A method and system for predicting target areas of a shallow coverage area gold mine

Abstract

The application relates to the technical field of intelligent exploration, in particular to a shallow coverage area gold mine target area prediction method and system, which comprises the following steps: inputting multi-source ore point data of a to-be-predicted area into a gold mine target area prediction model to obtain corresponding gold mine target area prediction results; the gold mine target area prediction model is trained through the following steps: acquiring first ore point data corresponding to a first ore point and second ore point data corresponding to a second ore point; based on the first ore point data, acquiring a first Au average value and a correlation coefficient corresponding to the first ore point, and based on the second ore point data, acquiring a first Au average value and a correlation coefficient corresponding to the second ore point; based on the first Au average value and the correlation coefficient, adjusting the first ore point data and the second ore point data, and based on the adjusted first ore point data and the second ore point data, obtaining a gold mine target area prediction model for predicting a gold mine target area in a shallow coverage area. The method effectively supplements the total amount of samples and significantly reduces the risk of target area misjudgment and misjudgment.

Description

Technical Field

[0001] This application relates to the field of intelligent exploration technology, and in particular to a method and system for predicting gold ore target areas in shallow overburden areas. Background Technology

[0002] Gold resources, as a vital mineral resource, play an irreplaceable role in technological development, financial security, high-end manufacturing, and the jewelry industry, significantly contributing to regional economic development and employment. With the increasing depletion of surface outcrops and the shrinking space for traditional prospecting, shallowly covered areas (areas where bedrock is covered by thin layers of sediment, soil, or vegetation) have become key target areas for finding new deposits. While these areas present greater exploration challenges, they often possess superior mineralization geological conditions and enormous prospecting potential. Therefore, developing efficient and accurate gold deposit prediction and exploration technologies suitable for shallowly covered areas is of great strategic significance for overcoming resource bottlenecks, expanding gold resource reserves, and ensuring resource security.

[0003] Current gold ore target area prediction technologies often employ deep learning algorithms such as convolutional neural networks, multilayer perceptrons, and multimodal fusion models. These models are trained by inputting multi-source data from the target area and outputting mineralization probability or potential classification results as the basis for target area delineation. However, this approach relies on local data from a single mining area for model training. When the number of known mineral deposits in the target mining area is small or the sample size is insufficient, the model is prone to overfitting. Although some methods attempt to introduce cross-domain data from other mining areas to supplement the sample, they fail to specifically adjust the core features of the cross-domain data. This results in distribution shifts in data from different mining areas due to subtle differences in the mineralization environment (such as mineralization intensity and element migration efficiency). Cross-domain data not only fails to supplement the data but also introduces "negative migration" interference, reducing the reliability of model predictions. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, this application provides a method and system for predicting gold ore target areas in shallow coverage areas. It solves the technical problems of requiring a large number of training samples and long training time, as well as the tendency to back-transfer, when using deep learning algorithms such as convolutional neural networks, multilayer perceptrons, and multimodal fusion models.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the main technical solutions adopted in this application include:

[0008] In a first aspect, embodiments of this application provide a method for predicting gold target areas in shallow-covered areas, including:

[0009] Input the multi-source mineral point data of the area to be predicted into the pre-trained gold mine target area prediction model to obtain the corresponding gold mine target area prediction results.

[0010] The gold mine target area prediction model is trained through the following steps:

[0011] Acquire the first mineral point data corresponding to the first mineral point and the second mineral point data corresponding to the second mineral point, and the mineralization mechanism and data acquisition specifications of the two different mineral points are consistent; the first mineral point data and the second mineral point data both include Au content and mineral point related data corresponding to multiple different sampling points;

[0012] Based on the data from the first mining site, the mean value of Au and the correlation coefficient corresponding to the first mining site are obtained. Based on the data from the second mining site, the mean value of Au and the correlation coefficient corresponding to the second mining site are obtained. The correlation coefficient represents the degree of correlation between Au content and relevant data of the mining site.

[0013] Based on the first Au mean and correlation coefficient corresponding to the first and second mining sites, the data of the first and second mining sites are adjusted to obtain the adjusted data of the first and second mining sites.

[0014] The gold target area prediction model was trained based on the adjusted data from the first and second mining sites to obtain a gold target area prediction model for shallow-covered areas.

[0015] Optionally, in one specific embodiment, the mineral deposit-related data includes As content and distance from the fault zone;

[0016] The correlation coefficient includes: the first correlation coefficient and the second correlation coefficient;

[0017] Based on the data from the first mining site, the mean Au value and correlation coefficient for the first mining site are obtained. Based on the data from the second mining site, the mean Au value and correlation coefficient for the second mining site are obtained, including:

[0018] Feature transformations are performed on the data from the first and second mineral deposits. The Au and As contents are transformed using a pre-defined formula (Formula 1), and the distance to the fault zone is transformed using a Z-score. Formula 1 is as follows:

[0019] x1=log 10 (x0+1);

[0020] Where x0 is the parameter value before feature transformation, and x1 is the parameter value after feature transformation;

[0021] Based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first Au mean value corresponding to the first and second mining sites is obtained.

[0022] Based on the first mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the first mining site are constructed. Based on the second mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the second mining site are constructed.

[0023] The first correlation coefficient represents the degree of correlation between Au content and As content, while the second correlation coefficient represents the degree of correlation between Au content and distance from the fault zone.

[0024] Optionally, in a specific embodiment, the mineral deposit related data further includes slope and Hg content in the soil; the correlation coefficient also includes a third correlation coefficient;

[0025] Based on the data from the first mining site, the mean Au value and correlation coefficient corresponding to the first mining site are obtained. Based on the data from the second mining site, the mean Au value and correlation coefficient corresponding to the second mining site are obtained. This also includes:

[0026] Using Formula 1, the soil Hg content at each sampling point in the first and second mining site data is transformed using a feature transformation, and the slope at each sampling point in the first and second mining site data is transformed using the Z-score.

[0027] Based on the soil Hg content and slope corresponding to each sampling point in the first and second mining site data after feature transformation, and using a pre-set formula two, the Hg correction data corresponding to each sampling point is obtained; the formula two is:

[0028] Hg 校正 =Hg 检测 ×(1+k×(Slope-5°));

[0029] Among them, Hg 检测 The Hg content in the soil corresponding to the sampling point. 校正 Here, Hg is the Hg correction data corresponding to the sampling point, Slope is the slope corresponding to the sampling point, and k is the preset slope correction parameter;

[0030] Based on the Hg-corrected data and Au content corresponding to each sampling point in the first and second mining site data, the third correlation coefficients corresponding to the first and second mining sites are obtained respectively; the third correlation coefficient is the degree of correlation between Au content and Hg-corrected data.

[0031] Optionally, in a specific embodiment, the data of the first and second mining sites are adjusted based on the first mean Au value and correlation coefficient corresponding to the first and second mining sites, including:

[0032] The average value of the first Au corresponding to the first mining site and the second mining site is averaged to obtain the average value of the second Au corresponding to the first mining site and the second mining site.

[0033] Based on the first and second mean Au values ​​corresponding to the first and second mining sites, Au adjustment parameters are obtained for the first and second mining sites, respectively; wherein, the Au adjustment parameter is the second mean Au divided by the first mean Au.

