Copper ore prediction method based on multi-source data

Abstract

The invention provides a copper mine prediction method based on multi-source data, and the method comprises the steps: S11, taking a copper mine geological report as a data source, and enabling the copper mine geological report to comprise a geological map, geophysical data, a geochemical sample analysis result and a remote sensing image; s12, respectively carrying out vectorization processing on the data obtained in the step S11; s13, inputting the information processed in the step S12 into a machine learning model for feature extraction, and filling the extracted features into a copper ore distribution diagram; s14, constructing a multi-modal copper mine prediction model, inputting the copper mine distribution diagram into the multi-modal copper mine prediction model for training, and obtaining a trained multi-modal copper mine prediction model; and S15, inputting the newly obtained copper ore geological report into the multi-mode copper ore prediction model to obtain a copper ore distribution prediction map of the target area. According to the method, the relevance and similarity of the multi-source data in the copper mine geological report are extracted, the multi-mode copper mine prediction model is established to predict the potential position of the copper mine, and the success rate and efficiency of prospecting are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of copper mine prediction, and in particular to a copper mine prediction method based on multi-source data. Background Art

[0002] Copper ore exploration is a complex and technology-intensive project, the purpose of which is to find copper-rich deposits in the earth's crust. With the growing demand for copper in new energy and high-tech industries, copper ore exploration and mining technologies are also constantly developing. Modern copper ore exploration technologies mainly include remote sensing technology, geophysical exploration, geochemical exploration, drilling technology, artificial intelligence and big data analysis. Remote sensing technology obtains surface information through satellite and aerial photography, and combines image processing and analysis to identify potential mineralized areas. Geophysical exploration includes methods such as seismic, electromagnetic, gravity and magnetic exploration to detect underground geological structures and ore body locations. Geochemical exploration searches for abnormal areas related to mineralization by analyzing the chemical composition of soil, water or rocks. Drilling technology is used to obtain underground core samples for direct observation and analysis of ore bodies.

[0003] The above-mentioned technologies for exploring the distribution of copper deposits have been proven to be effective in prospecting predictions, but there are certain limitations when predicting the distribution of copper deposits only through data sources in a single technical field. For example: although remote sensing technology can quickly obtain large-area surface information, identify surface rock types and vegetation coverage, and provide clues for copper mine exploration, it has obvious limitations in detecting deep minerals; although geophysical exploration methods are widely used, they are easily affected by factors such as particle size and porosity, and in the exploration of deep mineral resources, the general laws of its mineralization information are difficult to reflect, which increases the difficulty of exploration; although geochemical exploration can effectively identify and locate metal minerals, and predict mineral resources by analyzing the geochemical characteristics of soil, rocks and mineralized zones, it requires a large amount of samples and data processing, which is costly. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a copper ore prediction method based on multi-source data, by extracting the correlation and similarity of multi-source data in the copper mine geological report, establishing a multimodal copper ore prediction model to predict the potential location of the copper mine, and improving the success rate and efficiency of prospecting.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a copper mine prediction method based on multi-source data, the method comprising:

[0006] S11, using a copper mine geological report as a data source and preprocessing the data source, wherein the copper mine geological report includes a geological map, geophysical data, geochemical sample analysis results, and remote sensing images;

[0007] S12, respectively performing vector processing on the geological map, geophysical data, geochemical sample analysis results and remote sensing images to obtain corresponding geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information;

[0008] S13, respectively inputting geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information into a machine learning model for feature extraction, and filling the extracted features into a copper mine distribution map;

[0009] S14, constructing a multimodal copper ore prediction model, inputting the copper ore distribution map into the multimodal copper ore prediction model for training, and obtaining a trained multimodal copper ore prediction model;

[0010] S15. Input the newly acquired copper ore geological report into the multimodal copper ore prediction model to obtain a copper ore distribution prediction map for the target area.

[0011] Furthermore, in step S11, the data source is preprocessed, specifically including:

[0012] The geological maps, geophysical data, geochemical sample analysis results and remote sensing images were cleaned, standardized and normalized respectively.

