Deep mineral prediction method based on hyperspectral and geophysical data fusion

Abstract

The invention provides a deep mineral prediction method based on hyperspectral and geophysical data fusion, and the method comprises the steps: firstly collecting hyperspectral remote sensing data and geophysical data of a working area, and carrying out the preprocessing of the hyperspectral remote sensing data and the geophysical data; carrying out feature extraction on the hyperspectral remote sensing data and the geophysical data; carrying out fusion processing on the hyperspectral remote sensing data and the geophysical data; and finally, establishing a deep mineral prediction model, training and verifying the deep mineral prediction model, and carrying out deep anomaly recognition and target region delineation. The method is reasonable in conception, the hyperspectral data and the geophysical data can be conveniently fused, the advantages of the two methods are fully utilized, and the efficiency and the accuracy of mineral resource exploration can be improved; the method provides support in the aspects of enhancing the recognition capability, expanding the detection depth, improving the prediction precision, reducing the exploration cost, accelerating the exploration speed, improving the prospecting success rate and supporting sustainable development.

Description

Technical Field

[0001] The present invention belongs to the technical field of mineral prediction, and in particular relates to a deep mineral prediction method based on the fusion of hyperspectral and geophysical data. Background Art

[0002] Hyperspectral remote sensing is an advanced remote sensing technology that identifies and analyzes surface materials by acquiring the reflectance or radiation characteristics of surface objects in continuous spectral bands. This technology can detect very subtle spectral differences to distinguish different rocks, minerals, vegetation, etc. In mineral prediction, hyperspectral remote sensing technology can be used to identify altered minerals and rock types associated with mineralization, because these materials may show unique characteristics in the spectrum. For example, by analyzing aerial hyperspectral remote sensing data from Liuyuan-Fangshankou in Beishan, Gansu and Xuemistan area in Xinjiang, researchers were able to identify the presence of solid minerals and established prediction methods such as mineralization environment analysis method, ore deposit location model identification method and mineralization structure tracking method based on aerial hyperspectral remote sensing.

[0003] Geophysical exploration technology detects geological conditions such as stratum lithology and geological structure by studying and observing changes in various geophysical fields. Since different rock media that make up the earth's crust often differ in density, elasticity, conductivity, magnetism, radioactivity, and thermal conductivity, these differences will cause local changes in the corresponding geophysical fields. By measuring the distribution and variation characteristics of these physical fields and combining them with known geological data for analysis and research, the purpose of inferring geological properties can be achieved. Geophysical exploration techniques include gravity exploration, magnetic exploration, electrical exploration, and seismic exploration. These methods can be used to determine the potential location and distribution of mineral resources in mineral prediction.

[0004] Geophysical methods such as gravity, magnetic, electromagnetic and seismic exploration can detect deeper underground structures, while hyperspectral remote sensing technology is highly efficient and accurate in identifying surface materials. Data fusion helps expand the detection range, especially in deep mineral exploration, and can more effectively identify and predict deep hidden ore bodies, improve the accuracy and efficiency of mineral resource detection, and has important practical application value for the exploration and development of resources and energy.

[0005] However, the traditional or existing technologies mainly have the following disadvantages or shortcomings:

[0006] (1) Low data utilization efficiency

[0007] Traditional mineral exploration methods often only utilize limited data types and fail to fully utilize the potential of various data;

[0008] (2) Limited detection depth

[0009] Traditional surface exploration methods have difficulty detecting deep ore bodies;

[0010] (3) Low prediction accuracy

[0011] Due to the lack of effective data fusion and analysis methods, traditional forecasting models often have low prediction accuracy;

[0012] (4) High exploration costs

[0013] Traditional exploration methods often require a lot of manpower and material resources, resulting in high exploration costs;

[0014] (5) Slow exploration speed

[0015] Due to the limitations of exploration methods and technologies, traditional exploration is often slow. Summary of the invention

[0016] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a deep mineral prediction method based on the fusion of hyperspectral and geophysical data. The method has a reasonable conception and can conveniently integrate hyperspectral and geophysical data, make full use of the advantages of the two methods, and improve the efficiency and accuracy of mineral resource exploration. It can provide support in enhancing recognition ability, expanding detection depth, improving prediction accuracy, reducing exploration costs, accelerating exploration speed, improving the success rate of prospecting and supporting sustainable development.