[0034] Based on the Au adjustment parameter corresponding to the first mining site, the Au content of each sampling point in the first mining site is adjusted. Based on the Au adjustment parameter corresponding to the second mining site, the Au content of each sampling point in the second mining site is adjusted. The adjusted Au content is the product of the Au adjustment parameter and the Au content.

[0035] Based on the correlation coefficient corresponding to the first mining site and the adjusted Au content of each sampling point in the first mining site, the correlation coefficient corresponding to each sampling point in the first mining site is adjusted. Based on the correlation coefficient corresponding to the second mining site and the adjusted Au content of each sampling point in the second mining site, the correlation coefficient corresponding to each sampling point in the second mining site is adjusted.

[0036] Optionally, in a specific embodiment, based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first Au mean value corresponding to the first and second mining sites is obtained, including:

[0037] Based on the Au content corresponding to multiple sampling points in the first mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the first mining site is obtained.

[0038] Based on the Au content corresponding to multiple sampling points in the second mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the second mining site is obtained.

[0039] The first Au mean is the value corresponding to the 50th percentile.

[0040] Optionally, in a specific embodiment, the gold ore target area prediction model is trained based on the adjusted first and second ore point data to obtain a gold ore target area prediction model for predicting gold ore target areas in shallow-covered areas, including:

[0041] Mineral point labels are added to the adjusted first and second mineral point data to mark the corresponding mineral point labels in the first and second mineral point data.

[0042] Input the first and second mineral point data after marking the mineral point into the gold mine target area prediction model to obtain the prediction data corresponding to the first and second mineral points.

[0043] Based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the model parameters of the gold ore target area prediction model are adjusted to obtain a gold ore target area prediction model for shallow cover areas.

[0044] Optionally, in a specific embodiment, the model parameters of the gold ore target area prediction model are adjusted based on the predicted data corresponding to the first and second ore points, the ore point labels, and the second Au mean, including:

[0045] Based on the predicted data, mining point labels, and second Au average value corresponding to the first and second mining points, the corresponding total loss value is obtained.

[0046] Based on the total loss value and the pre-set BP algorithm, the model parameters of the gold mine target area prediction model are adjusted.

[0047] Optionally, in one specific embodiment, the prediction data includes a predicted ore-bearing probability and a predicted grade;

[0048] Based on the predicted data, mining point labels, and the second Au mean value corresponding to the first and second mining points, the corresponding total loss value is obtained, including:

[0049] Based on the predicted mineral-bearing probabilities and mineral-bearing labels corresponding to the first and second mineral deposits, and using a pre-set formula (Formula 3), the corresponding classification loss is obtained; formula 3 is:

[0050] ;

[0051] Where L1 is the classification loss, y1 is the mineral point label corresponding to the first mineral point, and y2 is the mineral point label corresponding to the second mineral point. This represents the predicted ore-bearing probability corresponding to the first ore deposit. This represents the predicted ore-bearing probability corresponding to the second mineral deposit.

[0052] Based on the predicted grades and the average Au values ​​of the first and second mining sites, and using a pre-set formula (Formula 4), the corresponding grade regression loss is obtained; formula 4 is:

[0053] ;

[0054] Where L2 represents the grade regression loss, G1 represents the second mean Au value corresponding to the first ore deposit, and G2 represents the second mean Au value corresponding to the second ore deposit. The predicted grade corresponding to the first mining site. The predicted grade corresponding to the second mining site;

[0055] The classification loss and grade regression loss are weighted and summed to obtain the corresponding total loss value.

[0056] Optionally, in a specific embodiment, obtaining the first mining point data corresponding to the pre-collected first mining point and the second mining point data corresponding to the second mining point includes:

[0057] Acquire the first mineral point data corresponding to the first mineral point and the second mineral point data corresponding to the second mineral point, and the mineralization mechanism and data acquisition specifications of the two different mineral points are consistent;

[0058] Based on the pre-set WGS84 coordinate system, the sampling coordinates of all sampling points are unified and resampled into a 10m×10m grid to match the sampling point space.

[0059] Secondly, embodiments of this application provide a shallow-covered gold ore target area prediction system, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the above-described shallow-covered gold ore target area prediction method.

[0060] (III) Beneficial Effects

[0061] This application presents a method for predicting gold ore target areas in shallow-covered areas. By defining a first and a second ore deposit, and using the correlation coefficient between the mean Au value, Au content, and relevant data of the ore deposits as quantitative criteria, the method makes targeted adjustments to cross-regional ore deposit data. This eliminates the interference of distributional offsets in data from different ore areas, effectively supplements the total sample size, and provides a high-quality data foundation for model training. The gold ore target area prediction model trained based on this data can more accurately capture weak mineralization signals in shallow-covered areas, significantly reduce the risk of missed or misjudged target areas, improve prediction accuracy and reliability, help focus on high-potential areas for exploration work, and reduce ineffective investment. Attached Figure Description

[0062] Figure 1 A flowchart of the training process for the gold mine target area prediction model provided in this application embodiment;

[0063] Figure 2 A flowchart illustrating the data adjustment process for the first / second mining site provided in this application embodiment. Detailed Implementation

[0064] To better explain and facilitate understanding of this application, the following detailed description of the application is provided in conjunction with the accompanying drawings and specific embodiments.

[0065] Gold mines, as important mineral resources, play an irreplaceable role in technological development, financial security, high-end manufacturing, and the jewelry industry. With the depletion of surface outcrops, shallowly covered areas have become the core target areas for gold prospecting. Developing efficient and accurate predictive exploration technologies is of great significance to safeguarding national resource security. Existing gold target area prediction methods mostly use deep learning algorithms, which rely on training with data from a single mining area, leading to overfitting when the sample size is insufficient. Introducing cross-domain data can cause distribution shifts and "negative migration" problems due to a lack of targeted feature adjustments, reducing prediction reliability. This application provides a method for predicting gold target areas in shallowly covered areas. By limiting cross-domain mining points with consistent mineralization mechanisms and data collection standards, and using the mean Au and correlation coefficient as quantitative criteria, the cross-domain data is targeted for adjustment. This effectively eliminates distribution shift interference, supplements the total sample size, and ensures data quality. The model trained based on this data can accurately capture weak mineralization signals, reduce the risk of missed or misjudged detections, improve prediction accuracy and exploration efficiency, and reduce ineffective investment.

[0066] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0067] This application provides a method for predicting gold target areas in shallowly covered areas, including:

[0068] Input the multi-source mineral point data of the area to be predicted into the pre-trained gold mine target area prediction model to obtain the corresponding gold mine target area prediction results.

[0069] like Figure 1 As shown, the gold mine target area prediction model is trained through the following steps:

[0070] S1. Obtain the first mineral point data corresponding to the first mineral point and the second mineral point data corresponding to the second mineral point, and the mineralization mechanism and data collection specifications of the two different mineral points are consistent; the first mineral point data and the second mineral point data both include Au content and mineral point related data corresponding to multiple different sampling points;

[0071] S2. Based on the data from the first mining site, obtain the first mean Au value and correlation coefficient corresponding to the first mining site; based on the data from the second mining site, obtain the first mean Au value and correlation coefficient corresponding to the second mining site; the correlation coefficient is the degree of correlation between Au content and relevant data of the mining site.

