A deep concealed ore deposit detection method based on multi-source geological data fusion

Abstract

The present application proposes an innovative deep concealed deposit detection method, which comprehensively uses geology, geophysics, geochemistry and remote sensing technology, and realizes the effective integration of multi-source geological data by developing an advanced data fusion algorithm. The method first standardizes the data collected by geological survey, wide area electromagnetic method exploration, hydrocarbon mercury superimposed halo analysis and aerial hyperspectral remote sensing for pretreatment to ensure the consistency and accuracy of the data. Then, using machine learning and statistical analysis technology, the fused data is analyzed in depth to identify and predict the mineralization mode and abnormal area. Finally, a three-dimensional geological model is constructed to intuitively show the shape and distribution of the deposit, providing a scientific basis for geological exploration. The innovation of the present application lies in its multi-source data fusion technology and the automation and precision of mineralization mode identification, which significantly improves the efficiency and accuracy of deep concealed deposit detection.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and in particular to a comprehensive detection method that combines geology, geophysics, geochemistry and remote sensing technologies to improve the detection accuracy and efficiency of deep concealed mineral deposits. Background Technology

[0002] In the field of geological exploration, traditional methods for detecting mineral deposits typically rely on single geological, geophysical, geochemical, or remote sensing technologies. These methods are often limited by the scope, depth, and accuracy of the exploration work. For example, geological methods mainly analyze the distribution of rocks and minerals through surface observation and sampling, but this method is difficult to use for deep underground exploration. Geophysical methods, such as seismic exploration or electromagnetic methods, can detect underground structures, but often require a large amount of computation and interpretation work, and have limited ability to identify deep, concealed mineral deposits. Geochemical methods look for mineralization signs by analyzing the chemical composition of soil and rocks, but these methods are easily affected by changes in the surface environment and also face challenges in deep mineral deposit detection. Remote sensing technologies, especially airborne or satellite remote sensing, can cover a large area of ​​the Earth's surface, but traditional remote sensing technologies have limited resolution and depth in identifying underground mineralization characteristics. To overcome these limitations and improve the efficiency and accuracy of detecting deep, concealed mineral deposits, there is an urgent need for an innovative method that integrates multiple geological data. This method should fully utilize the advantages of geological, geophysical, geochemical, and remote sensing data, and through advanced data processing and analysis techniques, achieve deeper and more precise detection of underground mineralization characteristics under different metallogenic geological conditions. This invention is proposed against this backdrop, aiming to address the shortcomings of existing technologies in the detection of deep, concealed mineral deposits through technological innovation. Summary of the Invention

[0003] The purpose of this invention is to provide a comprehensive method for detecting deep, concealed mineral deposits. This method utilizes the advantages of geology, geophysics, geochemistry, and remote sensing technologies, and through a series of innovative data processing and analysis procedures, it achieves high-precision detection of underground mineralization characteristics.

[0004] The methods for detecting deep, concealed mineral deposits include the following steps:

[0005] S1. Geological Data Acquisition: High-precision GPS equipment was used to record the latitude, longitude, and elevation information of the sampling points. Through systematic field observation and sample collection, we recorded in detail key geological information such as rock type, mineral composition, stratigraphic structure, and igneous rocks.

[0006] S2. Geophysical Data Acquisition: In the geophysical data acquisition phase, we employed wide-area electromagnetic method technology, using a ground coil system to measure the resistivity and electromagnetic wave propagation characteristics of the subsurface medium. The transmitting coil frequency range was set from 1Hz to 1kHz to detect subsurface structures at different depths, achieving a resistivity measurement accuracy of 1%, ensuring the accurate acquisition of the electrical characteristics of the subsurface medium.

[0007] S3. Geochemical Data Acquisition: In the geochemical data acquisition step, we used high-precision analytical instruments to determine the types of hydrocarbon minerals, mercury content, and other key elements in soil and rock samples. The mercury content analysis accuracy reached the nanogram level (ng / g), and the content analysis of other elements such as gold (Au), copper (Cu), lead (Pb), and zinc (Zn) also reached the corresponding high-precision standards. These data are crucial for identifying geochemical anomaly areas.

[0008] S4. Remote Sensing Data Acquisition: Airborne hyperspectral remote sensing technology is used to perform detailed spectral analysis of target areas. It can capture spectral data ranging from visible light to near-infrared or short-wave infrared, with resolution typically at the nanometer level. This data allows us to identify and analyze mineral species within the region, including their chemical composition and physical properties. By analyzing spectral characteristic curves, we can determine the unique spectral signature of specific minerals, thereby identifying and analyzing mineral types, surface locations, extent, and distribution characteristics.