[0013] Furthermore, in step S12, the geological map is vectorized, which specifically includes:

[0014] S21, dividing the geological map into m rows×n columns of grids according to a preset grid unit size, each grid being a geological unit, and each geological unit including a geological element and a geological attribute corresponding to the geological element;

[0015] S22, constructing a corpus based on regional geological data, and inputting the corpus constructed by the regional geological data into deep learning for training to obtain a geological language model;

[0016] S23, inputting the geological attributes of the geological elements in each geological unit into the geological language model to obtain a vectorized representation of the geological attributes of the geological elements in each geological unit;

[0017] S24. Superimpose the vectors of different geological elements, convert the geological attribute vector information of the geological elements into geological vector information through the geological language model, and assign the geological vector information to each corresponding geological unit.

[0018] Furthermore, in step S12, the geophysical data is vectorized, which specifically includes:

[0019] S31, digitizing the collected geophysical data, and converting the coordinates of the digitized geophysical data into geographic absolute coordinates;

[0020] S32, interpolating and resampling the data under the geographic absolute coordinates, completing the matching of the time coordinates and the spatial coordinates of the geophysical data, and establishing a unified coordinate database according to the matching results;

[0021] S33. The data in the coordinate database and the geological vector information in the geological unit are feature extracted and fused by using multi-source data information fusion technology to obtain corresponding geophysical vector information.

[0022] Furthermore, in step S12, the geochemical sample analysis results are vectorized, specifically including:

[0023] S41, performing principal component analysis on the geochemical sample analysis results to determine the correlation between each principal component and the target geochemical element variable;

[0024] S42, taking the score of each sample on the principal component as a new eigenvector to form a principal component feature matrix, wherein each row of the feature matrix represents a representation of a sample in the principal component space, i.e., a principal component eigenvector;

[0025] S43. The principal component characteristic vector is extracted and fused with the geological vector information in the geological unit through multi-source data information fusion technology to obtain the corresponding geochemical vector information.

[0026] Furthermore, in step S12, the remote sensing image is vectorized, which specifically includes:

[0027] S51, extracting features of the remote sensing image through a deep convolutional neural network to obtain a global image feature vector and a local image feature vector;

[0028] S52, performing cross attention calculation on the global image feature vector and the local image feature vector to determine the target feature vector;

[0029] S53. Extract and fuse the target feature vector with the geological vector information in the geological unit through multi-source data information fusion technology to obtain corresponding remote sensing vector information.

[0030] Further, in step S13, the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information are respectively input into the machine learning model for feature extraction, specifically including:

[0031] S61, dynamically learning the optimal weights of geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information through an automatic weighting framework of reinforcement learning;

[0032] S62. Based on the optimal weights, feature extraction is performed on the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information through a neural network to obtain geological features, geophysical features, geochemical features and remote sensing features respectively.

[0033] Furthermore, in step S13, the extracted features are respectively filled into the copper ore distribution map, specifically including:

[0034] S71, respectively performing correlation processing on the geological features, geophysical features, geochemical features and remote sensing features in pairs to obtain correlation information;

[0035] S72, taking the geological feature as the target feature, calculating the similarity between the geophysical feature, the geochemical feature, and the remote sensing feature and the target feature based on the correlation information, and obtaining similarity information;

[0036] S73, comparing the similarity information with a preset threshold, and filling the similarity information into the copper mine distribution map according to the comparison result based on GIS technology.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The invention provides a copper mine prediction method based on multi-source data, which vectorizes the multi-source data with the copper mine geological report as the data source, thereby improving the operability and analysis efficiency of the multi-source data. The feature extraction capability of the machine learning model can automatically identify the key features in the multi-source data after vectorization, and fill the extracted features into the copper mine distribution map in combination with GIS technology, thereby realizing data visualization, enabling geologists to intuitively understand the hidden meaning of the multi-source data, and constructing a multi-modal copper mine prediction model trained with the copper mine distribution map, integrating the advantage of predicting the distribution of potential copper mines through multi-source data. Finally, the multi-source data in the newly acquired copper mine geological report is input into the trained multi-modal copper mine prediction model to generate a copper mine distribution prediction map, thereby realizing the rapid evaluation and prediction of the distribution of copper mines in new areas, improving the exploration efficiency, reducing the exploration cost, and providing strong technical support for the rational development and utilization of copper mine resources. The invention improves the automation of data processing and analysis through the fusion of multi-source data and the deep mining of machine learning, realizes the prediction of the potential location of copper mines, and finally improves the efficiency and accuracy of copper mine mineral resource exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0040] Figure 1 A schematic flow chart of a copper ore prediction method based on multi-source data provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0042] Reference Figure 1 This embodiment provides a copper mine prediction method based on multi-source data, the method comprising:

[0043] S11. Using a copper mine geological report as a data source and preprocessing the data source, the copper mine geological report includes a geological map, geophysical data, geochemical sample analysis results, and remote sensing images.