[0072] S3. Based on the first Au mean and correlation coefficient corresponding to the first and second mining sites, adjust the data of the first and second mining sites to obtain the adjusted data of the first and second mining sites.

[0073] S4. Based on the adjusted data of the first and second mining sites, the gold target area prediction model is trained to obtain a gold target area prediction model for shallow-covered areas.

[0074] This embodiment provides a method for predicting gold deposit target areas in shallow-covered areas. By limiting the first and second mining points to the premise of "consistent mineralization mechanisms and standardized data collection," and using the correlation coefficient between the mean Au value and Au content and relevant data of the mining points as dual quantitative criteria, targeted adjustments are made to the cross-regional mining point data. This eliminates the interference of distributional offsets between different mining areas from the root, avoids the "negative migration" problem, effectively supplements the total sample size, alleviates overfitting caused by insufficient samples from a single mining area, and completely preserves the real mineralization correlation patterns in the data adjustment process, avoiding feature distortions from traditional preprocessing, and providing a high-quality data foundation for model training. The gold deposit target area prediction model trained based on this data can more accurately capture weak mineralization signals in shallow-covered areas, significantly reduce the risk of missed or misjudged target areas, improve prediction accuracy and reliability, help focus on high-potential areas for exploration work, reduce ineffective investment, and effectively improve the exploration efficiency and success rate of concealed gold deposits in shallow-covered areas.

[0075] Optionally, in one specific embodiment, the mineral deposit-related data includes As content and distance from the fault zone;

[0076] The correlation coefficient includes: the first correlation coefficient and the second correlation coefficient;

[0077] Then, based on the data from the first mining site, the mean value of the first Au and its correlation coefficient corresponding to the first mining site are obtained; based on the data from the second mining site, the mean value of the first Au and its correlation coefficient corresponding to the second mining site are obtained, including:

[0078] Feature transformations were performed on the data from the first and second mineral deposits. The Au and As contents were transformed using a pre-defined formula (Formula 1), and the distance to the fault zone was transformed using the Z-score. Formula 1 is as follows:

[0079] x1=log 10 (x0+1);

[0080] Where x0 is the parameter value before feature transformation, and x1 is the parameter value after feature transformation;

[0081] Based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first Au mean value corresponding to the first and second mining sites is obtained.

[0082] Based on the first mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the first mining site are constructed. Based on the second mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the second mining site are constructed.

[0083] The first correlation coefficient represents the degree of correlation between Au content and As content, while the second correlation coefficient represents the degree of correlation between Au content and distance from the fault zone.

[0084] Furthermore, the relevant data for the mining site also include slope and Hg content in the soil; the correlation coefficient also includes the third correlation coefficient;

[0085] Then, based on the data from the first mining site, the mean value of the first Au and the correlation coefficient corresponding to the first mining site are obtained; based on the data from the second mining site, the mean value of the first Au and the correlation coefficient corresponding to the second mining site are obtained; and the process also includes:

[0086] Formula 1 is used to perform feature transformation on the Hg content in the soil at each sampling point in the data from the first and second mining sites. Z-score is used to perform feature transformation on the slope at each sampling point in the data from the first and second mining sites.

[0087] Based on the soil Hg content and slope corresponding to each sampling point in the first and second mining point data after feature transformation, and using the pre-set Formula 2, the Hg correction data corresponding to each sampling point is obtained; Formula 2 is:

[0088] Hg 校正 =Hg 检测 ×(1+k×(Slope-5°));

[0089] Among them, Hg 检测 The Hg content in the soil corresponding to the sampling point. 校正 Here, Hg is the Hg correction data corresponding to the sampling point, Slope is the slope corresponding to the sampling point, and k is the preset slope correction parameter;

[0090] Based on the Hg-corrected data and Au content corresponding to each sampling point in the data from the first and second mining sites, the third correlation coefficients for the first and second mining sites are obtained respectively; the third correlation coefficient is the degree of correlation between Au content and Hg-corrected data.

[0091] Specifically, data collection focuses on the core objective of "penetrating shallow overburden interference and capturing deep mineralization information," integrating geochemical, deep-penetration, geological structural, topographical, and historical exploration data to ensure the data's relevance, completeness, and spatial correlation, including but not limited to:

[0092] Geochemical penetration core data, taking into account the special characteristics of shallow overburden areas (overburden thickness is usually 550m, mainly Quaternary loose deposits), focuses on collecting "deep-penetrating" samples that can reflect the mineralization information of the underlying bedrock, avoiding human pollution or weathering interference of the surface soil.

[0093] Fine-grained soil sample:

[0094] Collection specifications: At each sampling point, dig a soil profile to a depth of 1.52m (penetrating the topsoil / oxidized layer). Collect 500-1000g of soil from the lower part of the profile (30-50cm above the contact zone between the overburden and bedrock). Extract the <2μm clay particles (this component has a large specific surface area and a strong adsorption capacity for elements such as Au, As, Sb, and Hg migrated from deep ore-forming fluids, making it the core carrier for gold exploration in shallow overburden areas) through centrifugation and sedimentation.

[0095] Collection density: Adjusted according to terrain complexity, every 12km in flat areas. 2 One sampling point will be set up, with the density increased to every 0.5 km in mountainous / fault-developed areas. 2 One sampling point will be set up to ensure coverage of key areas such as the ore-controlling structural zone and the water system confluence area;

[0096] Detection elements: In addition to the target element Au content (detection limit must reach 0.1 ppb), the content of semi-natural indicator elements As and Hg content must be detected simultaneously to reflect the gold mineralization environment;

[0097] In addition, SiO2 and Al2O3 can be tested to distinguish the type of soil parent material; La, Ce and Nd content can be tested to indicate the intensity of ore-forming hydrothermal activity.

[0098] Soil gas measurement data:

[0099] Collection procedures: A soil gas collection well, 81.2m deep (to avoid interference from the surface atmosphere), was installed next to the soil sampling point. A sealed gas pump was used to extract soil gas, and the Hg content (expressed as concentration) in the soil was detected by gas chromatography-mass spectrometry (GC-MS). The detection limit was 0.01 ng / m³. 3 Alternatively, an electrostatic collection radon meter can be used to detect Rn concentration (unit: Bq / m³). 3 (This is supplementary data.)

[0100] Collection time: Select a sunny and windless day (8:00-10:00) to collect data, avoiding gas concentration fluctuations caused by diurnal temperature differences and precipitation. Collect data three times at each point and take the average value as the final data.

[0101] Significance of the data: The Hg content in the soil is a characteristic volatile element of gold deposits. Hg in deep mineralized bodies can migrate upwards along fault zones and fissures to the soil, forming anomalies.

[0102] Plant geochemical data:

[0103] Sample selection: Prioritize deep-rooted plants that are dominant in the study area (such as pine, poplar, and sea buckthorn, whose root systems can reach a depth of 25m and can penetrate the cover layer to obtain deep elements), and avoid shallow-rooted herbaceous plants or contaminated crops.

[0104] Collection specifications: Collect the same parts (such as the current year needles of pine trees and the leaves of poplar trees) from 35 healthy plants around each sampling point, wash and dry them, crush them, and use inductively coupled plasma mass spectrometry (ICPMS) to detect the Au and As content.