[0009] S5. Data Preprocessing: The data preprocessing steps involve cleaning, standardizing, and denoising the collected multi-source geological data. The cleaning process includes removing invalid data and outliers, while the standardization process ensures that all data have uniform dimensions and format. The denoising process applies filtering algorithms such as wavelet transform to minimize data noise, improving data quality and consistency.

[0010] S6. Multi-source Data Fusion: In the multi-source data fusion step, we developed a fusion algorithm that uses Principal Component Analysis (PCA) for data dimensionality reduction, retaining 95% of the components explained by variance, and Independent Component Analysis (ICA) for feature extraction. Furthermore, multi-scale analysis techniques are used to integrate data at different scales to comprehensively identify mineralization features.

[0011] S7. Pattern Recognition and Analysis: In the pattern recognition and analysis phase, we employed various machine learning algorithms, including Support Vector Machines (SVM), Random Forest, and deep learning networks. SVM used radial basis function (RBF) kernels, optimizing classification performance by adjusting parameters C and γ. Random Forest constructed a forest of 100 trees and used 10-fold cross-validation to optimize parameters. The deep learning network employed a convolutional neural network (CNN) structure with a depth of 5 layers, trained using the ReLU activation function and the Adam optimizer.

[0012] S8. 3D Geological Model Construction: The 3D geological model construction steps utilize professional geological modeling software, combined with geostatistical methods such as Kriging, to construct a 3D geological model of the target area. The model considers geological structure, geophysical response, and geochemical anomalies. Spatial correlation parameters are optimized through simulated annealing algorithms, predicting the spatial distribution characteristics of the ore deposit.

[0013] S9. Based on the model's predicted potential deposit locations, we planned and executed precise drilling operations to collect core samples for lithological description and mineral composition analysis. Simultaneously, geophysical logging techniques, such as resistivity and natural gamma logging, were applied to obtain the electrical characteristics of the subsurface medium. These measured data were meticulously compared with the model's predictions, and any deviations triggered further model adjustments. The adjustment process may involve re-evaluating the geological model's input parameters, optimizing machine learning algorithm parameters such as the number of layers and neuron configurations in DBNs, CNNs, and RNNs, or adjusting the variance range and nugget values ​​in geostatistical methods.

[0014] Preferably, the data preprocessing step begins with data calibration. This involves adjusting the raw data by comparing it with known standards or reference samples to eliminate measurement bias and systematic errors, ensuring the data calibration accuracy meets the expected standards. Next, outlier removal is performed using statistical methods such as standard deviation analysis or boxplots to identify outliers in the dataset. These outliers may be caused by measurement errors, sample contamination, or other atypical geological conditions. The outlier removal criteria are strictly set; for example, any value exceeding the mean ± 3 standard deviations is considered an outlier and excluded. Furthermore, robust mathematical transformations, such as logarithmic or Z-score transformations, are introduced to reduce the impact of extreme values ​​and improve the normality of the data.

[0015] Preferably, in this invention, the optimized implementation of the data fusion algorithm is achieved through precise principal component analysis (PCA) and independent component analysis (ICA) techniques. PCA is performed by calculating the covariance matrix Σ of the original dataset X, then performing eigenvalue decomposition to obtain eigenvectors vi and eigenvalues ​​λi. The top k eigenvectors with the largest eigenvalues ​​(k = 10) are selected to form the transformation matrix W, thereby projecting the original data into a new space while retaining principal components with a variance contribution rate exceeding 85%. The mathematical expression is: W = [v1, v2, ... … In the ICA phase, the algorithm iteratively maximizes the independence of independent source signals with non-Gaussian distributions, using curtosis or negentropy as the independence measure, and sets the convergence criterion as a change in the independence index less than ∈ = 0.001. The iterative update formula of ICA can be expressed as: Here, w is the weight vector, s is the signal extracted from the PCA-derived data, and t is the iteration number. This process continues until the convergence condition is met or the maximum number of iterations is reached, typically not exceeding 1000. By combining PCA and ICA methods, the data fusion algorithm not only improves the interpretability of the data but also significantly enhances the ability to identify key features of deep, concealed mineral deposits through precise mathematical models and strict parameter control. This method ensures the extraction of the most informative features from multi-source geological data, providing a solid foundation for subsequent mineral deposit exploration and resource assessment.