[0044] In this embodiment, the geological map provides the spatial distribution and geological structure information of the deposit, the geophysical data reflects the physical properties of the underground rock, the geochemical sample analysis results reveal the distribution characteristics of the elements, and the remote sensing image shows the landform and geological characteristics of the surface. The above data reflects the metallogenic environment and metallogenic conditions of the copper mine from different angles. Through comprehensive analysis, the formation process and distribution law of the copper deposit can be more comprehensively understood, thereby improving the accuracy of the prediction. For example, the abnormal element distribution in the geochemical sample analysis results may indicate the mineralization potential area, and the abnormal signal in the geophysical data may reveal the existence of the underground ore body. By fusing and analyzing these data, the potential mineralization area of ​​the copper mine can be more accurately delineated, providing guidance for subsequent exploration work. The processed data can be used to establish a multimodal copper mine prediction model in the future to predict the spatial distribution characteristics and mineralization law of the copper mine deposit, reasonably plan the location and depth of the exploration borehole, and accurately predict the distribution of the mineralization area and the ore body, which can reduce unnecessary exploration work, reduce exploration costs, and improve exploration efficiency.

[0045] In step S11, the data source is preprocessed, specifically including:

[0046] The geological maps, geophysical data, geochemical sample analysis results and remote sensing images were cleaned, standardized and normalized respectively.

[0047] In this embodiment, for geological maps, repeated, erroneous or inconsistent information in the geological maps can be removed by cleaning to ensure the accuracy and consistency of the data in the maps, for example, eliminating the erroneous place names or deviations of the location of features in the maps due to human error. Standardization and normalization processing can convert the data in the geological maps into a unified format and scale, which is convenient for comparison and comprehensive analysis between different geological maps, for example, unifying geological maps of different scales into a standard scale to facilitate comprehensive research on regional geological characteristics. The processed geological map data can be effectively integrated with geophysical, geochemical and remote sensing data to provide more comprehensive geological information for copper mine prediction.

[0048] For geophysical data, cleaning can effectively remove noise and interference signals in geophysical data, improve the signal-to-noise ratio of data, and make the data more accurate and reliable. For example, it can remove terrain gravity anomalies that are not related to mineralization in gravity data. Standardization and normalization can eliminate the differences between data collected by different sensors or at different times, making the data comparable. For example, electromagnetic data obtained by different sensors can be converted into a unified standard unit to facilitate comparative analysis and interpretation of data. Cleaned and standardized geophysical data can more clearly reflect the underground geological structure and ore body characteristics, providing more comprehensive geophysical information for copper mine prediction.

[0049] For geochemical sample analysis results, cleaning can remove outliers and errors in the sample analysis process and improve the accuracy of the analysis results, for example, eliminating abnormally high or low values ​​caused by sample contamination or analytical instrument failure. Standardization and normalization can convert the analysis results of different elements into a unified standard form, which is convenient for comparison and correlation analysis between elements. For example, converting the content data of different elements into standardized values ​​is convenient for identifying the correlation between elements and the abnormal distribution of ore-forming elements. Cleaned and standardized geochemical data can be used to establish geochemical models and provide more comprehensive geochemical information for copper mine prediction.

[0050] For remote sensing images, cleaning can remove noise and artifacts in remote sensing images and improve image clarity and quality, for example, removing image noise caused by sensor failure or atmospheric interference. Standardization and normalization processing can improve the contrast and brightness of images and make the features of ground objects in mining areas more obvious, for example, enhancing the contrast between vegetation and bare soil through normalization processing. Cleaned and standardized remote sensing image data can be fused with different data to provide basic information for dynamic monitoring and change analysis of the mineralization laws of copper mines.

[0051] S12, respectively performing vector processing on the geological map, geophysical data, geochemical sample analysis results and remote sensing images to obtain corresponding geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information.

[0052] In this embodiment, for geological maps, vectorization processing can convert the information in the geological maps into precise digital coordinates and attribute data, improve the accuracy and consistency of the data, and conduct complex spatial analysis, such as buffer analysis, overlay analysis, etc., which is helpful for in-depth research on the distribution characteristics of copper mineral resources, and can be accurately matched and analyzed with other data.

