as well as systematic errors in stratigraphic sequences, structural morphology, and orebody location. Simultaneously, a detailed underground geological survey is conducted to clarify the spatial relationships between rock masses, carbonate veins, and ore bodies, assess the impact of engineering disturbances on the original data, and establish a reliable geological data benchmark. This step completes the reliability assessment of historical data and the unification of spatial coordinates, eliminating systematic interference caused by the original mining operations on early exploration data. It provides accurate spatial positioning and a geological constraint framework for subsequent multi-source spectral acquisition, thereby preventing predictions from deviating from actual geological conditions due to distortion of basic data.

[0030] In step S1.2, the standardized acquisition process is implemented as follows: A rock and mineral sample acquisition specification is established, clarifying the point spacing, sample weight, and packaging and preservation requirements for systematically collecting rock and mineral samples from historical borehole cores and underground tunnels. Under a controlled indoor environment, short-wave infrared spectroscopy, thermal infrared spectroscopy, and X-ray fluorescence spectroscopy are performed sequentially. During the testing process, instrument parameter settings, environmental temperature and humidity conditions, and standard sample calibration procedures are standardized. The acquired multi-source spectral data undergoes denoising, correction, and normalization. Denoising uses a smoothing filter algorithm to eliminate instrument noise and environmental stray light interference. Correction uses a standard reflector or standard sample to correct spectral baseline drift. Normalization uses vector normalization or maximum-minimum normalization methods to unify the dimensions and numerical ranges between different spectral data, constructing a quality-controlled multi-source spectral geochemical dataset.

[0031] Step S1.2 establishes a standardized multi-source spectral data acquisition process, systematically collecting rock and mineral samples from historical borehole cores and underground tunnels. Short-wave infrared spectroscopy, thermal infrared spectroscopy, and X-ray fluorescence spectroscopy tests are then performed sequentially under controlled indoor conditions. The acquired multi-source spectral data undergoes denoising, correction, and normalization to construct a unified quality-controlled multi-source spectral geochemical dataset. This step standardizes and preprocesses the spectral data acquisition, eliminating data quality differences and dimensional biases between different test batches and different spectral equipment. This ensures that the three types of spectral data have a unified quality benchmark and comparability, thus providing standardized input for subsequent cross-source data fusion.

[0032] Existing technologies directly utilize uncorrected historical data, and the lack of unified quality control standards for spectral acquisition leads to spatial bias and quality heterogeneity in the basic data layer. Step S1 achieves dual improvement at the data source through the coordinated implementation of historical data correction and standardized acquisition. On the one hand, by comparing the differences between actual downhole surveys and historical data, systematic errors caused by coordinate deviations and engineering disturbances are eliminated, establishing a reliable geological data benchmark. On the other hand, through rigorous sampling procedures and controlled indoor testing, the quality consistency and spatial comparability of multi-source spectral data are ensured. As a result, subsequent multi-source data fusion is no longer limited by the interference of distorted basic data or heterogeneous spectral data, significantly improving the accuracy and reliability of the basic data layer in 3D mineralization prediction, and laying a solid data foundation for building a high-precision integrated comprehensive database.

[0033] Example 3: Please refer to Figure 1 In step S2, the specific method for data fusion and database construction is as follows; S2.1. Spatial registration and coordinate unification are performed between the standardized multi-source spectral geochemical dataset and the existing geological logging data, gold grade test results and geophysical interpretation data of the mine to eliminate spatial misalignment and coordinate system deviation caused by differences in acquisition benchmarks of multi-source data and establish a unified three-dimensional spatial coordinate framework. S2.2. Based on a unified three-dimensional spatial coordinate framework, construct an integrated comprehensive database, establish a multi-source heterogeneous data association index for spectral features, geochemical elements, geological structures and mineralization information, complete the fusion storage and collaborative management of multi-scale mineralization information under a unified spatial benchmark, and form a multi-source joint constraint mineralization association.

[0034] In this embodiment, the specific process of establishing the mineralization association network in step S2.2 is as follows: Using spatial grid cells or mineralization cells as network nodes, each node contains the spectral feature vector, geochemical element vector, and geological structural attribute vector of that spatial location; initial connection edges are established between nodes based on spatial adjacency; spatial adjacency edges are established when the Euclidean distance between two nodes is less than a preset spatial threshold; semantic association edges are established based on attribute similarity; the cosine similarity of spectral features and the Pearson correlation coefficient of geochemical element content between two nodes are calculated; semantic association edges are established when the similarity or correlation coefficient exceeds a preset threshold; edge weights are calculated by combining spatial adjacency and attribute similarity. The edge weights are obtained by weighted summation of spatial distance weights and attribute similarity weights. The closer the spatial distance and the more similar the attributes, the higher the edge weight. The higher the degree of the node, the greater the edge weight. By storing the node set and edge set through a graph data structure, a mineralization association network is formed. In this network, nodes with similar spectral characteristics are clustered through high-weight edges, nodes with related geochemical element content are connected through semantic association edges, and nodes with consistent geological structural attributes are aggregated through attribute constraint edges, realizing the collaborative management of multi-scale mineralization information. Furthermore, cross-type association mapping and semantic alignment rules are established to map alteration mineral types identified by short-wave infrared spectroscopy, rock and ore emissivity characteristics reflected by thermal infrared spectroscopy, element content determined by X-ray fluorescence spectroscopy, and structural attributes in geological logging to a unified semantic framework. This ensures that different data sources have consistent semantic expression in the mineralization association network, and improves the completeness and reliability of mineralization information identification through multi-source joint constraints.

[0035] Step S2.1 involves cross-source heterogeneous spatial registration of the standardized multi-source spectrogeochemical dataset with existing mine geological logging data, gold grade test results, and geophysical interpretation data. This step identifies and eliminates spatial misalignment and coordinate system deviations caused by differences in acquisition benchmarks among multi-source data. Through geometric correction and coordinate transformation, it aligns the positions of data from different sources under a unified spatial benchmark, establishing a unified three-dimensional spatial coordinate framework. This step unifies the spatial benchmark of heterogeneous data, solves the correlation distortion problem caused by inconsistencies in the coordinate systems of historical exploration data, and ensures accurate spatial correspondence between spectral test points, geological logging points, and grade test points, providing a precise spatial positioning foundation for the subsequent construction of an integrated database.