[0016] Preferably, in the pattern recognition and analysis steps, a series of advanced machine learning models are employed to improve the recognition accuracy of mineralization features. These models include Deep Belief Networks (DBNs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs), each optimized for different data characteristics and analytical needs. DBNs consist of multiple stacked Restricted Boltzmann Machines (RBMs), each containing 1000 neurons. They pre-train features through unsupervised learning and then fine-tune them through supervised learning to automatically extract complex patterns from the data. CNN models contain 5 convolutional layers and 3 fully connected layers, using the ReLU activation function and cross-entropy loss function. The convolutional kernel size is set to 3x3 with a stride of 1 to capture local spatial features and perform effective feature mapping. RNNs, particularly Long Short-Term Memory Networks (LSTMs), contain 3 layers with 128 neurons each, used to process sequential data and capture long-term dependencies in time series. All models used the Adam optimizer for parameter updates with a learning rate of 0.001. Overfitting was controlled using early stopping, and training was stopped when the performance improvement on the validation set was less than 0.01. The ensemble use of these models enabled us to analyze geological data from multiple perspectives and levels, significantly improving the accuracy and efficiency of mineral deposit exploration.

[0017] Preferably, geostatistical methods are employed to enhance the predictive accuracy of the 3D geological model during its construction. Specifically, kriging is used for spatial data interpolation and prediction. A semivariogram function is established to describe the correlation of spatial variables. A spherical model is selected as the semivariogram function, with parameters including a range of 500 meters, a sill value of 1.5 times the coefficient of variation, and a baseline effect of 10% of the sill value, to ensure the model's sensitivity to short-range spatial correlations. Furthermore, simulated annealing is applied to optimize the parameters of the geostatistical model to avoid local optima. The initial temperature is set to 1000 °C, the cooling rate to 0.98, and the Metropolis criterion is used to determine the acceptance probability of new solutions. Through these fine-tuning parameters and the application of geostatistical methods, the 3D geological model can more accurately reflect the spatial distribution characteristics of underground mineralization, thereby significantly improving the predictive accuracy and reliability of mineral deposit detection.

[0018] Preferably, the verification and optimization steps utilize comprehensive exploration data for in-situ verification and optimization of the model. Specifically, this step encompasses core drilling analysis, including detailed recording of lithological descriptions, mineral composition, structure, and mineralization characteristics of the cores. Drilling depths reach 1500 meters, with a core recovery rate exceeding 90%, ensuring accurate acquisition of subsurface geological information. Geophysical logging data collection employs various logging methods, including resistivity, natural gamma, density, and neutron porosity, with measurement accuracy controlled within 1% to verify the electrical characteristics of the geological model. Geochemical sampling results are analyzed by examining elemental content in soil, water, and rock samples, such as mercury, copper, lead, and zinc, with an accuracy reaching the ppb (parts per billion) level, to identify geochemical anomalies. This comprehensive analysis of exploration data provides practical geological, geophysical, and geochemical evidence for the geological model, enabling precise adjustments and optimizations, thereby improving the accuracy and reliability of mineral deposit detection.

[0019] The present invention proposes a highly integrated detection system for the deep concealed mineral deposit detection method. This system is carefully designed with multiple functional modules to achieve full-process automation and high efficiency. The system includes a data acquisition module responsible for field acquisition of geological, geophysical, geochemical, and remote sensing data; a data preprocessing module to ensure data quality by performing cleaning, standardization, and outlier removal, with data calibration accuracy controlled within 0.5%; a data fusion module applying PCA and ICA techniques for data dimensionality reduction and feature extraction, with PCA retaining over 90% of the variance components and the ICA independence index change threshold set to 0.0001; a pattern recognition module integrating machine learning algorithms such as DBN, CNN, and RNN, using the Adam optimizer with a learning rate set to 0.001; a 3D geological modeling module using Kriging for spatial data interpolation and simulated annealing algorithm to optimize parameters, with an initial temperature set to 1000°C and a cooling factor of 0.98; and a validation and optimization module combining 1500-meter-deep drilling core analysis, geophysical logging data with 1% accuracy, and geochemical sampling results with ppb-level accuracy to perform field validation and optimization of the model. The collaborative work of these modules ensures fully automated processing from data acquisition to final model verification, significantly improving the efficiency and accuracy of mineral deposit detection.

[0020] According to the deep concealed mineral deposit detection method described in claims 1 to 7, the computer program product of this invention is an integrated package containing professional software code designed to automate key steps such as data acquisition, preprocessing, fusion, pattern recognition, 3D geological modeling, and verification and optimization. The software is developed using an efficient programming language, ensuring rapid processing and high compatibility, and adheres to industry coding standards. The product is stored on a CD, hard drive, or solid-state drive, and users can deploy it on a personal computer or server through simple installation steps. The software interface is user-friendly, providing detailed user manuals and online help documentation to help users quickly master the operation methods. The program adopts a modular design, facilitating future functional expansion and updates, ensuring the long-term effectiveness and adaptability of the software. This computer program product provides users with a comprehensive, accurate, and easy-to-use solution for deep concealed mineral deposit detection, significantly improving the efficiency and accuracy of geological exploration work.