[0053] For geophysical data, vectorization processing converts geophysical data into structured vector data, which is convenient for mathematical operations and statistical analysis, helping geologists better understand the underground geological structure and the spatial distribution of copper ore bodies. The vectorized geophysical data can be integrated and analyzed with other geological, geochemical and other data to further provide more comprehensive information for copper mineral prediction.

[0054] For the analysis results of geochemical samples, vectorization processing converts geochemical data into standardized vector form, which is convenient for comparing and analyzing the contents of different elements. Vectorized data can be used for multivariate statistical analysis, such as cluster analysis, principal component analysis, etc., which helps to reveal the correlation and mineralization laws between elements. The geochemical data after vectorization can more intuitively display the spatial distribution characteristics of elements, and further provide a clearer basis for copper ore prediction.

[0055] For remote sensing images, vectorization processing can extract the boundaries and feature information of objects in remote sensing images, improve the accuracy of identifying objects around copper mines, and the vectorized remote sensing data can be used to dynamically monitor surface changes, such as changes in vegetation coverage and land use, to provide important information for environmental monitoring and resource management. At the same time, vectorized remote sensing data can be integrated with geological, geophysical and other data to provide more comprehensive multi-source information support for geological mapping, copper mine mineral prediction, etc.

[0056] In step S12, the geological map is vectorized, which specifically includes:

[0057] S21. Divide the geological map into m-row×n-column grids according to a preset grid unit size, each grid being a geological unit, and each geological unit including geological elements and geological attributes corresponding to the geological elements.

[0058] S22. Construct a corpus based on regional geological data, and input the corpus constructed with regional geological data into deep learning for training to obtain a geological language model.

[0059] S23, inputting the geological attributes of the geological elements in each geological unit into the geological language model to obtain a vectorized representation of the geological attributes of the geological elements in each geological unit.

[0060] S24. Superimpose the vectors of different geological elements, convert the geological attribute vector information of the geological elements into geological vector information through the geological language model, and assign the geological vector information to each corresponding geological unit.

[0061] In this embodiment, according to the geological map Properties of geological features , such as stratigraphic age, rock combination, etc., according to the preset grid unit size, the geological map is divided into m rows × n columns of grids, each grid is a geological unit, each geological unit includes geological elements and geological attributes corresponding to the geological elements, and these geological attributes are closely related to the distribution of copper deposits. After the geological map is divided into grid units, the geological elements and their attributes in each geological unit are clearly identified and organized, so that the originally complex and unstructured geological information is transformed into a structured data form, which is convenient for computer processing and analysis. Using deep learning to build a geological language model can capture the complex relationships and laws between the attributes of geological elements through the information in the corpus constructed by regional geological data, and dig out geological information characteristics that are difficult to discover with traditional methods, thereby improving the understanding and interpretation of geological phenomena.

[0062] The vectorized representation method of geological maps can represent the spatial, temporal and material composition information of geological elements. It can vectorize geological attributes through geological language models, unify the attributes of different geological elements into vector form, and superimpose the vector representations of different geological elements to form a complete expression of geological map information. It can comprehensively consider the various attributes of geological elements, provide a more comprehensive reflection of the comprehensive characteristics of geological elements for the subsequent superposition and conversion of vectors, and provide richer information for the prediction and analysis of copper mine distribution. Assigning geological vector information to each geological unit can provide refined data support for geological decision-making, helping decision makers to make more scientific and reasonable decisions at the specific geological unit level, such as the exploration layout of mineral resources and the formulation of mining plans.

[0063] In step S12, the geophysical data is vectorized, which specifically includes:

[0064] S31. The collected geophysical data are digitized and the coordinates of the digitized geophysical data are converted into geographic absolute coordinates.

[0065] S32. Interpolate and resample the data under the geographic absolute coordinates to complete the matching of the time coordinates and space coordinates of the geophysical data, and establish a unified coordinate database based on the matching results.

[0066] S33. The data in the coordinate database and the geological vector information in the geological unit are feature extracted and fused by using multi-source data information fusion technology to obtain corresponding geophysical vector information.