[0017] To solve the above technical problems, the present invention provides a deep mineral prediction method based on the fusion of hyperspectral and geophysical data, which first collects hyperspectral remote sensing data and geophysical data of the working area, and preprocesses the hyperspectral remote sensing data and geophysical data respectively; then performs feature extraction of the hyperspectral remote sensing data and geophysical data; then performs fusion processing on the hyperspectral remote sensing data and geophysical data; finally, establishes a deep mineral prediction model, trains and verifies the deep mineral prediction model, and carries out deep anomaly identification and target area delineation.

[0018] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the specific process of preprocessing the hyperspectral remote sensing data and the geophysical data respectively is: completing radiation correction, atmospheric correction and geometric correction for the hyperspectral remote sensing data to ensure data quality; at the same time, performing element content analysis on geochemical samples, and carrying out data cleaning, standardization and normalization processing.

[0019] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the process of extracting features from hyperspectral remote sensing data and geophysical data is: extracting mineral spectral features from hyperspectral remote sensing data, and extracting geophysical features from geophysical data.

[0020] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein: the mineral spectral characteristics include absorption characteristics and reflectivity; the geophysical characteristics include velocity, density and resistivity.

[0021] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the specific process of fusing the hyperspectral remote sensing data and the geophysical data is as follows:

[0022] (3.1) Statistical Fusion

[0023] Hyperspectral remote sensing data and geophysical data are both continuous data. They can be approximated as Gaussian distributions to calculate the statistical characteristics of the target minerals. Then, multiple Gaussian distributions are unified under the Bayesian framework to construct the posterior probability distribution function:

[0024] p(md)=const×p(dm)p(m);

[0025] Among them, the prior information is the product of the Gaussian distribution of hyperspectral remote sensing data and the Gaussian distribution of geophysical data, which can be expressed as:

[0026] p(m)=p1(m)p2(m),;

[0027] (3.2) Fusion of hyperspectral and geophysical data relationships

[0028] Integrate hyperspectral remote sensing data and geophysical data to establish the relationship between mineral spectral characteristics and geophysical characteristics, which facilitates the subsequent transformation of mineral spectral characteristics and geophysical characteristics.

[0029] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the process of approximating the hyperspectral remote sensing data and the geophysical data to Gaussian distribution in step (3.1) is:

[0030] Let d represent the data vector and m represent the model parameter vector; assume that the theoretical relationship between the data and the parameters of the model describing the underground conditions with physical parameters is represented by d = g(m); the observed value of the true value of d is represented by d obs Indicates that the prior information of m is expressed as m pri express;

[0031] It is assumed that the errors are independently distributed and conform to the zero-mean Gaussian distribution, with standard deviations of σ d and σ m , then the posterior probability distribution can be expressed as follows:

[0032]

[0033] Unify various Gaussian distributions into the Bayesian framework and select the random variable expression used in geophysical inversion. Let m and d be two random variables, p(dm) is the conditional probability of d to m, p(m) is the marginal probability of m, then the conditional probability of m to d is:

[0034]

[0035] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the specific process of step (3.2) is as follows:

[0036] (3.2.1) Feature extraction

[0037] Extracting hyperspectral data features, that is, extracting mineral spectral features, using principal component analysis to reduce dimensionality, and extracting principal components as features; extracting geophysical data features, that is, extracting geophysical features, using wavelet transform to extract time-frequency features of geophysical data;

[0038] (3.2.2) Selecting a deep learning model

[0039] A combined model of a convolutional neural network and a long short-term memory network is selected, and the model architecture of the combined model includes an input layer, a hidden layer, a fusion layer and an output layer; the input layer inputs the hyperspectral data features and the geophysical data features respectively; the hidden layer uses the convolutional neural network to extract the local features of the hyperspectral data, and uses the long short-term memory network to extract the time series features of the geophysical data; the fusion layer fuses the outputs of the convolutional neural network and the long short-term memory network; the output layer outputs the fusion features of the mineral spectral features and the geophysical features;

[0040] (3.2.3) Establishing feature relationships

[0041] Use known geological label data to train a combined model of convolutional neural network and long short-term memory network, learn the mapping relationship between hyperspectral data and geophysical data, and then use the cross entropy loss function to optimize the combined model of convolutional neural network and long short-term memory network;

[0042] (3.2.4) Feature Fusion

[0043] Use a deep fusion strategy to weight the output features of the convolutional neural network and the long short-term memory network, and then use the attention mechanism to automatically learn the importance of different features;

[0044] (3.2.5) Model training and validation

[0045] First, train the data set, which is divided into training set, validation set and test set; then select hyperparameters, including learning rate, batch size and number of iterations;

[0046] (3.2.6) Result analysis

[0047] Taking accuracy, recall and F1 score as evaluation indicators, the performance of the combined model of convolutional neural network and long short-term memory network was first evaluated on the validation set, the hyperparameters were adjusted, and then the generalization ability of the combined model of convolutional neural network and long short-term memory network was tested on the test set.