[0036] Step S2.2 constructs an integrated comprehensive database based on a unified three-dimensional spatial coordinate framework. Spectral features, geochemical elements, geological structures, and mineralization information are structured and organized according to a unified data architecture. A multi-source heterogeneous data association index is established within the database. Through cross-type association mapping, multi-scale mineralization information is fused, stored, and collaboratively managed under a unified spatial benchmark, forming multi-source jointly constrained mineralization associations. This step achieves deep fusion and integrated storage of multi-source heterogeneous data, breaking down the barriers of long-term dispersed storage of spectral, geochemical, and geological data. It realizes collaborative management of multi-scale mineralization information under precise spatial coordinates, providing machine learning models with spatially labeled, multi-dimensional fusion training samples.

[0037] In existing technologies, multi-source spectral data and geological data are stored separately and lack a unified spatial reference, leading to broken spatial correlations and noise in training sample labels. Step S2, through progressive implementation of spatial registration and database construction, first achieves spatial reference unification at the data layer, eliminating spatial misalignment caused by coordinate deviations; then, at the application layer, it achieves fused storage and collaborative management, establishing mineralization correlations across data types. Compared to the traditional decentralized management model, this step transforms independent data sources into an integrated comprehensive database with accurate three-dimensional spatial coordinates, enabling joint constraints on spectral features, geochemical elements, geological structures, and mineralization information within a unified framework. Therefore, subsequent metallogenic regularity analysis and model training can directly utilize spatially accurate multi-dimensional fused data, significantly improving the integrity and reliability of the data layer and providing a high-quality data support foundation for three-dimensional metallogenic prediction.

[0038] Example 4: Please refer to Figure 1 The specific method for step S2.1 is as follows: S2.11. Perform cross-source heterogeneous spatial registration between the spectral dataset and the existing geological logging data, gold grade test results, and geophysical interpretation data of the mine. Identify and eliminate spatial misalignment and coordinate deviation caused by differences in acquisition benchmarks of multi-source data, and realize position alignment and geometric correction of data from different sources under a unified spatial benchmark. S2.12. Based on the registered multi-source data, establish a three-dimensional spatial coordinate framework, incorporate spectral features, geochemical elements, geological structures, and mineralization information into the same coordinate system, eliminate spatial disconnect, establish a spatial correlation benchmark across data types, and realize accurate overlay and topological correlation of multi-source data in a unified space.

[0039] In this embodiment, the specific implementation of geometric correction and coordinate transformation in step S2.11 is as follows: An affine transformation model is used as the mathematical basis for spatial registration. The borehole coordinates and inclinometer data are used as corresponding control points. The translation parameters, rotation parameters, and scale parameters are solved to minimize the root mean square error of the corresponding control points of the spectral dataset, geological logging data, gold grade test results, and geophysical interpretation data. When the affine transformation cannot meet the local non-rigid deformation correction requirements, a quadratic polynomial transformation model is used to iteratively fit the residual error. Through point-by-point optimization, the positional deviation of each data layer under a unified spatial reference is less than the preset tolerance threshold, thereby achieving accurate positional alignment and geometric correction of data from different sources under a unified spatial reference.

[0040] Step S2.11 involves cross-source heterogeneous spatial registration of the spectral dataset with existing mine geological logging data, gold grade test results, and geophysical interpretation data to identify and eliminate spatial misalignment and coordinate deviation caused by differences in acquisition benchmarks among multi-source data. This step uses borehole coordinates and inclination data as spatial benchmarks, combines sample depth information to calculate the three-dimensional spatial coordinates of each data point, and achieves positional alignment of data from different sources under a unified spatial benchmark through geometric correction and coordinate transformation, thus realizing precise calibration of the spatial position of multi-source heterogeneous data. This eliminates spatial misalignment caused by inconsistencies in coordinate systems between historical exploration data and spectral test data, ensuring accurate spatial correspondence between spectral test points, geological logging points, and grade test points. It avoids incorrect matching between spectral features and mineralization labels due to spatial misalignment, providing a precise positional benchmark for the subsequent establishment of a unified three-dimensional spatial coordinate framework.

[0041] Step S2.12 establishes a three-dimensional spatial coordinate framework based on the registered multi-source data, incorporating spectral features, geochemical elements, geological structures, and mineralization information into the same coordinate system. This framework uses the mining area's independent coordinate system as a reference to establish a spatial correlation benchmark across data types. Through a spatial index structure, it achieves precise overlay and topological association of multi-source data in a unified space, realizing the unified construction of a spatial benchmark for multi-source data. This eliminates the spatial disconnect between spectral, geochemical, and geological data, providing data from different sources with a unified spatial coordinate identifier. This provides a spatial index foundation for the subsequent construction of an integrated database, ensuring precise overlay and topological association of multi-source data under a unified spatial benchmark.

[0042] In existing technologies, multi-source spectral data, geological logging data, and grade analysis results are stored in different coordinate systems, lacking a unified spatial benchmark, leading to broken spatial correlations and sample label noise. Step S2.1, through the progressive implementation of spatial registration and coordinate framework, eliminates spatial misalignment and coordinate deviation caused by differences in acquisition benchmarks at the data layer, achieving positional alignment of data from different sources; at the framework layer, spectral features, geochemical elements, geological structures, and mineralization information are incorporated into the same coordinate system, establishing a cross-data type spatial correlation benchmark. Compared to traditional methods, this step transforms independent data sources into a unified spatial benchmark, ensuring accurate spatial correspondence between spectral test points, geological logging points, and grade analysis points, improving the spatial consistency and correlation accuracy of multi-source data, and providing a spatial positioning foundation for subsequent database construction.

[0043] Example 5: Please refer to Figure 1 The specific method for step S2.2 is as follows: S2.21. Construct an integrated comprehensive database based on a three-dimensional spatial coordinate framework, organize spectral features, geochemical elements, geological structures and mineralization information in a structured manner according to a unified data architecture, perform fusion storage of multi-source heterogeneous data under precise spatial coordinate constraints, and establish a seamless access mechanism for cross-type data. S2.22. Establish a multi-source heterogeneous data association index within the integrated database, perform cross-type association mapping and semantic alignment of spectral features, geochemical elements, geological structures and mineralization information, and form a multi-source joint constraint mineralization association network through the collaborative management of multi-scale mineralization information.