[0067] In this embodiment, the geophysical data collected in the field are imported into a computer, and professional geophysical data processing software or general data processing tools such as Excel, MA TLAB, etc. are used to preliminarily organize the data, check the integrity and accuracy of the geophysical data and perform correction processing, and apply a filtering algorithm to filter the data to remove noise and interference signals in the data. Common filtering methods include: low-pass filtering, high-pass filtering, band-pass filtering, etc., and a suitable filtering method is selected according to the characteristics of the data and analysis requirements. The coordinates of the digitized geophysical data are converted into geographic absolute coordinates to ensure the accurate positioning of the data on the earth's surface, provide a reliable spatial reference for subsequent spatial analysis and data fusion, and avoid analysis deviations caused by inconsistent coordinates. Through coordinate conversion, the geophysical data and other spatial data (such as geological maps, remote sensing images, etc.) have a unified spatial coordinate system, which is convenient for spatial overlay analysis and fusion of multi-source data, and improves the compatibility of data.

[0068] The above data may be irregularly distributed in space. Through interpolation and resampling, regular grid data can be generated to fill the gaps in space, provide more complete and continuous spatial information, and improve the temporal and spatial integrity of the data. By matching the time coordinates and spatial coordinates of geophysical data, the temporal and spatial variation characteristics of geophysical phenomena can be accurately reflected. A unified coordinate database is established based on the matching results to centrally manage and store geophysical data and other related spatial data. In the unified coordinate database, the data is standardized and normalized, with a unified data format and structure, which is convenient for subsequent statistical analysis, model building, and decision support. Through multi-source data information fusion technology, the physical properties of the obtained geophysical vector information are combined with the geological properties of the geological vector information, and the characteristics and laws of copper ore geological information are comprehensively considered to provide information support for copper ore mineral prediction.

[0069] In step S12, the geochemical sample analysis results are vectorized, specifically including:

[0070] S41. Perform principal component analysis on the geochemical sample analysis results to determine the correlation between each principal component and the target geochemical element variable.

[0071] S42. The score of each sample on the principal component is used as a new eigenvector to form a principal component feature matrix, wherein each row of the feature matrix represents the representation of a sample in the principal component space, namely, the principal component eigenvector.

[0072] S43. The principal component characteristic vector is extracted and fused with the geological vector information in the geological unit through multi-source data information fusion technology to obtain the corresponding geochemical vector information.

[0073] In this embodiment, principal component analysis extracts the main features in the analysis results of geochemical samples. These features are linear combinations of the original variables and usually better reflect the essential structure of the data. For example, the first principal component may mainly reflect the distribution of copper elements, and the second principal component may reflect the joint distribution of lead and zinc. Determine the correlation between each principal component and the target geochemical element variable in the analysis results of geochemical samples, understand the interaction and influence between different elements, and provide a basis for the geological interpretation and mineralization mechanism research of copper mines. For example, it is found that the lead and zinc elements are strongly correlated on the second principal component, and it can be deduced that they have a common role in the copper mine mineralization process. The principal component score can explain the pattern in the data more intuitively. For example: a sample has a high score on the first principal component, which may mean that the sample is significantly higher than other samples in terms of copper content.

[0074] The projection value of each sample on each principal component is called the score of the sample on the principal component. These scores represent the position of the sample in the new coordinate system. The scores of each sample on all principal components are combined to form a new vector. This vector is the representation of the sample in the principal component space, that is, the new eigenvector. For example: suppose there are 100 samples, each of which measures 10 geochemical elements. After principal component analysis, we select the first 3 principal components because they explain most of the variance, that is, the score of each sample on the 3 principal components can be expressed as a 3-dimensional vector, which is the new eigenvector.

[0075] In the original data space, each sample is represented by its measured value on the original variable (such as geochemical elements such as Cu, Pb, Zn, Fe, etc.). For example, a sample may be measured as 10 on Cu, 5 on Pb, 8 on Zn, and 12 on Fe. These measured values ​​constitute the coordinates of the sample in the original data space. Through principal component analysis, we find a set of new coordinate axes (principal components), which are linear combinations of the original variables and are orthogonal (i.e., independent of each other). Each principal component explains a part of the variance in the original data. Usually, the first few principal components explain most of the variance. The projection value of each sample on these new coordinate axes (principal components) is called the score of the sample on the principal component. These scores constitute the coordinates of the sample in the principal component space. The scores of all samples on the principal components are combined to form a matrix, which is called the principal component feature matrix. Each row: Each row of the feature matrix corresponds to a sample, and each element in the row is the score of the sample on the corresponding principal component. Therefore, each row represents the representation of a sample in the principal component space.