[0105] Data correction: The elemental content was standardized according to the plant dry weight (unit: ng / g dry weight) to eliminate the influence of different plant growth cycles and water content on the data.

[0106] Groundwater chemical data:

[0107] Collection scope: Samples were collected from groundwater outcrops such as wells, springs, and stream sources in the study area, with at least 3 samples collected from each hydrological unit;

[0108] Detection indicators: In addition to dissolved Au content (detection limit 0.05ppb) and As content, pH value (mineralized fluids are mostly weakly acidic, pH 5.0-6.5), redox potential (Eh, gold mineralization mostly occurs in reducing environments, Eh < 0mV), and total dissolved solids (TDS, reflecting the intensity of water-rock interaction) can be detected simultaneously to supplement and correct the Au content, As content, and Hg content in the soil detected above.

[0109] Geological structure and topographic data:

[0110] The geological structural data source is based on the interpretation of high-resolution remote sensing imagery (such as Sentinel's 10m resolution imagery and unmanned aerial vehicle (UAV) 5m resolution imagery), combined with 1:50,000 geological mapping data, providing two types of structural information. Specifically, ore-controlling structures include regional large faults (length > 10km, trending in line with the regional metallogenic belt) and secondary faults (length 25km, mostly ore-fluid transport channels), with the fault's strike, dip, dip angle, and fracture zone width marked; ore-bearing structures include the axial and limb turning points of fold structures (easily forming fracture spaces, providing sites for gold deposits), with fold type (anticline / syncline) and core lithology marked; and fault zone data is used to obtain distances to fault zones.

[0111] Topographic data: Data source, extracted based on digital elevation model (DEM, 30m resolution, from ASTER GDEM or UAV LiDAR); core indicators are: slope: distinguishing between steep slopes (>25°, elements are easily lost) and gentle slopes (5°-25°, elements are easily deposited).

[0112] Historical exploration and known mineral deposit data (labeling basis), data collection, obtaining information on known gold deposits in the study area and surrounding areas through geological exploration reports and mineral resource reserve databases, must include:

[0113] Spatial location: accurate to latitude and longitude (error <10m), marking the outcrop location or borehole location of the mineralization point; mineralization attributes, mineralization type (quartz vein type, porphyry type, Carlin type, etc., different types have large differences in element combination), ore grade (Au content, unit g / t, distinguishing high-grade ore (>5g / t), medium-grade ore (15g / t), low-grade ore (0.31g / t)), ore body thickness (m) and reserve scale (small / medium / large); exploration methods, marking the discovery method of the mineralization point (surface outcrop, borehole, geophysical and geochemical anomaly verification) to ensure data reliability; labeling, adopting a two-class labeling system: two-class labeling, marking samples within a known mineralization point and a 500m radius (mineralization halo influence range) as "1" (mineralized), and marking other samples without mineralization records as "0" (non-mineralized).

[0114] The above data undergoes preprocessing to eliminate noise interference and optimize data quality. Preprocessing requires differentiated schemes tailored to the characteristics of different data types (such as the log-normal distribution of geochemical data and the volatility of soil gas data). The core objective is to eliminate missing values, outliers, and the influence of dimensions to ensure that the data meets the input requirements of deep learning models.

[0115] Data cleaning: accurately identify and handle data anomalies; missing value handling: select the corresponding strategy according to the missing value type to avoid deviations caused by simple data filling; outlier identification and correction: distinguish between data anomalies (measurement errors) and geological anomalies (mineralization signals) to avoid accidentally deleting valid information.

[0116] Univariate statistical identification (outlier identification): 3σ principle: For Zhengtai distribution data, samples exceeding "mean ± 3 × standard deviation" are marked as outliers; Interquartile Range (IQR): For log-normal distribution data, calculate IQR = Q3Q1, and mark samples exceeding Q3 + 1.5 × IQR or Q1 - 1.5 × IQR as outliers.

[0117] Spatial correlation verification: Local anomaly factor (LOF) is calculated as the ratio of the local density of each sample to the density of surrounding samples. Samples with LOF > 1.5 are considered spatial anomalies (possibly mineralization anomalies). Spatial autocorrelation analysis (Moran's I) is performed to analyze whether the samples within a 500m radius of the candidate anomaly show "high value clustering" (Moran's I > 0.3, indicating a geological anomaly, which is retained) or "isolated high values" (Moran's I < 0, indicating a data anomaly, which is corrected).

[0118] Outlier handling: For data anomalies, use "neighboring sample replacement" (e.g., if a sample has an Au content of 1000 ppb, which is much higher than the 110 ppb of surrounding samples, it is determined to be a detection error and replaced with the average of the 5 surrounding samples); Geological anomalies: retain and mark (e.g., if a sample has an Au content of 50 ppb, while surrounding samples mostly have 51.5 ppb and are located near a fault zone, it is determined to be a mineralization anomaly and the original data is retained).

[0119] Data standardization and transformation were performed to unify the data and adapt it to model assumptions. Data transformation (for non-normally distributed data): a logarithmic transformation was performed on the content of elements such as Au, As, and Hg in the soil that exhibit a log-normal distribution, using x1=log 10 The (x0+1) transformation (+1 avoids the fact that x=0 is meaningless) makes the data closer to a normal distribution and reduces the impact of extreme values ​​on the model.

[0120] Data standardization involves standardizing terrain data (slope) and distance to fault zones using Z-score.

[0121] For multimodal data (such as soil element + soil gas + plant data), standardize them by data type (e.g., one group for soil elements, one group for gas data) to avoid interference from the dimensional differences of different data types.

[0122] The data acquisition and preprocessing system in this application focuses on extracting information from deep gold deposits in shallowly covered areas. By organically integrating multi-source deep-penetrating samples (fine-grained soil, soil gas, plants, and groundwater) with geological structure, topography, and historical mining data, and combining targeted cleaning, spatial verification, and standardization strategies, it effectively removes surface interference, preserves real mineralization anomalies, and constructs a high-quality, highly interpretable input dataset. This significantly improves the ability of the gold target area prediction model to identify concealed ore bodies and the reliability of its predictions.

[0123] Furthermore, filtering is performed from a pre-deployed database;

[0124] Alternatively, real-time data collection can be conducted, specifically within the first and second mineral deposits (where the mineralization mechanism is consistent and data collection standards are unified), categorized as "0.5km of densely faulted areas". 2 / point, flat area 1km2 Sampling points are deployed at a density of " / points", with no fewer than 60 effective sampling points deployed at each mining site to ensure coverage of the mineralized zone, transition zone, and background zone; the coordinates of the sampling points are located using the WGS84 coordinate system, and the spatial relationship between the sampling points and the surrounding ore-controlling faults is recorded simultaneously.