[0044] In this embodiment: Step S2.21 constructs an integrated comprehensive database based on a three-dimensional spatial coordinate framework. The spatially registered spectral features, geochemical elements, geological structures, and mineralization information are organized in a structured manner according to a unified data architecture. A relational spatial database management system is used to perform fusion storage of multi-source heterogeneous data under precise spatial coordinate constraints, establishing a seamless access mechanism for cross-type data and realizing the construction of an integrated storage architecture for multi-source heterogeneous data. This step eliminates the barriers to long-term dispersed storage of spectral, geochemical, and geological data, enabling multi-source data to have a consistent data organization format and access interface under a unified spatial reference, providing structured data support for subsequent metallogenic regularity analysis and machine learning model training.

[0045] Step S2.22 establishes a multi-source heterogeneous data association index within the integrated database, performing cross-type association mapping and semantic alignment of spectral features, geochemical elements, geological structures, and mineralization information. It constructs primary key and foreign key association relationships and a spatial topological index, achieving semantic correspondence between different data types at the mineralization unit scale. Through the collaborative management of multi-scale mineralization information, a multi-source jointly constrained mineralization association network is formed, realizing the establishment of semantic layer associations for multi-source data. This step enables the effective extraction of weak mineralization anomalies that cannot be identified by a single data source through multi-source joint responses, improving the completeness and reliability of mineralization information identification and providing multi-dimensional jointly constrained training samples for machine learning models.

[0046] In existing technologies, multi-source spectral data, geological logging data, and grade analysis results are typically stored in independent databases or file systems, lacking a unified data architecture and spatial association mechanism. This results in low data retrieval efficiency and difficulties in cross-type joint analysis. Step S2.2, through the progressive implementation of integrated database construction and association index establishment, firstly, organizes multi-source heterogeneous data into a structured manner according to a unified architecture at the storage layer, achieving integrated storage and seamless access. Subsequently, a cross-type semantic mapping and mineralization association network is established at the association layer, enabling spectral features, geochemical elements, geological structures, and mineralization information to form multi-source joint constraints. Compared to the traditional distributed storage model, this step realizes the transformation of independent data sources into an integrated comprehensive database, giving multi-source data a unified organizational format, spatial identification, and semantic association, significantly improving data retrieval efficiency and cross-type joint analysis capabilities, and providing high-quality data layer support for three-dimensional mineralization prediction.

[0047] Example 6: Please refer to Figure 1 In step S3, the specific method for analyzing the mineralization regularity is as follows; S3.1. Call upon multi-source fusion data from the integrated database to determine the spectral-geochemical characteristics of ore bodies and ores, identify ore-forming rock bodies and accurately date them in conjunction with the genetic studies of typical regional ore deposits, analyze the control and destructive effects of ore-controlling structures on mineralization, summarize the mineralization regularity and condense the geological prior knowledge. S3.2 Construct a quantitative mapping relationship between the spatial distribution of ore bodies, alteration zoning and multi-source spectral response characteristics, and synergistically correlate the planar and vertical combination forms of ore bodies, pinch-out and recurrence patterns and the spatial distribution of lateral veins with spectral characteristics, so as to transform the mineralization regularity into the geological a priori constraints of the machine learning model.

[0048] In this embodiment, the statistical analysis method in step S3.1 is implemented as follows: spectral characteristic parameters of known mineralized sections and surrounding rock sections are extracted from the integrated database, and the mean and standard deviation of each characteristic parameter in the mineralized section and surrounding rock section are calculated respectively; the box plot method is used to identify the abnormal lower limit of the spectral characteristic parameters, and the spectral response that exceeds the abnormal lower limit and is consistent with the distribution trend of the mean of the mineralized section is regarded as a mineralization indicator feature, or the mean plus three times the standard deviation method is used to determine the mineralization indicator threshold range, and the spectral characteristics that meet the threshold range are included in the mineralization indicator feature set, so as to provide a quantitative discrimination benchmark for determining the spectral-geochemical characteristic features of ore bodies and ores.

[0049] Step S3.1 systematically defines the spectral-geochemical characteristics of ore bodies and ores by calling multi-source fusion data from the integrated database. This step extracts information on sericite and chlorite alteration mineral assemblages identified by short-wave infrared spectroscopy, emissivity data of rocks and ores reflected by thermal infrared spectroscopy, and gold and associated element content data determined by X-ray fluorescence spectroscopy from the database. Statistical analysis methods are used to determine the threshold range of mineralization indicator characteristics. Combined with the genetic studies of typical regional deposits, zircon U-Pb dating technology is used to identify rock bodies closely related to mineralization and accurately define the mineralization age. The controlling effects of ore-controlling faults and subsequent tectonic activity on the formation and destruction of ore bodies are analyzed, and the combination forms and distribution patterns of ore bodies in the planar and vertical directions are clarified, effectively condensing prior geological knowledge. This transforms the tectonic ore-controlling mechanism and alteration zoning patterns into calculable geological constraints, providing geological theoretical support for subsequent machine learning model training and preventing model training from deviating from geological reality.

[0050] The specific construction method of the quantitative mapping relationship in step S3.2 is as follows: First, extract the spatial distribution parameters of known mineralized bodies from the integrated database, including the planar extension length, vertical extension thickness, pinch-out and recurrence interval, and horizontal offset of lateral veins. Simultaneously, extract the multi-source spectral response characteristics at the corresponding locations, including the absorption peak depth of altered minerals in shortwave infrared spectroscopy, the emissivity values ​​of rocks and minerals in thermal infrared spectroscopy, and the gold and associated element content values ​​in X-ray fluorescence spectroscopy. A quantitative mapping function between the spatial distribution parameters of the ore body and the spectral response characteristics is constructed using a multivariate nonlinear regression method, with spectral characteristics as the independent variable and mineralization degree indicators as the dependent variable. The nonlinear mapping relationship is fitted, and the regression coefficients are calculated. The planar and vertical combination of ore bodies, pinch-out and recurrence patterns, and spatial distribution and spectral characteristics of lateral veins are synergistically correlated to establish three types of geological a priori constraints: ore-controlling fault distance weighting coefficient, alteration mineral assemblage threshold, and elemental content gradient parameter. The ore-controlling fault distance weighting coefficient is calculated based on the vertical distance between the sample point and the known ore-controlling fault; the closer the distance, the larger the weighting coefficient. The alteration mineral assemblage threshold is determined statistically based on the spectral characteristics of typical mineralized sections. The elemental content gradient parameter is calculated based on the rate of change of elemental content between adjacent sample points. These quantitative mapping relationships and constraint parameters are written into a geological a priori knowledge base as external constraints during the training process of the machine learning model, ensuring that the model training is guided by mineralization laws.