[0076] The fusion of the principal component characteristic matrix and the geological vector information in the geological unit can comprehensively consider geochemical and geological information. The fused geochemical vector information can be used for more complex geological modeling and mineral prediction, and improve the comprehensive analysis ability of geological phenomena. For example: for a certain geological unit, its geological vector information shows that the geological structure of the area is conducive to mineralization, and the principal component characteristic vector shows that the samples of the unit have a higher score on the first principal component (copper element related). Combining this information, we can judge that the geological unit has a high copper mineralization potential.

[0077] In step S12, the remote sensing image is vectorized, which specifically includes:

[0078] S51, extracting features of the remote sensing image through a deep convolutional neural network to obtain a global image feature vector and a local image feature vector;

[0079] S52, performing cross attention calculation on the global image feature vector and the local image feature vector to determine the target feature vector;

[0080] S53. Extract and fuse the target feature vector with the geological vector information in the geological unit through multi-source data information fusion technology to obtain corresponding remote sensing vector information.

[0081] In this embodiment, the remote sensing image is feature extracted by a deep convolutional neural network (CNN) to obtain a global image feature vector and a local image feature vector, capturing the overall features of the image and the features of a specific area in the image, such as global features such as the overall texture and color distribution of the image, and local features such as buildings, rivers, and vegetation. The cross-attention mechanism can help the model focus on the most important part of the image, and by calculating the weights between the global features and the local features, determine which features are more important, and finally combine the global image feature vector and the local image feature vector to obtain a target feature vector, so that the global and local information can be more comprehensively represented. The target feature vector is feature extracted and fused with the geological vector information in the geological unit through multi-source data information fusion technology to form a more comprehensive feature representation. The geological vector information includes geological structure, lithology, mineralization information, etc. The fused remote sensing vector information is used for the subsequent construction of a multimodal copper mine prediction model and for mineral prediction of copper mines.

[0082] S13. The geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information are respectively input into the machine learning model for feature extraction, and the extracted features are filled into the copper mine distribution map.

[0083] In this embodiment, the extracted and fused feature vectors are filled into the copper ore distribution map. In this embodiment, the feature-extracted vector information is matched with the geographic spatial data through GIS software (such as ArcGIS) to generate a copper ore distribution map. For example, the feature vector is matched with the geological unit to generate a feature representation of each geological unit, and visualized on the copper ore distribution map.

[0084] In step S13, the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information are respectively input into the machine learning model for feature extraction, specifically including:

[0085] S61. Through the automatic weighting framework of reinforcement learning, the optimal weights of geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information are dynamically learned.

[0086] S62. Based on the optimal weights, feature extraction is performed on the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information through a neural network to obtain geological features, geophysical features, geochemical features and remote sensing features respectively.

[0087] In this embodiment, the automatic weighting framework of reinforcement learning can be used to dynamically learn the optimal weights of geological vector information, geophysical vector information, geochemical vector information, and remote sensing vector information. Reinforcement learning adjusts the weights according to environmental feedback through trial and error methods to maximize the cumulative reward. For example, using algorithms such as Q-learning or PolicyGradients, the weight adjustment process is regarded as a decision-making process, and the optimal weight configuration is found by constantly trying different weight combinations; the environment can be the spatial distribution of geological units, and the feature vector of each geological unit includes geological, geophysical, geochemical, and remote sensing features; the reward can be the accuracy of the prediction, for example, if the mineralized area predicted by the model is highly consistent with the actual mineralized area, a positive reward is given; otherwise, a negative reward is given. Each vector information is input into the neural network, and features are extracted through forward propagation. Each layer of the neural network extracts features from low-level edge and texture features to high-level semantic features. For example, for geochemical data, a fully connected neural network can be used to extract features and convert element content into feature vectors.

[0088] In step S13, the extracted features are respectively filled into the copper ore distribution map, specifically including:

[0089] S71. Correlation processing is performed on geological features, geophysical features, geochemical features and remote sensing features respectively to obtain correlation information.

[0090] S72. Taking geological features as target features, the similarities between geophysical features, geochemical features, and remote sensing features and the target features are calculated based on the correlation information to obtain similarity information.

[0091] S73, comparing the similarity information with a preset threshold, and filling the similarity information into the copper mine distribution map according to the comparison result based on GIS technology.