[0021] Furthermore, in this invention, the method or system for detecting deep concealed mineral deposits also includes an advanced user interface, providing users with an intuitive and easy-to-use interactive platform. This interface allows users to input geological parameters, such as rock type, mineral composition, and stratigraphic age, and select data preprocessing options, fusion algorithms, machine learning models, and other analysis methods. Users can view the data processing flow, pattern recognition results, and the construction process and final results of the 3D geological model in real time. All this information is displayed graphically, ensuring clarity and comprehensibility. The user interface also features model adjustment capabilities, providing graphical adjustment tools such as sliders, drop-down menus, and buttons, allowing users to directly fine-tune model parameters within the interface without complex programming or command input. The adjusted results are immediately displayed on the interface, facilitating rapid evaluation of the adjustment effect. The interface also integrates log recording and history recording functions, recording all operation steps and model status, supporting users to backtrack and compare versions. To ensure result sharing and further analysis, the user interface supports multiple data output formats, including PDF, CSV, and DXF, enabling users to easily export model results and related data. Overall, this user interface design greatly enhances the system's usability and flexibility, making the exploration of deep, concealed mineral deposits more efficient and accurate, while providing users with a rich interactive experience.

[0022] The present invention also includes an innovative geological exploration decision support tool, in addition to the aforementioned deep concealed mineral deposit detection method or system. This tool is designed to intelligently recommend and prioritize potential mineral deposits based on the prediction results obtained from the aforementioned method, and to allocate resources accordingly. It analyzes three-dimensional geological models, geochemical anomalies, geophysical responses, and other relevant geological data, employing advanced algorithms to assess the mineralization potential and exploration value of various regions. The decision support tool uses a multi-criteria decision analysis method, comprehensively considering factors such as geological characteristics, exploration costs, expected returns, and risk levels, to automatically generate an exploration priority list. It can also provide customized resource allocation schemes based on the size of the exploration team, equipment configuration, and project schedule, ensuring the efficient organization and execution of exploration activities. The tool features a user-friendly interface, allowing geological exploration teams to easily input relevant parameters, adjust evaluation criteria, and optimize recommendation results. It also supports integration with existing exploration management software and databases, enabling seamless data exchange and real-time updates. Attached Figure Description

[0023] Figure 1 This is a flowchart of the data preprocessing process for the deep concealed mineral deposit detection method of the present invention.

[0024] Figure 2 This is a flowchart illustrating the steps of the deep concealed mineral deposit detection method of the present invention. Detailed Implementation

[0025] The purpose of this invention is to provide a comprehensive method for detecting deep, concealed mineral deposits. This method utilizes the advantages of geology, geophysics, geochemistry, and remote sensing technologies, and through a series of innovative data processing and analysis procedures, it achieves high-precision detection of underground mineralization characteristics.

[0026] The methods for detecting deep, concealed mineral deposits include the following steps:

[0027] S1. Geological Data Acquisition: High-precision GPS equipment was used to record the latitude, longitude, and elevation information of sampling points. Through systematic field observation and sample collection, we recorded in detail key geological information such as rock type, mineral composition, stratigraphic structure, and igneous rocks. Geological samples underwent laboratory analysis, such as X-ray diffraction (XRD), to determine the quantitative composition of minerals, providing basic geological information for subsequent data processing and ore deposit prediction.

[0028] S2. Geophysical Data Acquisition: In the geophysical data acquisition phase, we employed wide-area electromagnetic method technology, using a ground coil system to measure the resistivity and electromagnetic wave propagation characteristics of the subsurface medium. The transmitting coil frequency range was set from 1Hz to 1kHz to detect subsurface structures at different depths, achieving a resistivity measurement accuracy of 1%, ensuring the accurate acquisition of the electrical characteristics of the subsurface medium.

[0029] S3. Geochemical Data Acquisition: In the geochemical data acquisition step, we used high-precision analytical instruments to determine the mercury content and other key elements in soil, water, and rock samples. The mercury content analysis in soil samples reached the nanogram level (ng / g), and the content analysis of other elements such as gold (Au), copper (Cu), lead (Pb), and zinc (Zn) also reached the corresponding high-precision standards. These data are crucial for identifying geochemical anomaly areas.