[0051] Step S3.2, based on the geological prior knowledge refined in S3.1, constructs a quantitative mapping relationship between the spatial distribution of ore bodies, alteration zoning, and multi-source spectral response characteristics. This step synergistically correlates the planar and vertical combination forms of ore bodies, pinch-out and recurrence patterns, and the spatial distribution of lateral veins with shortwave infrared alteration mineral characteristics, thermal infrared emissivity characteristics, and elemental content characteristics. A multivariate regression analysis method is used to establish a quantitative conversion function between spectral response intensity and mineralization degree. Furthermore, the mineralization regularity is transformed into geological prior constraints for the machine learning model, specifically including the weight coefficient of ore-controlling fault distance, alteration mineral combination threshold, and elemental content gradient parameters. This ensures that the model training process is constrained by geological regularity, achieving a high degree of consistency between the prediction results and geological regularity. This ensures that the spatial distribution of mineralization probability is consistent with the ore-controlling structures and alteration zoning distribution characteristics, improving the geological rationality of the model predictions.

[0052] In existing technologies, machine learning model training often adopts a purely data-driven approach, without incorporating prior geological knowledge. This results in predictions that are frequently disconnected from ore-controlling structures and alteration zoning patterns. Step S3, through a progressive implementation of metallogenic regularity analysis and quantitative mapping, firstly calls upon multi-source fused data from an integrated database at the knowledge layer to define spectral-geochemical characteristic markers and analyze ore-controlling structures, thus refining prior geological knowledge. Subsequently, at the constraint layer, a quantitative mapping relationship between ore body distribution and spectral response is constructed, transforming metallogenic regularities into calculable constraints. Compared to the traditional purely data-driven model, this step embeds geological mechanism knowledge into the pre-training stage of the model, guiding the machine learning model with geological regularities from the initial training phase. This significantly improves the consistency between prediction results and geological reality, avoiding data fitting without physical meaning.

[0053] Example 7: Please refer to Figure 1 In step S4, the specific method for constructing the interpretable model is as follows; S4.1. Geological prior constraints are embedded into the machine learning model training process. The tectonic ore-controlling law and alteration zoning characteristics are introduced as constraints into the model optimization objective function. A mineralization probability prediction model is constructed by combining mineralization label samples and the parameters are iteratively optimized. A nonlinear mapping relationship between multi-source spectral features and mineralization probability is established. S4.2. Verify the interpretability of the prediction model. Quantitatively analyze the independent contribution and synergistic gain of each spectral feature to mineralization indication through feature attribution analysis. Evaluate the reliability of the model based on geological prior constraints. Ensure that the prediction results are consistent with the ore-controlling structures and alteration zoning distribution characteristics. Output the mineralization probability of each spatial location in the mining area.

[0054] In this embodiment, the specific calculation method of the geological consistency penalty term in step S4.1 is as follows: For each spatial location in the training samples, the deviation value between its predicted mineralization probability and the geological prior constraint is calculated. The deviation value includes the product term of the normalized distance from the location to the nearest ore-controlling fault and the predicted mineralization probability of the location, and the cosine distance term between the alteration mineral assemblage characteristics of the location and the alteration characteristics of the typical mineralization section. The product term and the cosine distance term are weighted and summed according to a preset weight coefficient to obtain the single-sample penalty value. Then, the penalty values ​​of all training samples are averaged to obtain the geological consistency penalty term. The geological consistency penalty term is superimposed with the standard loss function according to a preset ratio to form a comprehensive loss function to guide the model learning direction to conform to geological laws.

[0055] The specific transformation rules for the computable constraints in step S4.1 are as follows: The spatial constraint matrix uses the spatial coordinates of each training sample as rows and columns, and the matrix elements are the reciprocal of the difference in the weight coefficients of the ore-controlling fracture distance between two samples. When a sample is located within the ore-controlling fracture influence zone, the matrix elements are assigned high weights, and when located outside the influence zone, they are assigned low weights. The feature screening rules retain feature dimensions in the shortwave infrared spectrum whose absorption peak depth exceeds the threshold of the alteration mineral assemblage, and remove weak response dimensions that are below the threshold. The sample weight adjustment factor is determined based on the ratio of the element content gradient parameter of the sample point to the average gradient of the region. Samples located in the high gradient zone are assigned higher weights, and samples located in the low gradient zone are assigned lower weights.

[0056] The specific implementation of embedding geological prior constraints in step S4.1 is as follows: Geological prior knowledge obtained from the analysis of mineralization regularities is transformed into computable constraints. Specifically, this includes converting the weight coefficients of ore-controlling fault distances into spatial constraint matrices, converting alteration mineral combination thresholds into feature selection rules, and converting element content gradient parameters into sample weight adjustment factors. During the training of the machine learning model, the above constraints are embedded into the model optimization objective function. A geological consistency penalty term is added to the standard loss function. This penalty term is calculated based on the degree of deviation between the model prediction results and the geological prior constraints. When the prediction results conflict with the known distribution of ore-controlling faults or alteration zoning patterns, the loss value is dynamically increased to constrain the model's learning direction. Using the multi-dimensional fusion features in the integrated database as input, mineralization label samples are obtained by binarizing the gold grade test results after threshold determination. The spatial constraint matrix and feature selection rules are applied to the input feature layer, and the sample weight adjustment factors are applied to the sampling probability of the training samples to construct a mineralization probability prediction model. Under geological prior constraints, an initial mapping between multi-source spectral features and mineralization response is established.