[0125] Au and As content: Fine-grained soil samples (< 2μm clay particles) were collected from sampling points at a depth of 1.5-2m. ICP-MS was used for detection, with detection limits of ≤0.1ppb (Au) and ≤1ppb (As), respectively. The original detection values ​​were recorded. Distance from fault zone: Using GIS spatial analysis tools, based on the ore-controlling fault vector data interpreted from the 1:50,000 geological map of the mining area, the straight-line distance from each sampling point to the nearest fault zone was calculated, with units uniformly expressed in meters (m). Slope: Based on 30m resolution ASTER GDEM data, the slope value of the grid where the sampling point is located was extracted using the ArcGIS terrain analysis module, with units uniformly expressed in degrees (°). Hg content in soil: A 1.0m deep sampling well was set up next to the soil sampling point. From 8:30 to 9:30 (the period with minimal environmental interference), air was collected from the soil using a sealed air pump. The Hg content was detected using GC-MS, and the average value of three collections was recorded, with units uniformly expressed in ng / m³. 3 .

[0126] According to formula one, x1=log 10 (x0+1) transforms the original test values ​​of the three types of indicators (x0 is the original value, x1 is the transformed value), where "+1" is used to avoid the original value being meaningless when it is 0. The transformed data conforms to the characteristics of a normal distribution and eliminates the interference of extreme values. Standardization is performed according to the Z-score formula to transform the indicator values ​​into standardized data with mean = 0 and standardization = 1, thus unifying the numerical range of different beam steel indicators.

[0127] The first mean Au value is obtained by taking the Au content data of all sampling points in the first and second mining sites after feature transformation, and then obtaining the first mean Au value to reflect the overall statistical level of Au content in each mining site.

[0128] The first correlation coefficient (Au-As): Calculated using the Pearson correlation coefficient formula based on the Au and As content data after feature transformation, quantifying the degree of linear and synergistic correlation between the two; the second correlation coefficient (Au - distance from the fault zone): Calculated using the Pearson correlation coefficient formula based on the Au content and distance from the fault zone data after feature transformation, quantifying the influence of tectonic factors on Au distribution; the third correlation coefficient (Au-Hg corrected data): First calculated according to formula two (Hg)... 校正 =Hg 检测The Hg correction data is calculated using the formula ×(1+k×(Slope-5°)) (where k is the slope correction parameter, determined by verification of Hg anomalies at known mining sites, with a value range of 0.005-0.01). Then, based on the Au content after feature transformation and the Hg correction data, the Pearson correlation coefficient formula is used to calculate and quantify the degree of mineralization correlation between Hg and Au after correcting for topographic interference.

[0129] For example: Given that 80 sampling points were deployed at the first mining site and 60 sampling points at the second mining site, and that the mineralization mechanism (quartz vein type, fracture-controlled ore) and data acquisition specifications (sampling depth, detection method) are completely consistent, some sampling point data are as follows:

[0130] First mineral deposit, A1 (sampling ID), 0.6 (Au content, ppb), 4.2 (As content, ppb), 200 (distance from fault zone, m), 12 (slope, °), 1.2 (Hg content in soil, ng / m³) 3 First mining site, A2 (sampling ID), 85.0 (Au content, ppb), 52.0 (As content, ppb), 150 (distance from fault zone, m), 8 (slope, °), 5.8 (Hg content in soil, ng / m³) 3 First mining site, A3 (sampling ID), 0.2 (Au content, ppb), 2.1 (As content, ppb), 800 (distance from fault zone, m), 22 (slope, °), 0.9 (Hg content in soil, ng / m³) 3 );

[0131] Second mining site, D1 (sampling ID), 0.7 (Au content, ppb), 3.8 (As content, ppb), 250 (distance from fault zone, m), 15 (slope, °), 1.1 (soil Hg content, ng / m³) 3 Second mining site, D2 (sampling ID), 35.0 (Au content, ppb), 22.0 (As content, ppb), 180 (distance from fault zone, m), 10 (slope, °), 3.2 (soil Hg content, ng / m³) 3 Second mining site, D3 (sampling ID), 0.3 (Au content, ppb), 2.5 (As content, ppb), 900 (distance from fault zone, m), 20 (slope, °), 0.8 (soil Hg content, ng / m³). 3 ).

[0132] Logarithmic transformation, applicable to Au content, As content, and Hg content in soil, eliminates the interference of extreme values ​​in the log-normal distribution. The results are as follows:

[0133] For example, for the first mining site, A1, log10(0.6+1)=0.2 (Au content), log10(4.2+1)=0.72 (As content), log10(1.2+1)=0.34 (Hg content in the soil). Other examples are calculated in the same way as above.

[0134] Z-score standardization is applicable to distance from the fault zone and slope. Taking the first mining point A1 as an example, the distance from the fault zone is: sample mean = 450m, standard deviation 280m; A1 is (200-450) / 280≈-0.89; other data are calculated in the same way.

[0135] First Au mean calculation (arithmetic mean of Au content after feature transformation).

[0136] Correlation coefficient calculation (Pearson correlation coefficient, quantifying linear association):

[0137] This is achieved through the correlation coefficient calculation formula:

[0138] ;

[0139] Where r is the calculated correlation coefficient, and n is the number of sampling points. The sum of the products of the two correlation parameters that need to be calculated for all sampling points. The sum of the first parameter of the two correlation parameters required to be calculated for all sampling points. The sum of the second parameter among the two correlation parameters required to be calculated for all sampling points. The sum of the squares of the first parameter among the two correlation parameters required to be calculated for all sampling points. The sum of the squares of the second parameter among the two correlation parameters required to be calculated for all sampling points.

[0140] Taking the calculation of primary correlation as an example, let's illustrate:

[0141] ;

[0142] Similarly, calculate the second correlation.

[0143] The third correlation coefficient, calculated using Formula 2, yields Hg correction data with k=0.008, which has been pre-verified as correct using known mining sites. For example:

[0144] Hg 检测 =0.9ng / m 3 Slope = 22°, Hg 校正 =0.9×(1+0.008×(22-5))≈1.02ng / m 3Then, perform the transformation according to Formula 1 on the Hg-corrected data and substitute it into the correlation coefficient calculation formula to calculate the third correlation coefficient.

[0145] This embodiment obtains cross-domain mineral deposit data consistent with mineralization mechanisms and data specifications through "database filtering or standardized real-time acquisition," categorized into "0.5km of densely populated fault zones." 2 / point, flat area 1km 2 The scientifically densityed sampling points, using a " / point" approach, combined with high-precision detection methods such as ICP-MS and GCMS, and GIS spatial analysis technology, ensure the reliability and consistency of multi-source data, including Au content, As content, and distance from fault zones. Subsequently, logarithmic transformation is used to eliminate extreme value interference, Z-score standardization unifies the dimensions, and the mean Au value reflects the mineralization statistical level. Pearson correlation coefficients are used to quantify the mineralization correlation between Au and As, distance from fault zones, and slope-corrected Hg content. This effectively integrates cross-domain data to supplement the total sample size, avoiding model overfitting due to insufficient samples from a single mining area. Furthermore, standardization and mechanism-oriented correlation quantification eliminate cross-domain data distribution offsets and terrain interference, ensuring data quality and consistency with mineralization patterns. Ultimately, models trained on this data can more accurately capture weak mineralization signals in shallow-covered areas, significantly reducing the risk of missed or misjudged targets, improving prediction accuracy and exploration efficiency, and reducing ineffective investment.

[0146] Optionally, in a specific embodiment, the data for the first and second mining sites are adjusted based on the mean value of the first Au and the correlation coefficient corresponding to the first and second mining sites, such as... Figure 2 As shown, it includes:

[0147] S31. Average the first Au values ​​corresponding to the first and second mining sites to obtain the second Au values ​​corresponding to the first and second mining sites.