[0030] S4. Remote Sensing Data Acquisition: In the remote sensing data acquisition phase, we utilize airborne hyperspectral remote sensing technology to collect spectral characteristic data of the target area. The sensors used have a high spatial resolution of better than 5 meters and a wide wavelength range of 400 nm to 2500 nm, which provides rich information for analyzing the surface mineral composition and surface cover, thereby identifying and analyzing minerals, types, surface locations, extent, and distribution characteristics.

[0031] S5. Data Preprocessing: The data preprocessing steps involve cleaning, standardizing, and denoising the collected multi-source geological data. The cleaning process includes removing invalid data and outliers, while the standardization process ensures that all data have uniform dimensions and format. The denoising process applies filtering algorithms such as wavelet transform to minimize data noise, improving data quality and consistency.

[0032] S6. Multi-source Data Fusion: In the multi-source data fusion step, we developed a fusion algorithm that uses Principal Component Analysis (PCA) for data dimensionality reduction, retaining 95% of the components explained by variance, and Independent Component Analysis (ICA) for feature extraction. Furthermore, multi-scale analysis techniques are used to integrate data at different scales to comprehensively identify mineralization features.

[0033] S7. Pattern Recognition and Analysis: In the pattern recognition and analysis phase, we employed various machine learning algorithms, including Support Vector Machines (SVM), Random Forest, and deep learning networks. SVM used radial basis function (RBF) kernels, optimizing classification performance by adjusting parameters C and γ. Random Forest constructed a forest of 100 trees and used 10-fold cross-validation to optimize parameters. The deep learning network employed a convolutional neural network (CNN) structure with a depth of 5 layers, trained using the ReLU activation function and the Adam optimizer.

[0034] S8. 3D Geological Model Construction: The 3D geological model construction steps utilize professional geological modeling software, combined with geostatistical methods such as Kriging, to construct a 3D geological model of the target area. The model considers geological structure, geophysical response, and geochemical anomalies. Spatial correlation parameters are optimized through simulated annealing algorithms, predicting the spatial distribution characteristics of the ore deposit.

[0035] In step S9, Verification and Optimization, we accurately verified the model-predicted deposit location through field drilling and geophysical logging. Specifically, drilling was conducted in the mineralized areas predicted by the model, with drilling depths determined based on geological model predictions and geological structure analysis, typically reaching 1500 meters, and core recovery exceeding 95%. During drilling, detailed lithological descriptions and mineral composition analyses were performed every 10 meters, using techniques such as X-ray diffraction (XRD) and electron probe microanalysis (EPMA) to ensure accurate identification of mineral components. Geophysical logging data, including resistivity and natural gamma logging, were collected with an accuracy controlled within 1% to obtain the electrical characteristics of the subsurface medium. The collected drilling and geophysical data were compared and analyzed with the model predictions; any deviation between the predictions and actual data triggered model adjustments. For example, if the mineralized area predicted by the model is not found during drilling, we will re-evaluate the input parameters of the geological model and adjust the parameters in the machine learning algorithm. For instance, in a DBN, the number of layers may be increased from 3 to 5; in a CNN, the kernel size may be adjusted from 3x3 to 5x5, and the stride from 1 to 2; or in an RNN, the time step may be increased from 5 to 10, and the number of neurons from 128 to 256. Kriging parameters in geostatistical methods, such as the variance range from 200 meters to 500 meters and the nugget value from 1.5 times the coefficient of variation to 2 times, will also be optimized based on actual data. Through this iterative verification and optimization process based on parameter values, we can significantly improve the predictive accuracy and reliability of the mineral deposit detection model, ensuring its effective application in actual geological exploration.

[0036] Data preprocessing begins with data calibration. Raw measurements are adjusted to within the standard error range, typically within ±0.5%, by comparing them against international geological sample standards. Next, outlier removal is performed. A threshold of 1.5 times the IQR (interquartile range) is determined by calculating the dataset's quartiles and interquartile ranges. Any data point exceeding the 25th percentile minus 1.5 times the IQR or exceeding the 75th percentile plus 1.5 times the IQR is considered an outlier and excluded. Furthermore, the Z-score method is applied to further identify and remove outliers; points with an absolute Z-score greater than 3 are considered outliers. Through these precise parameter settings and rigorous quality control measures, the data preprocessing steps ensure the accuracy and reliability of the dataset, providing a solid foundation for subsequent multi-source data fusion and pattern recognition.