[0057] Step S4.1 embeds geological prior constraints into the machine learning model training process, using multi-dimensional fused features from an integrated database as input samples, and constructs a mineralization probability prediction model by combining mineralization label samples. This step transforms tectonic ore-controlling laws and alteration zoning characteristics into constraints introduced into the model's objective function. A geological consistency penalty term is added to the loss function; when the model's prediction results conflict with known tectonic ore-controlling laws, the loss value is dynamically increased, guiding the model's learning direction to conform to geological laws. The model adopts a gradient boosting architecture, with 500 weak learners, a maximum tree depth of 6, a learning rate of 0.05, and a subsampling ratio of 0.8. Parameters are iteratively optimized through five-fold cross-validation, establishing a nonlinear mapping relationship between multi-source spectral features and mineralization probability, achieving deep integration of geological prior knowledge and a data-driven model. This avoids physically meaningless fitting caused by purely data-driven approaches, ensuring that the spatial distribution of the model's prediction results conforms to the ore-controlling structures and alteration zoning laws, thus improving the geological rationality of mineralization predictions.

[0058] The specific implementation of interpretability verification in step S4.2 is as follows: Using the Shapley additive interpretation value as the feature attribution analysis method, for each predicted sample, the marginal contribution value of each input feature under different feature subset combinations is calculated. The Shapley value of each spectral feature is obtained through weighted averaging, and this value represents the independent contribution of that feature to the mineralization prediction result of the sample. Further, the synergistic gain index between features is calculated. By comparing the difference between the prediction contribution of the combined feature combination and the sum of the prediction contributions of individual features, the synergistic gain relationship between shortwave infrared, thermal infrared, and X-ray fluorescence spectral features is quantified. Based on the Shapley value and synergistic gain index of each feature, a quantitative correlation between feature importance and mineralization response is established, identifying the key spectral response mechanisms driving mineralization prediction and making the model decision-making process transparent.

[0059] Step S4.2 verifies the interpretability of the mineralization probability prediction model by employing feature attribution analysis to quantitatively analyze the independent contributions and synergistic gain relationships of short-wave infrared alteration mineral characteristics, thermal infrared emissivity characteristics, and X-ray fluorescence elemental content characteristics to mineralization indication. This step calculates the marginal contribution values ​​of each input feature in different mineralization samples, identifies the key spectral response mechanisms driving mineralization prediction, and assesses the model's reliability based on geological prior constraints. The prediction results are then compared and verified for consistency with ore-controlling structures and alteration zoning characteristics. When the model outputs a high mineralization probability, the corresponding feature contribution distribution is simultaneously output to verify whether it is driven by the proximity of ore-controlling structures and alteration mineral combinations, thus achieving transparency in the prediction process and credibility of the results. This overcomes the limitation of traditional black-box models that cannot be verified against geological laws, enabling geologists to understand the prediction basis based on feature contribution, significantly increasing trust in intelligent prediction results.

[0060] In existing technologies, machine learning prediction models are mostly black-box architectures, lacking prior geological knowledge during training. This often leads to prediction results that are disconnected from ore-controlling structures and alteration zoning patterns, and the model's decision-making process is uninterpretable, making it difficult for geologists to trust the predicted target area. Step S4, through the progressive implementation of embedding geological prior constraints and verifying interpretability, firstly introduces ore-controlling structures and alteration zoning characteristics as constraints into the model optimization objective at the training layer, guiding the model's learning direction towards geological laws. Subsequently, at the verification layer, feature attribution analysis is used to quantitatively analyze the contribution of spectral features, ensuring a high degree of consistency between the prediction results and geological laws. Compared to the traditional purely data-driven model, this step achieves a leap from experience-driven to data- and knowledge-driven prediction, giving the mineralization probability prediction results clear geological mechanism support and interpretability, significantly improving the geological rationality and engineering application credibility of the model prediction.

[0061] Example 8: Please refer to Figure 1 The specific method for step S4.1 is as follows: S4.11. Transform the geological prior knowledge obtained from the analysis of mineralization regularity into computable constraints and embed it into the training process of machine learning model. Use the multi-dimensional fusion features in the integrated database as input, combine mineralization label samples to construct a mineralization probability prediction model, and establish an initial mapping between multi-source spectral features and mineralization response under geological prior constraints. S4.12. Perform iterative optimization on the mineralization probability prediction model, adjust the model parameters and hyperparameters under geological prior constraints, so that the spatial distribution of the prediction results conforms to the ore-controlling structure and alteration zoning law, and establish a stable nonlinear mapping between multi-source spectral characteristics and mineralization probability by cross-validation to constrain the model's generalization ability.

[0062] In this embodiment, the specific method for obtaining mineralization label samples in step S4.11 is as follows: using the industrial grade of the mining area as the benchmark threshold, samples with gold grade test results higher than or equal to the industrial grade are marked as positive samples, and samples with gold grade lower than the industrial grade are marked as negative samples. This completes the binarization of mineralization labels, so that mineralization label samples and multidimensional fusion features form a one-to-one correspondence in the sample space, providing supervision labels for training the mineralization probability prediction model.

[0063] Step S4.11 transforms the geological prior knowledge obtained from the analysis of mineralization regularities into computable constraints and embeds it into the machine learning model training process. This step uses multi-dimensional fusion features from an integrated database as input. Mineralization label samples are obtained by binarizing gold grade test results after thresholding. Ore-controlling fault distances and alteration zoning characteristics are assigned initial weights higher than spectral features as prior features. Simultaneously, a geological consistency penalty term is added to the loss function. A mineralization probability prediction model is constructed based on the mineralization label samples. Under geological prior constraints, an initial mapping between multi-source spectral features and mineralization response is established, achieving the initial fusion of geological prior knowledge and a data-driven model. This guides the model training direction to geological regularities, avoiding numerical fitting lacking geological basis and providing an initial model architecture consistent with geological realities for subsequent iterative optimization.

[0064] The specific implementation of iterative optimization in step S4.12 is as follows: In each iteration, the standard loss value is calculated first, and then the geological consistency penalty term is calculated. The two are weighted and summed to obtain the comprehensive loss value. The model parameters are adjusted according to the gradient direction of the comprehensive loss value. Under the geological prior constraints, the model hyperparameters are adjusted, including the number of weak learners, the maximum tree depth, the learning rate, and the subsampling ratio, so that the spatial distribution of the prediction results conforms to the ore-controlling structure and alteration zoning law. The five-fold cross-validation strategy is used to constrain the model's generalization ability. The training dataset is divided into five subsets. Four subsets are used for training and one subset for validation. After five iterations, the average validation accuracy is taken as the evaluation index of generalization ability. When the validation accuracy meets the preset threshold and the spatial distribution of the prediction results meets the preset standard of consistency with the geological prior constraints, the iteration is stopped and the model parameters are output to establish a stable nonlinear mapping between multi-source spectral features and mineralization probability.