[0048] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the specific process of establishing the deep mineral prediction model is: selecting a suitable machine learning algorithm to construct a prediction model, that is, a combination model of a convolutional neural network and a long short-term memory network, and then tuning the parameters of the combination model of the convolutional neural network and the long short-term memory network to find the optimal parameter combination to improve the prediction performance of the combination model of the convolutional neural network and the long short-term memory network.

[0049] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the tuning of the model parameters is to use Bayesian optimization to tune the parameters of the combined model of the convolutional neural network and the long short-term memory network, and the specific process is:

[0050] (4.01) Define optimization goals

[0051] Define an objective function, select the validation set loss rate or validation set accuracy of the combined model of convolutional neural network and long short-term memory network as the optimization target, and then determine the parameters to be optimized and their value ranges;

[0052] (4.02) Select the Bayesian optimization library;

[0053] (4.03) Preparing the model

[0054] Define the combined model architecture of convolutional neural network and long short-term memory network, and prepare training data and validation data;

[0055] (4.04) Define the hyperparameter search space

[0056] Define convolutional neural network parameters, long short-term memory network parameters and training parameters;

[0057] (4.05) Configuring Bayesian Optimization

[0058] Set up a random exploration step to obtain initial data points, choose an acquisition function to guide subsequent exploration, and choose a Bayesian optimization algorithm;

[0059] (4.06) Run Bayesian Optimization

[0060] Use the Bayesian optimization algorithm to find the optimal parameters in the defined search space, set the number of iterations, and each iteration will select new parameters for trial based on the acquisition function;

[0061] (4.07) Model training and evaluation

[0062] Use the parameter training model recommended by Bayesian optimization and evaluate the performance of the parameter training model on the validation set;

[0063] (4.08) Result Analysis

[0064] Record and analyze the optimal parameters found by Bayesian optimization, and compare the performance of the combined model of convolutional neural network and long short-term memory network under the optimal parameters with the baseline model;

[0065] (4.09) Model verification

[0066] Use the test set to verify the performance of the combined model of convolutional neural network and long short-term memory network under the optimal parameters;

[0067] (4.10) Model deployment

[0068] The trained combined model of convolutional neural network and long short-term memory network is saved for subsequent use, and the combined model of convolutional neural network and long short-term memory network is applied to actual mineral prediction tasks.

[0069] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein: the machine learning algorithm includes support vector machine, random forest, neural network and LightGBM; the method for tuning the model parameters includes grid search, random search and Bayesian optimization.

[0070] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the process of model training and verification is: using known geological and mineral data to train the prediction model, and using geochemical methods to verify unknown points to evaluate the prediction accuracy of the prediction model.

[0071] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data, wherein the process of carrying out deep anomaly identification and target area delineation is: by fusing hyperspectral remote sensing data and geophysical data, establishing a deep mineral prediction model, and carrying out training and verification of the deep mineral prediction model, hyperspectral anomalies, geophysical anomalies and deep mineralization probability maps are obtained; and the prediction model is used to identify the areas of geophysical anomalies and hyperspectral anomalies;

[0072] A constraint condition must be added in the deep anomaly identification link: it is assumed that the distribution of solid minerals within a certain scale has a certain continuity, and a Markov random field neighborhood system is used to constrain it. When determining whether a certain point is an anomaly, it is necessary to consider the relationship between hyperspectral remote sensing data and geophysical data established by machine learning in the neighborhood system to make a comprehensive judgment; the prospecting target area is delineated by fusing hyperspectral remote sensing data and geophysical data, establishing a deep mineral prediction model, and training and verifying the deep mineral prediction model.

[0073] By adopting the above technical solution, the present invention has the following beneficial effects:

[0074] The present invention utilizes the fusion of hyperspectral data and geophysical data to carry out solid mineral prediction, innovatively integrates the two types of data, and introduces geochemical element analysis for verification, because it can provide more comprehensive and accurate geological information and can improve the accuracy of mineral exploration.

[0075] The present invention makes full use of the low-cost exploration technology of hyperspectral to expand the detection range. On this basis, it can also make full use of the advantages of geophysical exploration methods in detection depth to improve the success rate of deep mineral exploration.