[0037] The data fusion algorithm achieves dimensionality reduction and feature selection through precise Principal Component Analysis (PCA) and Independent Component Analysis (ICA). Specifically, in the PCA stage, the algorithm first calculates the eigenvalues ​​and eigenvectors of the covariance matrix, then sorts them according to the magnitude of the eigenvalues ​​and selects the top 5 principal components. These components have a cumulative contribution rate exceeding 90%, ensuring that the main information of the dataset is preserved. In the ICA stage, the algorithm initializes the weight matrix and iteratively optimizes it until the change in independence indices (such as mutual information) is less than 0.0001, ensuring the stability and reliability of the independent components. The number of ICA iterations is typically limited to no more than 1000 to avoid overfitting. By combining PCA and ICA, the data fusion algorithm not only significantly reduces the dimensionality of the data but also accurately extracts features crucial for mineral deposit exploration, providing high-quality input data for subsequent pattern recognition and geological model construction.

[0038] In the pattern recognition and analysis steps of our deep concealed mineral deposit detection method, we meticulously designed and implemented a series of machine learning models to achieve efficient identification of mineralization features in geological data. The Deep Belief Network (DBN) consists of five stacked Restricted Boltzmann Machines (RBMs), each containing 500 neurons. It was pre-trained using a contrastive divergence algorithm with a learning rate of 0.01, and then fine-tuned to optimize the extraction of mineralization features. The Convolutional Neural Network (CNN) consists of three convolutional layers and two fully connected layers. The convolutional layers use 3x3 kernels with a stride of 1. Each convolutional layer is followed by a max-pooling layer to reduce feature dimensionality and enhance feature invariance. The ReLU activation function is used to introduce non-linearity, with a learning rate of 0.001. A recurrent neural network (RNN), specifically a long short-term memory network (LSTM), consisting of 3 layers with 128 neurons each, was used to process geological time-series data, capturing the long-term dependencies of geological events. The time step was set to 5, and the learning rates for the forget gate and input gate were set to 0.0005. All models used cross-entropy as the loss function and were trained using the Adam optimizer, with β1 and β2 parameters set to 0.9 and 0.999, respectively, to adaptively adjust the learning rate.

[0039] In the three-dimensional geological model construction step of our deep concealed mineral deposit detection method, we employed geostatistical methods, including kriging and simulated annealing, to significantly improve the model's prediction accuracy. In kriging, we set parameters for the semivariogram function, such as a range parameter of 500 meters, a sill parameter set to 1.5 times the data coefficient of variation, and a baseline effect parameter set to 5% of the sill effect, to accurately capture the correlation of spatial data. The simulated annealing algorithm's parameter settings included an initial temperature of 1000 degrees Celsius, followed by a gradual decrease in temperature with a cooling factor of 0.98, ensuring the search process traversed a sufficient number of solution spaces and avoided getting trapped in local optima. The probability of accepting a new solution was determined by the Metropolis criterion, where the temperature parameter and energy difference jointly determined the acceptance probability.

[0040] In the verification and optimization phase, we adopted a series of specific exploration data collection and analysis measures. First, the drilling core analysis depth was set at 1500 meters, with a core recovery rate target of 95%, ensuring the comprehensiveness and continuity of subsurface geological information. During drilling, detailed core description and sampling were conducted every 10 meters, including recording rock type, mineral composition, structure, and mineralization characteristics. Second, geophysical logging data collection covered parameters such as resistivity, natural gamma, density, and neutron porosity, with the measurement accuracy of each parameter controlled within 1% to ensure high-precision comparison with the electrical characteristics of the geological model. Furthermore, geochemical sampling results were analyzed to determine the content of key elements such as mercury, copper, lead, and zinc in soil, water, and rock samples, achieving an accuracy at the ppb level. For example, the accuracy of mercury content analysis was 1 ppb, and the accuracy of copper content analysis was 5 ppb, to identify geochemical anomalies. Through this detailed exploration data and precise analysis, we were able to rigorously verify and specifically optimize the geological model, thereby significantly improving the accuracy and reliability of mineral deposit detection.

[0041] The system implemented in this invention has a series of precisely configured modules to ensure the efficiency and accuracy of the entire detection process. The system includes a data acquisition module for collecting geological, geophysical, geochemical, and remote sensing data; a data preprocessing module for data cleaning, standardization, and outlier removal, with data calibration accuracy controlled within ±0.5% and outlier removal criteria set at ±3 standard deviations above the mean; a data fusion module using PCA to retain components explaining more than 90% of the variance, with the ICA independence index change threshold set at 0.0001; and a pattern recognition module using algorithms such as DBN, CNN, and RNN, where the CNN contains 3 convolutional layers. The model incorporates a layered structure and two fully connected layers, with each layer of the LSTM network containing 128 neurons. All models utilize the Adam optimizer with a learning rate set to 0.001. The 3D geological modeling module sets the kriging range parameter to 500 meters, the nugget parameter to 1.5 times the data coefficient of variation, and the simulated annealing algorithm's initial temperature to 1000°C with a cooling factor of 0.98. The validation and optimization module, combined with core analysis from a 1500-meter depth, maintains geophysical logging data accuracy within 1% and geochemical analysis accuracy to 1 ppb. This tight integration and precise parameter settings enable full automation and optimization from data acquisition to model validation, significantly improving the efficiency and accuracy of mineral deposit exploration.