[0092] In this embodiment, the correlation between geological features and geophysical features is calculated by statistical analysis or machine learning methods, for example, using Pearson correlation coefficient or Spearman rank correlation coefficient to measure the linear or nonlinear relationship between the two; the correlation between geological features and geochemical features is calculated, using similar methods, such as correlation coefficient or mutual information; the correlation between geological features and remote sensing features is calculated, using image processing and feature matching techniques, such as SIFT or SURF feature matching; the correlation between geophysical features and geochemical features is calculated, using multivariate statistical analysis methods, such as principal component analysis (PCA) or factor analysis; the correlation between geophysical features and remote sensing features is calculated, using image processing and feature matching techniques, such as spectral feature matching; the correlation between geochemical features and remote sensing features is calculated, using multivariate statistical analysis methods, such as linear discriminant analysis (LDA) or support vector machine (SVM). Through the above method, the correlation information between each pair of features is obtained, and the correlation information is represented in the form of a matrix, wherein each element represents the correlation between a pair of features.

[0093] Based on the correlation information, the similarity between geophysical features, geochemical features, remote sensing features and geological features is calculated, for example, using methods such as cosine similarity, Euclidean distance or Jaccard similarity. The calculated similarity information is compared with the preset threshold to screen out features with a similarity greater than the threshold. For example, the similarity threshold is set to 0.7, indicating that only features with a similarity greater than 0.7 are considered to be highly correlated with geological features. GIS software (such as ArcGIS) is used to match the screened similarity information with geographic spatial data to generate a copper mine distribution map. The similarity information is displayed on the map through the visualization function of the GIS software, and different colors or symbols are used to represent different similarity values, helping geologists to understand the data more intuitively. Useful information is extracted from multi-source data, that is, by correlating geological features, geophysical features, geochemical features and remote sensing features, similarity information is calculated, and the similarity information is filled into the copper mine distribution map based on GIS technology, and a comprehensive analysis is performed in combination with geological information to improve the efficiency and accuracy of copper mineral resource exploration.

[0094] S14, constructing a multimodal copper ore prediction model, inputting the copper ore distribution map into the multimodal copper ore prediction model for training, and obtaining a trained multimodal copper ore prediction model.

[0095] In this embodiment, the copper ore distribution map is input into the constructed multimodal copper ore prediction model for training to obtain a trained model. First, image features are extracted from the copper ore distribution map by using methods such as convolutional neural networks. Secondly, based on a deep learning model of multimodal fusion, the extracted multimodal features are fused by methods such as weighted averaging, splicing, and attention mechanisms. The training set data is used to train the model, and the hyperparameters are adjusted to improve the convergence speed and prediction accuracy. The model performance is then evaluated using the validation set data, and relevant indicators are calculated to ensure a good prediction effect. Finally, a trained multimodal copper ore prediction model is obtained.

[0096] S15. Input the newly acquired copper ore geological report into the multimodal copper ore prediction model to obtain a copper ore distribution prediction map for the target area.

[0097] In this embodiment, the copper mine geological report data after pretreatment and feature extraction is input into the trained multimodal copper mine prediction model, a copper mine distribution prediction map is generated and a favorable prediction area is delineated, and the reliability of the prediction results is analyzed in combination with the mineral geological data, so as to determine the prospecting prediction area to identify the potential copper mine area, and compare with the known area to verify the model effect, and the potential area can also be selected for field exploration to further verify the accuracy and reliability of the model, so as to improve the accuracy and efficiency of copper mine prediction. Finally, the model and data are continuously updated according to the field exploration results, and the prediction accuracy of the model is continuously optimized so that it can adapt to the new geological environment and data changes. The copper mine distribution prediction map contains the reserves and distribution of copper mine resources, and marks different types of copper deposits, such as porphyry type, sedimentary ore-bearing type, etc. In addition, the prediction map will also show the proportion of the predicted copper deposit area in the study area, delineate the prospective mineralization area, and provide scientific basis and direction for geological exploration and resource development. These data are obtained through the analysis and prediction of the multimodal copper mine prediction model, which helps to understand the potential distribution and development value of copper mine resources.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A copper mine prediction method based on multi-source data, characterized in that: The method comprises: S11, using a copper mine geological report as a data source and preprocessing the data source, wherein the copper mine geological report includes a geological map, geophysical data, geochemical sample analysis results, and remote sensing images; S12, respectively performing vector processing on the geological map, geophysical data, geochemical sample analysis results and remote sensing images to obtain corresponding geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information; S13, respectively inputting geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information into a machine learning model for feature extraction, and filling the extracted features into a copper mine distribution map; S14, constructing a multimodal copper ore prediction model, inputting the copper ore distribution map into the multimodal copper ore prediction model for training, and obtaining a trained multimodal copper ore prediction model; S15. Input the newly acquired copper ore geological report into the multimodal copper ore prediction model to obtain a copper ore distribution prediction map for the target area.