[0148] S32. Based on the first and second mean Au values ​​corresponding to the first and second mining sites, obtain the Au adjustment parameters corresponding to the first and second mining sites respectively; wherein, the Au adjustment parameter is the second mean Au divided by the first mean Au.

[0149] S33. Based on the Au adjustment parameter corresponding to the first mining point, adjust the Au content of each sampling point in the first mining point; based on the Au adjustment parameter corresponding to the second mining point, adjust the Au content of each sampling point in the second mining point; the adjusted Au content is the product of the Au adjustment parameter and the Au content.

[0150] S34. Based on the correlation coefficient corresponding to the first mining point and the adjusted Au content of each sampling point in the first mining point, adjust the correlation coefficient corresponding to each sampling point in the first mining point. Based on the correlation coefficient corresponding to the second mining point and the adjusted Au content of each sampling point in the second mining point, adjust the correlation coefficient corresponding to each sampling point in the second mining point.

[0151] Furthermore, based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first Au mean value corresponding to the first and second mining sites is obtained, including:

[0152] Based on the Au content corresponding to multiple sampling points in the first mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the first mining site is obtained.

[0153] Based on the Au content corresponding to multiple sampling points in the second mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the second mining site is obtained.

[0154] The first Au mean is the value corresponding to the 50th percentile.

[0155] Specifically, the Au contents of the two mining sites after transformation are arranged in ascending order (Mining Site 1: [0.08, 0.09, ..., 0.18, ..., 2.08]; Mining Site 2: [0.09, 0.10, ..., 0.23, ..., 1.56]). For Mining Site 1, the mean Au content of the 40th and 41st samples after transformation is 0.18, i.e., the mean Au content of the first mining site is 0.18. For Mining Site 2, the mean Au content of the 30th and 31st samples after transformation is 0.23, i.e., the mean Au content of the first mining site is 0.23.

[0156] The quantile mapping algorithm avoids the influence of extreme values ​​by sorting and extracting median values, ensuring that the first mean Au value can reflect the core distribution level of Au content in the mining site (which is superior to the arithmetic mean, which is easily affected by outliers).

[0157] Calculate the second mean Au value: Second mean Au value = (First mean Au value of the first ore deposit + First mean Au value of the second ore deposit) / 2 = (0.18 + 0.23) / 2 = 0.205 ≈ 0.21;

[0158] The Au adjustment parameters are calculated as follows: Au adjustment parameter for the first mining site = 0.21 / 0.18 ≈ 1.1167, and Au adjustment parameter for the second mining site is approximately 0.913.

[0159] Adjust the Au content at each sampling point (adjusted Au content = adjustment parameter × original transformed Au content);

[0160] After verification of the adjustment results, the 50th percentile of Au content at the first mining site was approximately 0.21 (consistent with the mean Au value at the second mining site); the 50th percentile of Au content at the second mining site was approximately 0.21 (consistent with the mean Au value at the second mining site).

[0161] The correlation coefficient was adjusted based on the constraint between the adjusted Au content and the original correlation coefficient.

[0162] Maintain the positive correlation between Au and As content (first correlation coefficient, 0.72 for the first mining site, 0.68 for the second mining site), recalculate the covariance based on the adjusted Au content, and ensure that the correlation coefficient deviation is ≤5%;

[0163] Adjustment formula (derived based on the Pearson correlation coefficient definition): Adjusted correlation coefficient r' = r × (σ_Au' / σ_Au) × (σ_As / σ_As') × (Cov(Au',As) / Cov(Au,As)), where σ is the standard deviation, Cov is the covariance, r is the original first correlation coefficient, σ_As is the standard deviation of As content before adjustment, σ_As' is the standard deviation of As content after adjustment, σ_Au is the standard deviation of Au content before adjustment, σ_Au' is the standard deviation of Au content after adjustment, Cov(Au', As) is the covariance of Au content after adjustment and As content before adjustment, and Cov(Au, As) is the covariance of Au content before adjustment and As content before adjustment. Example:

[0164] σ_Au=0.12, σ_Au'=0.12×1.167≈0.14, σ_As=σ_As'=0.15, Cov(Au, As)=0.0126, Cov(Au', As)=0.0126×1.167≈0.0147, then after adjustment r'=0.72×(0.14 / 0.12)×(0.15 / 0.15)×(0.0147 / 0.0126)≈0.98 (the deviation exceeds 5%, and correction is required);

[0165] The logic was corrected because the As content was not adjusted. It is necessary to ensure that r'=r by "constraining the covariance ratio". After the adjustment, the first correlation coefficient of the first mining point is 0.71 and that of the second mining point is 0.67.

[0166] The adjustment logic for the second and third correlation coefficients is the same as that for the first correlation coefficient.

[0167] This application avoids interference from extreme values ​​through quantile mapping algorithms, ensuring the accuracy of the core distribution level of Au content. Then, it calculates a second Au mean value for a unified benchmark by averaging the first Au mean values ​​from two mining sites. Combined with Au adjustment parameters, it scales and adjusts the Au content at each sampling point of the two mining sites, achieving precise alignment of the Au content distribution center across mining areas. This fundamentally eliminates the distribution offset problem caused by differences in mineralization intensity in different mining areas. Simultaneously, in the correlation coefficient adjustment, the original correlation coefficient is used as a constraint, and the changes in covariance and standard deviation of the adjusted Au content are adapted to ensure... The mineralization correlation trends (positive / negative correlation) and intensity deviations of Au and As, distance from fault zones, and Hg correction data are ≤5%, fully preserving the true mineralization regularity. This adjustment method effectively integrates cross-domain data to supplement the total sample size, solving the model overfitting problem caused by insufficient samples from a single mining area. It also ensures the consistency and geological mechanism adaptability of the adjusted data, providing high-quality and highly collaborative fusion data for model training. Ultimately, it significantly improves the generalization ability and prediction accuracy of the gold target area prediction model in shallow overburden areas, reduces the risk of missed and misjudgments, and reduces ineffective exploration investment.

[0168] Optionally, in a specific embodiment, the gold ore target area prediction model is trained based on the adjusted first and second ore point data to obtain a gold ore target area prediction model for predicting gold ore target areas in shallow-covered areas, including:

[0169] Mineral point labels are added to the adjusted first and second mineral point data to mark the corresponding mineral point labels in the first and second mineral point data.

[0170] Input the first and second mineral point data after marking the mineral point into the gold mine target area prediction model to obtain the prediction data corresponding to the first and second mineral points.

[0171] Based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the model parameters of the gold ore target area prediction model are adjusted to obtain a gold ore target area prediction model for shallow cover areas.

[0172] Furthermore, based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the model parameters of the gold ore target area prediction model are adjusted, including:

[0173] Based on the predicted data, mining point labels, and second Au average value corresponding to the first and second mining points, the corresponding total loss value is obtained.

[0174] The model parameters of the gold mine target area prediction model are adjusted based on the total loss value and the pre-set BP algorithm.