[0042] This invention provides a highly specialized software package containing precisely written code capable of automating a series of complex operations, including data acquisition, preprocessing, data fusion, pattern recognition, 3D geological modeling, and verification and optimization. The software code employs a modular structure and is implemented in efficient programming languages ​​such as C++ or Python, ensuring algorithm execution efficiency and cross-platform compatibility. The program provides a user interface supporting at least 1024x768 resolution displays, ensuring clear visuals on various devices. The software is stored on computer-readable media, such as DVDs (capacity no less than 4.7GB), hard drives (speed no less than 7200RPM), or solid-state drives (write speed no less than 500MB / s), to accommodate different users' storage and retrieval needs. During installation, the software provides a detailed installation wizard, supports at least Windows 10 and Linux Ubuntu 20.04 operating systems, and requires at least 4GB of RAM, ensuring stable operation. The package also includes a user manual of no less than 200,000 words, detailing the software's installation, configuration, operation procedures, and troubleshooting guide. The online help documentation provides answers to at least 1,000 frequently asked questions, ensuring that users can quickly resolve any problems they encounter while using the product.

[0043] The system of this invention is equipped with a powerful user interface that allows users to input geological parameters, select analysis methods, and view and adjust the results of the 3D geological model in real time through a concise and clear graphical user interface (GUI). The user interface provides parameter input fields, including but not limited to a rock type selector, a mineral composition input box (supporting input of up to 100 minerals), stratigraphic time stamps (accurate to millions of years), and a detailed description area for geological structure data. Users can select data preprocessing algorithms via drop-down menus. For example, when selecting Principal Component Analysis (PCA), users can choose the number of components to retain (e.g., the first 5 principal components), or set the independence index change threshold (e.g., 0.0001) in Independent Component Analysis (ICA). In the pattern recognition module, users can select different machine learning models, such as Deep Belief Networks (DBNs) and Convolutional Neural Networks (CNNs), and set the number of network layers and neurons. The number of convolutional layers in a CNN can be set to 3 to 5, and the kernel size of each layer can be set to 3x3 or 5x5. The user interface also includes a real-time results display area that shows rendered images of the 3D geological model, supporting a resolution of at least 1024x768 to ensure clear visibility of model details. Model adjustment tools allow users to adjust kriging parameters via sliders; for example, the range can be adjusted from 100 meters to 1000 meters, and the si-ll and nugget effect parameters can be adjusted using percentage sliders. The user interface also features detailed logging, recording each user action and model status, supporting historical data review and version comparison. Data output supports exporting results as PDF reports, CSV data tables, and DXF drawings, ensuring users can easily share and further analyze the results.

[0044] This invention provides a geological exploration decision support tool that, based on prediction results, employs a quantitative decision-making algorithm to provide priority ranking and resource allocation suggestions for mineral deposit exploration. This tool comprehensively considers multi-dimensional characteristics of geological data, such as probability values ​​of geological models (e.g., a mineralization probability threshold set to exceed 70%), the intensity of geochemical anomalies (e.g., mercury content exceeding 1 ppb), resistivity anomalies from geophysical measurements (set to be 20% lower than the surrounding background value), as well as exploration costs and expected economic benefits. It automatically evaluates the exploration value of each potential mineral deposit through a weighted scoring system.

[0045] The decision support tool uses Monte Carlo simulations to assess risk and uncertainty, with the number of simulations set to 1000 to ensure the reliability of the statistical results. It also incorporates multi-objective optimization algorithms, such as genetic algorithms, with 100 iterations, crossover rate, and mutation rate set to 0.8 and 0.01, respectively, to find the optimal resource allocation scheme.

[0046] The user interface allows exploration teams to adjust parameters based on actual needs, such as adjusting the weighting of exploration costs or the discount rate of expected returns, ensuring that recommended solutions align with actual exploration strategies. The tool also provides a real-time feedback mechanism, dynamically updating recommendations based on exploration progress and market changes.