2. The copper mine prediction method based on multi-source data according to claim 1, characterized in that: In step S11, the data source is preprocessed, specifically including: The geological maps, geophysical data, geochemical sample analysis results and remote sensing images were cleaned, standardized and normalized respectively.

3. The copper mine prediction method based on multi-source data according to claim 1 is characterized in that: In step S12, the geological map is vectorized, which specifically includes: S21, dividing the geological map into m rows×n columns of grids according to a preset grid unit size, each grid being a geological unit, and each geological unit including a geological element and a geological attribute corresponding to the geological element; S22, constructing a corpus based on regional geological data, and inputting the corpus constructed by the regional geological data into deep learning for training to obtain a geological language model; S23, inputting the geological attributes of the geological elements in each geological unit into the geological language model to obtain a vectorized representation of the geological attributes of the geological elements in each geological unit; S24. Superimpose the vectors of different geological elements, convert the geological attribute vector information of the geological elements into geological vector information through the geological language model, and assign the geological vector information to each corresponding geological unit.

4. The copper mine prediction method based on multi-source data according to claim 3 is characterized in that: In step S12, the geophysical data is vectorized, which specifically includes: S31, digitizing the collected geophysical data, and converting the coordinates of the digitized geophysical data into geographic absolute coordinates; S32, interpolating and resampling the data under the geographic absolute coordinates, completing the matching of the time coordinates and the spatial coordinates of the geophysical data, and establishing a unified coordinate database according to the matching results; S33. The data in the coordinate database and the geological vector information in the geological unit are feature extracted and fused by using multi-source data information fusion technology to obtain corresponding geophysical vector information.

5. The copper mine prediction method based on multi-source data according to claim 3 is characterized in that: In step S12, the geochemical sample analysis results are vectorized, specifically including: S41, performing principal component analysis on the geochemical sample analysis results to determine the correlation between each principal component and the target geochemical element variable; S42, taking the score of each sample on the principal component as a new eigenvector to form a principal component feature matrix, wherein each row of the feature matrix represents a representation of a sample in the principal component space, i.e., a principal component eigenvector; S43. The principal component characteristic vector is extracted and fused with the geological vector information in the geological unit through multi-source data information fusion technology to obtain the corresponding geochemical vector information.

6. The copper mine prediction method based on multi-source data according to claim 1, characterized in that: In step S12, the remote sensing image is vectorized, which specifically includes: S51, extracting features of the remote sensing image through a deep convolutional neural network to obtain a global image feature vector and a local image feature vector; S52, performing cross attention calculation on the global image feature vector and the local image feature vector to determine the target feature vector; S53. Extract and fuse the target feature vector with the geological vector information in the geological unit through multi-source data information fusion technology to obtain corresponding remote sensing vector information.

7. The copper mine prediction method based on multi-source data according to claim 1, characterized in that: In step S13, the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information are respectively input into the machine learning model for feature extraction, specifically including: S61, dynamically learning the optimal weights of geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information through an automatic weighting framework of reinforcement learning; S62. Based on the optimal weights, feature extraction is performed on the geological vector information, geophysical vector information, geochemical vector information and remote sensing vector information through a neural network to obtain geological features, geophysical features, geochemical features and remote sensing features respectively.

8. The copper mine prediction method based on multi-source data according to claim 1, characterized in that: In step S13, the extracted features are respectively filled into the copper ore distribution map, specifically including: S71, respectively performing correlation processing on the geological features, geophysical features, geochemical features and remote sensing features in pairs to obtain correlation information; S72, taking the geological feature as the target feature, calculating the similarity between the geophysical feature, the geochemical feature, and the remote sensing feature and the target feature based on the correlation information, and obtaining similarity information; S73. Compare the similarity information with a preset threshold value, and fill the similarity information into the copper mine distribution map according to the comparison result based on GIS technology.

Images