[0175] Furthermore, the forecast data includes predicted ore-bearing probability and predicted grade;

[0176] Then, based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the corresponding total loss value is obtained, including:

[0177] Based on the predicted mineral-bearing probabilities and mineral-bearing labels corresponding to the first and second mineral deposits, and the pre-set Formula 3, the corresponding classification loss is obtained; Formula 3 is:

[0178] ;

[0179] Where L1 is the classification loss, y1 is the mineral point label corresponding to the first mineral point, and y2 is the mineral point label corresponding to the second mineral point. This represents the predicted ore-bearing probability corresponding to the first ore deposit. This represents the predicted ore-bearing probability corresponding to the second mineral deposit.

[0180] Based on the predicted grades of the first and second mining sites and the average of the second Au, and using the pre-set Formula 4, the corresponding grade regression loss is obtained; Formula 4 is:

[0181] ;

[0182] Where L2 represents the grade regression loss, G1 represents the second mean Au value corresponding to the first ore deposit, and G2 represents the second mean Au value corresponding to the second ore deposit. The predicted grade corresponding to the first mining site. The predicted grade corresponding to the second mining site;

[0183] The classification loss and grade regression loss are weighted and summed to obtain the corresponding total loss value.

[0184] Specifically, the adjusted data for the first and second mining sites are labeled to clarify the "true attributes" of each sampling point, serving as a supervisory signal for model training. The labeling rule is as follows: if a sampling point is located within a known mineralized area (such as an area verified by boreholes as a mineralized body), it is labeled as a positive sample (mineral point label y=1); if it is located in a background area without mineralization, it is labeled as a negative sample (mineral point label y=0).

[0185] The labeled data from the two mining sites (a total of 140 records) were input into the gold mine target area prediction model (the model adopts a CNN+MLP hybrid architecture: CNN extracts spatial features, and MLP learns element-related features). The model outputs two types of prediction results:

[0186] Predicted mineralization probability (y'): The probability that the sampling point is a mineralization point as determined by the model, with a value range of [0,1] (the closer to 1, the higher the probability of mineralization).

[0187] Predicted grade (G'): The model predicts the Au grade (unit: g / t) corresponding to this sampling point, which is obtained by inverse calculation based on the adjusted Au content (it needs to be combined with the previous logarithmic transformation for inverse operation).

[0188] We use a weighted sphere method combining classification loss and grade loss to comprehensively evaluate the model's predictive performance. The smaller the total loss value, the better the model's performance. For example:

[0189] y1=1, Substituting 0.85 into the equation, we get [1×log(0.85)+(1-1)×log(1-0.85)]≈0.1625. Therefore, for the second mineral deposit, y2=1. =0.78, and similarly we can get -0.2485, so the classification loss is 0.2055;

[0190] G1 = 2.1 g / t =2.3g / t, substituting this into the equation yields (2.1-2.3). 2 =0.04, G2=2.1g / t, =1.9g / t, and similarly we can get 0.04, so the taste loss will be L2=0.04.

[0191] L=0.7×0.2055+0.3×0.04≈0.1559.

[0192] Backpropagation calculates the gradient of the total loss value with respect to each parameter of the model (such as the convolutional kernel weights of CNN and the weight matrix of MLP); parameter update uses gradient descent (such as the Adam optimizer) to adjust the parameters in the direction of the gradient and reduce the total loss value (for example, if the learning rate is set to 0.001, after 100 iterations, the total loss value converges from the initial 0.5 to 0.08).

[0193] Iterative training: Repeat the process of "input data → prediction → loss calculation → parameter adjustment" until the model's performance on the validation set reaches its optimal level (e.g., ore-bearing prediction accuracy ≥ 85%, grade prediction error ≤ 0.2 g / t); Finally, after training, save the optimal parameters to obtain a model that can be directly used for gold ore target area prediction in shallow overburden areas.

[0194] This embodiment simultaneously optimizes "ore-bearing judgment" and "grade prediction". The model can not only locate the target area, but also assess the potential of the target area (high-grade target areas are given priority for exploration). It also calculates the regression loss based on a unified second Au mean, ensuring that the grade prediction benchmarks of the two ore points are consistent and improving the model's generalization ability.

[0195] Furthermore, the optimizer uses the AdamW optimizer (Adam + weight decay), with an initial learning rate of 1e4 and a weight decay coefficient of 1e5 (to prevent overfitting). For learning rate scheduling, cosine annealing LR is used, where the learning rate is decayed to 1e6 along a cosine curve every 50 iterations, and then increased back to 1e4 to avoid the model getting trapped in local optima.

[0196] Optionally, in a specific embodiment, obtaining the first mining point data corresponding to the pre-collected first mining point and the second mining point data corresponding to the second mining point includes:

[0197] Acquire the first mineral point data corresponding to the first mineral point and the second mineral point data corresponding to the second mineral point, and the mineralization mechanism and data acquisition specifications of the two different mineral points are consistent;

[0198] Based on the pre-set WGS84 coordinate system, the sampling coordinates of all sampling points are unified and resampled into a 10m×10m grid to match the sampling point space.

[0199] This embodiment selects cross-domain mineral point data with consistent mineralization mechanisms and data collection standards. Data quality is optimized through logarithmic transformation, Z-score standardization, and Hg slope correction. A quantile mapping algorithm is used to accurately extract the first Au mean, which is then combined with the second Au mean and adjusted parameters to align the cross-domain Au content distribution. Simultaneously, the correlation coefficient deviation is constrained to ≤5% to preserve mineralization patterns. Furthermore, through mineral point labeling, dual-task prediction of ore-bearing probability and grade, and total loss calculation using a weighted sum of "classification loss + regression loss," model parameters are optimized based on the BP algorithm. This effectively integrates cross-domain data to supplement the total sample size and solves the overfitting problem caused by insufficient samples from a single mineral area. It also ensures data consistency and geological mechanism adaptability, ultimately significantly improving the generalization ability and prediction accuracy of the shallow-cover gold ore target area prediction model, reducing the risk of missed or misjudged detections, providing precise guidance for exploration work, and reducing ineffective investment.

[0200] Furthermore, this application provides a shallow-covered gold ore target area prediction system, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the above-described shallow-covered gold ore target area prediction method.

[0201] In the description of this application, it should be understood that the terms "first" and "second" are used 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" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0202] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0203] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0204] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0205] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for predicting gold target areas in shallow-covered areas, characterized in that, include: Input the multi-source mineral point data of the area to be predicted into the pre-trained gold mine target area prediction model to obtain the corresponding gold mine target area prediction results. The gold mine target area prediction model is trained through the following steps: Acquire the first mineral point data corresponding to the first mineral point and the second mineral point data corresponding to the second mineral point, and the mineralization mechanism and data acquisition specifications of the two different mineral points are consistent; the first mineral point data and the second mineral point data both include Au content and mineral point related data corresponding to multiple different sampling points; Based on the data from the first mining site, the mean value of Au and the correlation coefficient corresponding to the first mining site are obtained. Based on the data from the second mining site, the mean value of Au and the correlation coefficient corresponding to the second mining site are obtained. The correlation coefficient represents the degree of correlation between Au content and relevant data of the mining site. Based on the first Au mean and correlation coefficient corresponding to the first and second mining sites, the data of the first and second mining sites are adjusted to obtain the adjusted data of the first and second mining sites. The gold target area prediction model was trained based on the adjusted data from the first and second mining sites to obtain a gold target area prediction model for shallow-covered areas.