[0065] Step S4.12 involves iterative optimization of the mineralization probability prediction model, adjusting model parameters and hyperparameters under geological prior constraints. This step employs a gradient boosting decision tree architecture, setting the number of weak learners to 500, the maximum tree depth to 6, the learning rate to 0.05, and the subsampling ratio to 0.8. A five-fold cross-validation strategy is used to constrain the model's generalization ability, ensuring that the spatial distribution of the prediction results conforms to the ore-controlling structures and alteration zoning patterns. A stable nonlinear mapping between multi-source spectral characteristics and mineralization probability is established, achieving a simultaneous improvement in model prediction stability and geological rationality. This ensures consistent prediction performance across different ore sections, avoids regional prediction bias, and outputs mineralization probability prediction results with spatial distribution conforming to geological patterns.

[0066] In existing technologies, machine learning model training often employs a purely data-driven approach, neglecting to incorporate prior geological knowledge. This results in predictions that are frequently disconnected from ore-controlling structures and alteration zoning patterns, and model parameter optimization lacks geological constraints. Step S4.1, through a progressive implementation of geological prior embedding and iterative optimization, firstly transforms ore-forming laws into computable constraints for model training during the initial mapping construction stage, guiding the model's learning direction towards geological laws. Subsequently, in the parameter optimization stage, cross-validation constrains generalization capabilities, ensuring that the spatial distribution of prediction results conforms to geological laws. Compared to the traditional purely data-driven model, this step achieves deep coupling between geological prior knowledge and data-driven algorithms, enabling the mineralization probability prediction model to possess geologically reasonable constraints and stable generalization capabilities, significantly improving the consistency between prediction results and geological realities.

[0067] Example 9: Please refer to Figure 1 The specific method for step S4.2 is as follows: S4.21. Conduct interpretability verification of the mineralization probability prediction model, use the characteristic attribution analysis method to analyze the independent contribution and synergistic gain of short-wave infrared, thermal infrared and X-ray fluorescence spectral characteristics to mineralization indication, reveal the mechanism of multi-source spectroscopy driving mineralization prediction, and establish a quantitative correlation between characteristic importance and mineralization response. S4.22. Evaluate the reliability of model predictions under geological prior constraints, verify the consistency between the mineralization probability distribution and the ore-controlling structures and alteration zoning characteristics, ensure that the prediction results conform to the spatial constraints of metallogenic regularity, output the mineralization probability of each spatial location in the mining area, and provide data input for three-dimensional ore body morphology modeling.

[0068] In this embodiment: Step S4.21 verifies the interpretability of the mineralization probability prediction model by using feature attribution analysis to calculate the marginal contributions of shortwave infrared, thermal infrared, and X-ray fluorescence spectral features. This step analyzes the independent contribution and synergistic gain relationship of each spectral feature to mineralization indication, identifies key feature combinations driving mineralization prediction, reveals the mechanism of multi-source spectral-driven mineralization prediction, and establishes a quantitative correlation between feature importance and mineralization response. This completes the transparent analysis of the model decision-making process, providing clear spectral feature support for the prediction results. Geologists can understand the model's prediction logic based on feature contribution, significantly increasing trust in the intelligent prediction results and avoiding the shortcomings of black-box models that cannot be verified.

[0069] The specific implementation method of consistency verification in step S4.22 is as follows: Spatial overlay analysis is performed on the mineralization probability spatial distribution map, the ore-controlling structure distribution map, and the alteration zoning map. The spatial overlap rate between the predicted high value area and the known ore-controlling fault zone and alteration mineral assemblage distribution area is calculated. When the overlap rate exceeds the preset threshold, the prediction result is determined to conform to the spatial constraints of the metallogenic law. When the overlap rate is lower than the preset threshold, the prediction result is determined to conflict with the geological law and is screened out. The abnormal prediction values ​​after screening are marked and their characteristic attribution results are analyzed retrospectively to identify data anomalies or model limitations that cause prediction deviations. For the mineralization probability prediction results that pass the consistency verification, the mineralization probability values ​​of each location in the mining area are output according to the spatial coordinates to form regular three-dimensional grid data, providing geologically reliable data input for subsequent three-dimensional ore body morphology modeling.

[0070] Step S4.22 assesses the reliability of the model predictions under geological prior constraints by verifying the consistency between the spatial distribution of mineralization probability and the characteristics of ore-controlling structures and alteration zoning. This step compares the spatial superposition relationship between predicted high-value areas and known ore-controlling fault zones and alteration mineral assemblages to verify whether the prediction results conform to the spatial constraints of metallogenic regularities, filters out abnormal prediction values ​​that conflict with geological regularities, and outputs the verified mineralization probability for each spatial location in the mining area. This ensures that the prediction results are highly consistent with geological prior knowledge, providing geologically reliable data input for subsequent 3D orebody morphology modeling and avoiding target area deviations caused by building 3D models based on erroneous prediction results.

[0071] In existing technologies, machine learning prediction models are mostly black-box architectures, with uninterpretable decision-making processes. Geologists cannot determine whether the prediction results are driven by actual mineralization characteristics, and there is a lack of mechanisms to verify the consistency between prediction results and geological laws, resulting in insufficient credibility of the predicted target area. Step S4.2, through the progressive implementation of feature attribution analysis and consistency verification, firstly analyzes the marginal contribution of each spectral feature to mineralization indication at the mechanistic level, establishing a quantitative correlation between feature importance and mineralization response, making the model prediction logic transparent. Subsequently, at the verification level, the mineralization probability distribution is spatially overlaid and verified with the ore-controlling structures and alteration zoning distribution, filtering out abnormal predictions that conflict with geological laws. Compared with the traditional black-box prediction mode, this step achieves a dual improvement in the interpretability of the prediction process and the geological credibility of the results, giving the mineralization probability prediction results clear spectral feature support and geological law verification, significantly enhancing the practical application value of intelligent prediction results in drilling engineering deployment.