[0076] The present invention can effectively improve prediction efficiency and reduce exploration costs, and at the same time establishes a new data integration method to improve data utilization efficiency.

[0077] The present invention aims to overcome the above-mentioned shortcomings and achieve more efficient, accurate and economical mineral exploration by introducing the fusion of hyperspectral and geophysical data and the application of machine learning algorithms.

[0078] The present invention can conveniently integrate hyperspectral and geophysical data, fully utilize the advantages of the two methods, and improve the efficiency and accuracy of mineral resource exploration; it can provide support in enhancing recognition capabilities, expanding detection depth, improving prediction accuracy, reducing exploration costs, accelerating exploration speed, increasing the success rate of prospecting, and supporting sustainable development.

[0079] Compared with the conventional technology, the present invention also has the following characteristics or advantages:

[0080] (1) Improve the efficiency and accuracy of mineral resource exploration

[0081] Traditional mineral exploration methods are often inefficient and have limited accuracy. The present invention can more comprehensively analyze geological information by fusing hyperspectral and geophysical data, thereby improving the efficiency and accuracy of exploration.

[0082] (2) Enhance recognition capabilities and expand detection depth

[0083] Hyperspectral remote sensing technology has high efficiency and high precision in identifying surface materials, while geophysical exploration technology has advantages in detection depth; the present invention can enhance the recognition capability and expand the detection depth through data fusion;

[0084] (3) Reduce exploration costs and speed up exploration

[0085] Traditional exploration methods are often costly and time-consuming; the present invention utilizes the low-cost advantage of hyperspectral data and combines it with geophysical data to speed up exploration while reducing costs;

[0086] (4) Improve the success rate of mineral exploration

[0087] By combining the advantages of the two types of data, the present invention can more accurately identify and predict deep hidden ore bodies, thereby improving the success rate of prospecting;

[0088] (5) Supporting sustainable development

[0089] The rational development and utilization of mineral resources are crucial to sustainable development; the present invention helps to achieve the rational development of mineral resources by improving exploration efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0091] Figure 1 It is a schematic diagram of a Markov random field neighborhood system involved in the deep mineral prediction method based on the fusion of hyperspectral and geophysical data of the present invention;

[0092] Figure 2 It is a schematic diagram of Gaussian probability distribution involved in the deep mineral prediction method based on the fusion of hyperspectral and geophysical data of the present invention. DETAILED DESCRIPTION

[0093] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0094] The present invention is further explained below in conjunction with specific implementation modes.

[0095] like Figure 1 As shown, the present embodiment provides a deep mineral prediction method based on the fusion of hyperspectral and geophysical data, which is to first collect hyperspectral remote sensing data and geophysical data of the working area, and pre-process the hyperspectral remote sensing data and geophysical data respectively; then perform feature extraction of the hyperspectral remote sensing data and geophysical data; then perform fusion processing on the hyperspectral remote sensing data and geophysical data; finally, establish a deep mineral prediction model, train and verify the deep mineral prediction model, and perform deep anomaly identification and target area delineation.

[0096] The deep mineral prediction method based on the fusion of hyperspectral and geophysical data of the present invention specifically includes the following steps:

[0097] (1) Data collection and preprocessing

[0098] Collect hyperspectral remote sensing data and geophysical data in the work area; complete radiation correction, atmospheric correction and geometric correction (these correction methods are all conventional correction methods) on the hyperspectral remote sensing data to ensure data quality; analyze the element content of geochemical samples (geochemical samples are separately collected data that can directly point to whether they contain minerals, so they are used for verification), and carry out data cleaning, standardization and normalization (these processing methods all use conventional means commonly used in big data mining).

[0099] (2) Feature extraction

[0100] Extract mineral spectral features, such as absorption features, reflectivity, etc., from hyperspectral remote sensing data; extract geophysical features, such as velocity, density, resistivity, etc., from geophysical data.