2. The method for predicting gold target areas in shallow-covered areas according to claim 1, characterized in that, The relevant data for the mineral deposits include As content and distance from the fault zone; The correlation coefficient includes: the first correlation coefficient and the second correlation coefficient; Based on the data from the first mining site, the mean Au value and correlation coefficient for the first mining site are obtained. Based on the data from the second mining site, the mean Au value and correlation coefficient for the second mining site are obtained, including: Feature transformations are performed on the data from the first and second mineral deposits. The Au and As contents are transformed using a pre-defined formula (Formula 1), and the distance to the fault zone is transformed using a Z-score. Formula 1 is as follows: x1=log 10 (x0+1); Where x0 is the parameter value before feature transformation, and x1 is the parameter value after feature transformation; Based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first Au mean value corresponding to the first and second mining sites is obtained. Based on the first mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the first mining site are constructed. Based on the second mining site data after feature transformation, a first correlation coefficient and a second correlation coefficient for the second mining site are constructed. The first correlation coefficient represents the degree of correlation between Au content and As content, while the second correlation coefficient represents the degree of correlation between Au content and distance from the fault zone.

3. The method for predicting gold target areas in shallow-covered areas according to claim 2, characterized in that, The relevant data for the mining site also includes slope and Hg content in the soil; the correlation coefficient also includes a third correlation coefficient. Then, based on the data from the first mining site, the mean value of the first Au and the correlation coefficient corresponding to the first mining site are obtained; based on the data from the second mining site, the mean value of the first Au and the correlation coefficient corresponding to the second mining site are obtained; and the process also includes: Using Formula 1, the soil Hg content at each sampling point in the first and second mining site data is transformed using a feature transformation, and the slope at each sampling point in the first and second mining site data is transformed using the Z-score. Based on the soil Hg content and slope corresponding to each sampling point in the first and second mining site data after feature transformation, and using a pre-set formula two, the Hg correction data corresponding to each sampling point is obtained; the formula two is: Hg 校正 =Hg 检测 ×(1+k×(Slope-5°)); Among them, Hg 检测 The Hg content in the soil corresponding to the sampling point. 校正 Here, Hg is the Hg correction data corresponding to the sampling point, Slope is the slope corresponding to the sampling point, and k is the preset slope correction parameter; Based on the Hg-corrected data and Au content corresponding to each sampling point in the first and second mining site data, the third correlation coefficients corresponding to the first and second mining sites are obtained respectively; the third correlation coefficient is the degree of correlation between Au content and Hg-corrected data.

4. The method for predicting gold target areas in shallow-covered areas according to claim 1, characterized in that, Based on the mean and correlation coefficient of the first Au value corresponding to the first and second mining sites, adjustments were made to the data for the first and second mining sites, including: The average value of the first Au corresponding to the first mining site and the second mining site is averaged to obtain the average value of the second Au corresponding to the first mining site and the second mining site. Based on the first and second mean Au values ​​corresponding to the first and second mining sites, Au adjustment parameters are obtained for the first and second mining sites, respectively; wherein, the Au adjustment parameter is the second mean Au divided by the first mean Au. Based on the Au adjustment parameter corresponding to the first mining site, the Au content of each sampling point in the first mining site is adjusted. Based on the Au adjustment parameter corresponding to the second mining site, the Au content of each sampling point in the second mining site is adjusted. The adjusted Au content is the product of the Au adjustment parameter and the Au content. Based on the correlation coefficient corresponding to the first mining site and the adjusted Au content of each sampling point in the first mining site, the correlation coefficient corresponding to each sampling point in the first mining site is adjusted. Based on the correlation coefficient corresponding to the second mining site and the adjusted Au content of each sampling point in the second mining site, the correlation coefficient corresponding to each sampling point in the second mining site is adjusted.

5. The method for predicting gold target areas in shallow-covered areas according to claim 2, characterized in that, Based on the Au content corresponding to multiple sampling points in the first and second mining site data after feature transformation, the first mean Au value corresponding to the first and second mining sites is obtained, including: Based on the Au content corresponding to multiple sampling points in the first mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the first mining site is obtained. Based on the Au content corresponding to multiple sampling points in the second mining site data after feature transformation and the pre-set quantile mapping algorithm, the first mean Au value corresponding to the second mining site is obtained. The first Au mean is the value corresponding to the 50th percentile.

6. The method for predicting gold target areas in shallow-covered areas according to claim 4, characterized in that, The gold target area prediction model was trained based on the adjusted data from the first and second mining sites to obtain a gold target area prediction model for shallow-covered areas, including: Mineral point labels are added to the adjusted first and second mineral point data to mark the corresponding mineral point labels in the first and second mineral point data. Input the first and second mineral point data after marking the mineral point into the gold mine target area prediction model to obtain the prediction data corresponding to the first and second mineral points. Based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the model parameters of the gold ore target area prediction model are adjusted to obtain a gold ore target area prediction model for shallow cover areas.

7. The method for predicting gold target areas in shallow-covered areas according to claim 6, characterized in that, Based on the predicted data corresponding to the first and second mining sites, the mining site labels, and the second Au mean, the model parameters of the gold ore target area prediction model were adjusted, including: Based on the predicted data, mining point labels, and second Au average value corresponding to the first and second mining points, the corresponding total loss value is obtained. Based on the total loss value and the pre-set BP algorithm, the model parameters of the gold mine target area prediction model are adjusted.

8. The method for predicting gold target areas in shallow-covered areas according to claim 7, characterized in that, The prediction data includes predicted ore-bearing probability and predicted grade; Based on the predicted data, mining point labels, and the second Au mean value corresponding to the first and second mining points, the corresponding total loss value is obtained, including: Based on the predicted mineral-bearing probabilities and mineral-bearing labels corresponding to the first and second mineral deposits, and using a pre-set formula (Formula 3), the corresponding classification loss is obtained; formula 3 is: ； Where L1 is the classification loss, y1 is the mineral point label corresponding to the first mineral point, and y2 is the mineral point label corresponding to the second mineral point. This represents the predicted ore-bearing probability corresponding to the first ore deposit. This represents the predicted ore-bearing probability corresponding to the second mineral deposit. Based on the predicted grades and the average Au values ​​of the first and second mining sites, and using a pre-set formula (Formula 4), the corresponding grade regression loss is obtained; formula 4 is: ； Where L2 represents the grade regression loss, G1 represents the second mean Au value corresponding to the first ore deposit, and G2 represents the second mean Au value corresponding to the second ore deposit. The predicted grade corresponding to the first mining site. The predicted grade corresponding to the second mining site; The classification loss and grade regression loss are weighted and summed to obtain the corresponding total loss value.

9. The method for predicting gold target areas in shallow overburdened areas according to claim 1, characterized in that, The method further includes: After obtaining the first mining point data corresponding to the first mining point and the second mining point data corresponding to the second mining point, the sampling coordinates of all sampling points are unified based on the pre-set WGS84 coordinate system and resampled into a 10m×10m grid to match the sampling point space.

10. A gold ore target area prediction system in shallow overburden areas, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the shallow-cover gold ore target area prediction method according to any one of claims 1 to 9.

Images