[0072] Example 10: Please refer to Figure 1 In step S5, the specific method for delineating the three-dimensional target area is as follows; S5.1 Receive the mineralization probability prediction results and couple them to the three-dimensional geological model of the mining area as a mineralization constraint field. Under the constraint of the spatial distribution of mineralization probability, infer the three-dimensional geometric shape of the ore body in the plane and vertical combination form, construct the three-dimensional spatial morphology model of the ore body, and minimize the subjective intervention of manual ore connection. S5.2. Generate three-dimensional visualized mineralization prediction results based on the three-dimensional spatial morphology model of the ore body, delineate deep exploration target areas with high mineralization potential according to the mineralization probability distribution characteristics, and output the spatial coordinates of the target area and the mineralization potential classification information to provide direct and clear spatial positioning guidance for drilling engineering verification.

[0073] In this embodiment, the specific implementation of mineralization constraint field coupling in step S5.1 is as follows: First, using borehole data as hard constraints, a three-dimensional geological framework model of the mining area is constructed by combining stratigraphic occurrence and structural elements to determine the spatial equations of stratigraphic interfaces and structural surfaces. The predicted mineralization probability values ​​of discrete points are extended into three-dimensional probability volume data using ordinary kriging interpolation. During the interpolation process, the mineralization probability value is used as a regional variable, and the spatial distance is used as a structure function variable. A variogram model is fitted and the interpolation weights are solved. Based on the three-dimensional probability volume data obtained by interpolation, the moving cube algorithm is used to extract mineralization probability isosurfaces. A probability threshold parameter is set, and isosurfaces that meet the threshold conditions in the three-dimensional probability field are tracked. Under the constraint of the spatial distribution of mineralization probability, the three-dimensional geometric shape of the ore body in planar and vertical combination is inferred, and a three-dimensional spatial morphology model of the ore body is constructed. This process replaces manual ore connection with data-driven methods, minimizing model deviations caused by differences in the subjective experience of geologists, and making the inference of the three-dimensional morphology of the ore body controlled by the joint constraints of the spatial distribution of mineralization probability and the geological framework.

[0074] Step S5.1 receives the mineralization probability prediction results output from step S4 and couples them as a mineralization constraint field to the three-dimensional geological model of the mining area. This step constructs a three-dimensional framework of strata and structure using borehole data as hard constraints. It then uses Kriging interpolation to expand the mineralization probability prediction values ​​of discrete points into three-dimensional probability volume data. Subsequently, it extracts isosurfaces using the moving cube algorithm and infers the three-dimensional geometric shape of the ore body's planar and vertical combinations under the constraints of the mineralization probability spatial distribution, constructing a three-dimensional spatial morphology model of the ore body. This effectively replaces the subjective inference method of manual ore connection. This minimizes model deviations caused by differences in the experience of geologists, ensuring that the inference of the ore body's three-dimensional morphology is controlled by the joint constraints of the mineralization probability spatial distribution and the geological framework, thus improving the consistency between the three-dimensional ore body model and the actual situation.

[0075] Step S5.2 generates 3D visualized mineralization prediction results based on a 3D spatial morphology model of the ore body. This step sets threshold parameters based on the mineralization probability distribution characteristics, delineating continuous areas with mineralization probabilities higher than the threshold and spatial topological correlations with the main controlling fault zone as high-mineralization-potential deep exploration target areas. It outputs the target area's spatial coordinates, volumetric parameters, and mineralization potential classification information, providing direct spatial positioning guidance for drilling engineering verification. This ensures that the target area delineation results have clear mineralization probability support and structural correlation basis, avoiding the arbitrariness of traditional manual target area delineation and improving the targeting and economic efficiency of exploration deployment.

[0076] In existing technologies, the inference of 3D orebody morphology mainly relies on manual ore connection by geologists based on sparse borehole data. This method is heavily constrained by subjective experience, resulting in significant differences in inference results among different personnel. Furthermore, the target area delineation lacks quantitative basis, leading to a large amount of ineffective drilling verification work. Step S5 implements a progressive approach through mineralization constraint field coupling and visualized target area delineation. First, at the modeling layer, mineralization probability volume data is interpolated and isosurfaces are extracted to form a 3D morphology model, replacing experience-driven subjective inference. Subsequently, at the application layer, the target area is delineated based on mineralization probability thresholds and structural topology correlation, and spatial coordinates are output. Compared to the traditional manual method, this step realizes the transformation of 3D orebody morphology inference from experience-driven to data-driven, enabling target area delineation to have quantitative support from mineralization probability, significantly reducing the ineffective cost of drilling verification, and improving the scientific rigor and economic benefits of deep exploration in complex structural gold deposits.

[0077] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0078] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A three-dimensional mineralization prediction method based on multi-source spectral data fusion, characterized in that: The specific steps are as follows: S1. Data Correction and Acquisition: Correct errors in historical exploration data, establish standardized acquisition procedures, collect rock and mineral samples, and complete spectral testing preprocessing. S2. Data Fusion and Database Construction: Spatially register multi-source spectral data with geological logging data and gold grade test results to build an integrated comprehensive database, establish related indexes, and achieve fusion storage; S3. Analysis of metallogenic regularity: Based on the database, the spectral-geochemical characteristics of ore bodies are identified, ore-forming rock bodies are identified, the ore-controlling structural features and metallogenic regularity are analyzed, the mapping relationship between spectral response and ore body distribution is established, and geological a priori constraints are formed. In step S3, the specific method for analyzing the mineralization regularity is as follows; S3.

1. Call upon multi-source fusion data from the integrated database to determine the spectral-geochemical characteristics of ore bodies and ores, identify ore-forming rock bodies and accurately date them in conjunction with the genetic studies of typical regional ore deposits, analyze the control and destructive effects of ore-controlling structures on mineralization, summarize the mineralization regularity and condense the geological prior knowledge. S3.2 Construct a quantitative mapping relationship between the spatial distribution of ore bodies, alteration zoning and multi-source spectral response characteristics, and synergistically correlate the planar and vertical combination forms of ore bodies, pinch-out and recurrence patterns and the spatial distribution of lateral veins with spectral characteristics, and transform the mineralization regularity into the geological a priori constraints of the machine learning model. S4. Interpretable Model Construction: Using multi-dimensional fusion features as input, geological prior constraints are integrated into the training of machine learning models to construct a mineralization probability prediction model, analyze the contribution of each spectral feature to mineralization indication, and output the mineralization probability prediction results. In step S4, the specific method for constructing the interpretable model is as follows; S4.