[0101] (3) Data Fusion

[0102] (3.1) Statistical Fusion

[0103] Hyperspectral remote sensing data and geophysical data are both continuous data. Hyperspectral remote sensing data and geophysical data can be approximated as Gaussian distribution to calculate the statistical characteristics of the target minerals. The aforementioned process of approximating hyperspectral remote sensing data and geophysical data as Gaussian distribution is:

[0104] Let d represent the data vector and m represent the model parameter vector; assume that the theoretical relationship between the data and the parameters of the model describing the underground conditions with physical parameters is represented by d = g(m); the observed value of the true value of d is represented by d obs Indicates that the prior information of m is expressed as m pri express;

[0105] It is assumed that the errors are independently distributed and conform to the zero-mean Gaussian distribution, with standard deviations of σ d and σ m, then the posterior probability distribution can be expressed as follows:

[0106]

[0107] Various Gaussian distributions are unified under the Bayesian framework, and the Bayesian formula has several forms. Here we mainly introduce the expression of random variables used in geophysical inversion. Let m and d be two random variables, p(dm) is the conditional probability of d to m, p(m) is the marginal probability of m, so the conditional probability of m to d is:

[0108]

[0109] Construct the posterior probability distribution function:

[0110] p(md)=const×p(dm)p(m);

[0111] Among them, the prior information is the product of the Gaussian distribution of hyperspectral remote sensing data and the Gaussian distribution of geophysical data, which can be expressed as:

[0112] p(m)=p1(m)p2(m),.

[0113] (3.2) Fusion of hyperspectral remote sensing data and geophysical data

[0114] Through deep learning methods, we integrate hyperspectral data and seismic data to establish the relationship between mineral spectral characteristics and geophysical characteristics, which facilitates the subsequent transformation between the two. The specific process is as follows:

[0115] (3.2.1) Feature extraction

[0116] Hyperspectral data feature extraction: To extract mineral spectral features, such as absorption features, reflectivity, etc., use principal component analysis (PCA) to reduce dimensionality, and extract the first few principal components (such as absorption features, reflectivity, etc.) as features; Geophysical data feature extraction: Extract geophysical features such as seismic wave velocity, amplitude, frequency, etc., and use wavelet transform to extract the time-frequency characteristics of geophysical data.

[0117] (3.2.2) Selecting a deep learning model

[0118] Model selection: Select a combined model of convolutional neural network (CNN) and long short-term memory network (LSTM).

[0119] The model architecture is:

[0120] Input layer: input hyperspectral data features and geophysical data features respectively;

[0121] Hidden layer: CNN is used to extract local features of hyperspectral data, and LSTM is used to extract time series features of geophysical data;

[0122] Fusion layer: fuses the outputs of CNN and LSTM;

[0123] Output layer: Outputs the fusion features of mineral spectral features and geophysical features.

[0124] (3.2.3) Establishing feature relationships

[0125] Model training: Use known geological label data to train a combined model of convolutional neural networks and long short-term memory networks to learn the mapping relationship between hyperspectral data and geophysical data; use the cross entropy loss function to optimize the above combined model.

[0126] (3.2.4) Feature Fusion

[0127] Feature fusion strategy: Use the deep fusion strategy to weightedly fuse the output features of CNN and LSTM; use the attention mechanism to automatically learn the importance of different features.

[0128] (3.2.5) Model training and validation

[0129] Training data set: The data set is divided into training set (70%), validation set (15%) and test set (15%); hyperparameter selection: learning rate is 0.001; batch size is 32; number of iterations is 100

[0130] (3.2.6) Result analysis

[0131] Evaluation indicators: accuracy, recall, F1 score (F1 score is an indicator for evaluating the performance of classification models, especially when the data set is imbalanced; it takes into account the accuracy and recall of the model and provides a single number to measure the overall performance of the model); Model evaluation: Evaluate model performance on the validation set and adjust hyperparameters; Test the generalization ability of the model on the test set.

[0132] (4) Establishing a deep mineral prediction model

[0133] Select a suitable machine learning algorithm to build a prediction model; among them, commonly used machine learning algorithms include support vector machine (SVM), random forest, neural network and LightGBM, etc.; using the parameters of the above fusion process, select the combined model of convolutional neural network (CNN) and long short-term memory network (LSTM) supplemented above as the prediction model; use methods such as grid search, random search or Bayesian optimization to tune the parameters of the prediction model and find the optimal parameter combination to improve the prediction performance of the prediction model; in terms of parameter tuning, Bayesian optimization is a very effective method, especially when dealing with complex models with a large number of parameters; the following is a specific process of using Bayesian optimization to tune the parameters of the combined model of convolutional neural network (CNN) and long short-term memory network (LSTM):

[0134] (4.01) Define optimization goals

[0135] Objective function: Define an objective function, usually the validation set loss or validation set accuracy of the combined model of the convolutional neural network and the long short-term memory network, as the optimization target.

[0136] Parameter range: Determine the parameters that need to be optimized and their value range.

[0137] (4.02) Select the Bayesian Optimization Library

[0138] Library selection: Choose a Bayesian optimization library such as GPyOpt, Hyperopt, or Sc i kit-opt imi ze.