[0047] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting deep, concealed mineral deposits, characterized in that, The method includes the following steps: S1. Geological Data Acquisition: Collect and obtain key geological information on rock types, mineral composition, stratigraphic structure, and igneous rocks in the target area through data collection, field surveys, and sampling; S2. Geophysical Data Acquisition: Using wide-area electromagnetic method technology, geophysical parameters such as resistivity and electromagnetic wave propagation characteristics of subsurface media are measured to invert deep geological features; S3. Geochemical Data Acquisition: Using hydrocarbon-mercury superimposed halo analysis technology, collect hydrocarbon mineral assemblages, mercury content, and geochemical indicators such as gold, copper, lead, and zinc in soil and rocks; S4. Remote sensing data acquisition: Using airborne hyperspectral remote sensing technology, acquire spectral characteristic data of the target area to identify and analyze mineral types, surface locations, extent and distribution characteristics; S5. Data Preprocessing: Cleaning, standardizing, and denoising the collected geological, geophysical, geochemical, and remote sensing data to improve data quality and consistency; S6. Multi-source data fusion: Develop a data fusion algorithm to extract features and perform multi-scale analysis on preprocessed data, and integrate them into a unified dataset; S7. Pattern Recognition and Analysis: Using machine learning algorithms to perform pattern recognition on the fused data, analyze mineralization characteristics and abnormal areas, wherein the machine learning algorithms include support vector machines, random forests or deep learning networks; S8. Construction of 3D Geological Model: Based on the analysis results of mineralization characteristics and anomalous areas, a 3D geological model of the target area is constructed using geological modeling software to predict the spatial distribution of mineral deposits; S9. Verification and Optimization: Verify the predicted mineral deposit locations in the field through drilling or geophysical verification methods, and adjust and optimize the model based on the verification results; The data fusion algorithm specifically includes principal component analysis (PCA) and independent component analysis (ICA) techniques, used for data dimensionality reduction and feature selection; In the pattern recognition and analysis step, the deep learning network includes a deep belief network (DBN), a convolutional neural network (CNN), or a recurrent neural network (RNN); The three-dimensional geological model construction step also includes using geostatistical methods to improve the prediction accuracy of the model, including Kriging or simulated annealing algorithm; In the verification and optimization steps, the actual exploration data includes drilling core analysis, geophysical logging data, and geochemical sampling results. The optimized implementation of the data fusion algorithm is achieved through precise principal component analysis (PCA) and independent component analysis (ICA) techniques. PCA is performed by calculating the covariance matrix Σ of the original dataset X, and then performing eigenvalue decomposition to obtain the eigenvector v. i and eigenvalues ​​λ i We select the k eigenvectors with the largest eigenvalues ​​(k=10) to form the transformation matrix W, thus projecting the original data into a new space while retaining principal components with a variance contribution rate exceeding 85%. The mathematical expression is: W=[v1,v2,...,v...] k ], X new =XW; In the ICA phase, the algorithm iteratively maximizes the independence of independent source signals with non-Gaussian distributions, using negative entropy or kurtosis as the independence measure, and sets the convergence criterion as a change in the independence index less than XW. ; In the pattern recognition and analysis steps, the models used include Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), or Recurrent Neural Networks (RNN). DBNs consist of multiple stacked Restricted Boltzmann Machines (RBMs), each containing 1000 neurons. They are pre-trained using unsupervised learning of features, followed by supervised fine-tuning to automatically extract complex patterns from the data. CNN models contain 5 convolutional layers and 3 fully connected layers, using ReLU activation and cross-entropy loss functions. The kernel size is set to 3x3 with a stride of 1 to capture local spatial features and perform effective feature mapping. RNNs are Long Short-Term Memory (LSTM) networks, containing 3 layers with 128 neurons each, used to process sequential data and capture long-term dependencies in time series. All models use the Adam optimizer for parameter updates, with a learning rate set to 0.

001. Overfitting is controlled using early stopping; training stops when the performance improvement on the validation set is less than 0.

01.

2. The method according to claim 1, characterized in that, The data preprocessing steps further include data calibration and outlier removal to ensure the accuracy and reliability of the data.

3. A system for implementing the method according to any one of claims 1 to 2, characterized in that, The system includes the following modules: data acquisition module, data preprocessing module, data fusion module, pattern recognition module, 3D geological modeling module, and verification and optimization module.

4. The system according to claim 3, characterized in that, The system also includes a user interface that allows users to input geological parameters, select analysis methods, view model results, and make model adjustments.

5. A computer program product, characterized in that, The software includes software code for performing the method of claim 1, the code being stored on a computer-readable medium, including an optical disc, a hard disk, or a solid-state drive.

6. A decision support tool for geological exploration, characterized in that, Based on the prediction results obtained by using the method described in claim 1, priority ranking and resource allocation suggestions for mineral deposit exploration are provided.

Images