1. Geological prior constraints are embedded into the machine learning model training process. The tectonic ore-controlling law and alteration zoning characteristics are introduced as constraints into the model optimization objective function. A mineralization probability prediction model is constructed by combining mineralization label samples and the parameters are iteratively optimized. A nonlinear mapping relationship between multi-source spectral features and mineralization probability is established. S4.

2. Verify the interpretability of the prediction model. Quantitatively analyze the independent contribution and synergistic gain of each spectral feature to mineralization indication through feature attribution analysis. Evaluate the reliability of the model based on geological prior constraints. Ensure that the prediction results are consistent with the ore-controlling structures and alteration zoning distribution characteristics. Output the mineralization probability of each spatial location in the mining area. S5. Three-dimensional target area delineation: Based on the mineralization probability prediction results, a three-dimensional morphological model of the ore body is constructed to generate visualization results, delineate the deep exploration target area, and guide the verification of drilling projects. In step S5, the specific method for delineating the three-dimensional target area is as follows; S5.1 Receive the mineralization probability prediction results and couple them to the three-dimensional geological model of the mining area as a mineralization constraint field. Under the constraint of the spatial distribution of mineralization probability, infer the three-dimensional geometric shape of the ore body in the plane and vertical combination form, construct the three-dimensional spatial morphology model of the ore body, and minimize the subjective intervention of manual ore connection. S5.

2. Generate three-dimensional visualized mineralization prediction results based on the three-dimensional spatial morphology model of the ore body, delineate deep exploration target areas with high mineralization potential according to the mineralization probability distribution characteristics, and output the spatial coordinates of the target area and the mineralization potential classification information to provide direct and clear spatial positioning guidance for drilling engineering verification.

2. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 1, characterized in that, In step S1, the specific method for data correction and collection is as follows: S1.1 Based on the comparison between historical exploration data and actual underground geological conditions, correct systematic errors in coordinate deviation, stratigraphic sequence, structural morphology and ore body positioning, and simultaneously conduct detailed underground geological surveys to clarify the spatial occurrence relationship between rock masses, carbonate veins and ore bodies, assess the degree of impact of engineering disturbance on the original data, and establish geological data benchmarks. S1.2 Establish a standardized multi-source spectral data acquisition process, systematically collect rock and mineral samples from historical borehole cores and underground roadways, complete short-wave infrared spectroscopy, thermal infrared spectroscopy and X-ray fluorescence spectroscopy tests in a controlled indoor environment, and perform noise reduction, correction and normalization processing on the multi-source spectral data to construct a multi-source spectral geochemical dataset.

3. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 2, characterized in that, In step S2, the specific method for data fusion and database construction is as follows; S2.

1. Spatial registration and coordinate unification are performed between the standardized multi-source spectral geochemical dataset and the existing geological logging data, gold grade test results and geophysical interpretation data of the mine to eliminate spatial misalignment and coordinate system deviation caused by differences in acquisition benchmarks of multi-source data and establish a unified three-dimensional spatial coordinate framework. S2.

2. Based on a unified three-dimensional spatial coordinate framework, construct an integrated comprehensive database, establish a multi-source heterogeneous data association index for spectral features, geochemical elements, geological structures and mineralization information, complete the fusion storage and collaborative management of multi-scale mineralization information under a unified spatial benchmark, and form a multi-source joint constraint mineralization association.

4. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 3, characterized in that, The specific method for step S2.1 is as follows: S2.

11. Perform cross-source heterogeneous spatial registration between the spectral dataset and the existing geological logging data, gold grade test results, and geophysical interpretation data of the mine. Identify and eliminate spatial misalignment and coordinate deviation caused by differences in acquisition benchmarks of multi-source data, and realize position alignment and geometric correction of data from different sources under a unified spatial benchmark. S2.

12. Based on the registered multi-source data, establish a three-dimensional spatial coordinate framework, incorporate spectral features, geochemical elements, geological structures, and mineralization information into the same coordinate system, eliminate spatial disconnect, establish a spatial correlation benchmark across data types, and realize accurate overlay and topological correlation of multi-source data in a unified space.

5. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 4, characterized in that, The specific method for step S2.2 is as follows: S2.

21. Construct an integrated comprehensive database based on a three-dimensional spatial coordinate framework, organize spectral features, geochemical elements, geological structures and mineralization information in a structured manner according to a unified data architecture, perform fusion storage of multi-source heterogeneous data under precise spatial coordinate constraints, and establish a seamless access mechanism for cross-type data. S2.

22. Establish a multi-source heterogeneous data association index within the integrated database, perform cross-type association mapping and semantic alignment of spectral features, geochemical elements, geological structures and mineralization information, and form a multi-source joint constraint mineralization association network through the collaborative management of multi-scale mineralization information.

6. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 5, characterized in that, The specific method for step S4.1 is as follows: S4.

11. Transform the geological prior knowledge obtained from the analysis of mineralization regularity into computable constraints and embed it into the training process of machine learning model. Use the multi-dimensional fusion features in the integrated database as input, combine mineralization label samples to construct a mineralization probability prediction model, and establish an initial mapping between multi-source spectral features and mineralization response under geological prior constraints. S4.

12. Perform iterative optimization on the mineralization probability prediction model, adjust the model parameters and hyperparameters under geological prior constraints, so that the spatial distribution of the prediction results conforms to the ore-controlling structure and alteration zoning law, and establish a stable nonlinear mapping between multi-source spectral characteristics and mineralization probability by cross-validation to constrain the model's generalization ability.

7. The three-dimensional mineralization prediction method based on multi-source spectral data fusion according to claim 6, characterized in that, The specific method for step S4.2 is as follows: S4.

21. Conduct interpretability verification of the mineralization probability prediction model, use the characteristic attribution analysis method to analyze the independent contribution and synergistic gain of short-wave infrared, thermal infrared and X-ray fluorescence spectral characteristics to mineralization indication, reveal the mechanism of multi-source spectroscopy driving mineralization prediction, and establish a quantitative correlation between characteristic importance and mineralization response. S4.

22. Evaluate the reliability of model predictions under geological prior constraints, verify the consistency between the mineralization probability distribution and the ore-controlling structures and alteration zoning characteristics, ensure that the prediction results conform to the spatial constraints of metallogenic regularity, output the mineralization probability of each spatial location in the mining area, and provide data input for three-dimensional ore body morphology modeling.

Images