[0139] (4.03) Preparing the model

[0140] Model definition: Define the combined model architecture of CNN and LSTM.

[0141] Data preparation: Prepare training data and validation data.

[0142] (4.04) Define the hyperparameter search space

[0143] CNN parameters: such as the number of convolutional layers, the number of filters in each layer, the filter size, the type of pooling layer, etc.

[0144] LSTM parameters: such as the number of LSTM units, the number of layers, etc.

[0145] Training parameters: such as learning rate, batch size, regularization parameters, etc.

[0146] (4.05) Configuring Bayesian Optimization

[0147] Initial random exploration: Set a certain number of random exploration steps to obtain initial data points.

[0148] Collection Function: Select a collection function, such as expected improvement (EI), probability density on a Gaussian process, etc., to guide subsequent exploration.

[0149] Optimization Algorithm: Select a Bayesian optimization algorithm such as Gaussian Process (GP).

[0150] (4.06) Run Bayesian Optimization

[0151] Optimization process: Use the Bayesian optimization algorithm to find the optimal parameters in the defined search space.

[0152] Iterations: Set the number of iterations. Each iteration will select new parameters for trial based on the acquisition function.

[0153] (4.07) Model training and evaluation

[0154] Train Model: Train the model using the parameters recommended by Bayesian optimization.

[0155] Evaluate Model: Evaluate the performance of the parameter trained model on the validation set.

[0156] (4.08) Result Analysis

[0157] Optimal parameters: Record and analyze the optimal parameters found by Bayesian optimization.

[0158] Performance comparison: The performance of the combined model of Convolutional Neural Network and Long Short-Term Memory Network with optimal parameters is compared with the baseline model.

[0159] (4.09) Model verification

[0160] Test set verification: Use the test set to verify the performance of the combined model of convolutional neural network and long short-term memory network under the optimal parameters.

[0161] (4.10) Model deployment

[0162] Model saving: Save the trained convolutional neural network and long short-term memory network combined model for subsequent use.

[0163] Application model: The combined model of convolutional neural network and long short-term memory network is applied to actual mineral prediction tasks.

[0164] (5) Model training and verification

[0165] The prediction model is trained using known geological and mineral data (conventional training methods are used, and the whole process includes dividing the training set and validation set, model selection, parameter initialization, parameter optimization, etc.), and the unknown points are verified using geochemical methods to evaluate the prediction accuracy of the prediction model.

[0166] (6) Deep abnormality identification and target area delineation

[0167] Through the above steps (3)-(5), three results are obtained: hyperspectral anomaly, geophysical anomaly, and deep mineralization probability map; the prediction model is used to identify geophysical anomaly and hyperspectral anomaly areas; a constraint condition is added to the deep anomaly identification link: it is believed that the distribution of solid minerals within a certain scale has a certain continuity, and the Markov random field neighborhood system is used for constraint. When determining whether a certain point is an abnormal point, it is necessary to consider the relationship between hyperspectral and geophysical data established by machine learning in the neighborhood system to make a comprehensive judgment (the above step (3)); the prospecting target area is delineated by carrying out the above steps (3)-(5);

[0168] Markov random field is an extension of the one-dimensional Markov process; for the underground profile on the two-dimensional plane, it can be regarded as a Markov random field;

[0169] Let S = {(i, j)|1≤i≤M,1≤j≤N} (M and N are the number of rows and columns in the discrete two-dimensional space) be a set of spatial locations defined on the underground profile, if the set δ = {δ(s)|s∈S} satisfies the following conditions:

[0170]

[0171] Then δ is called the set of neighborhood systems defined on S. In this case, s is the center point, δ(s) is the neighborhood point, and the order of the neighborhood system is defined by the distance between the neighborhood point and the center point. The distance here is usually the Euclidean distance. Figure 1 As shown in the figure, whether each central point contains minerals or not is related to the points in its neighborhood system. Through this constraint, a certain degree of continuity can be guaranteed.

[0172] The present invention has a reasonable concept, can conveniently integrate hyperspectral and geophysical data, make full use of the advantages of the two methods, and can improve the efficiency and accuracy of mineral resource exploration; it provides support in enhancing recognition capabilities, expanding detection depth, improving prediction accuracy, reducing exploration costs, accelerating exploration speed, increasing the success rate of prospecting, and supporting sustainable development.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep mineral prediction method based on the fusion of hyperspectral and geophysical data, characterized by: First, collect the hyperspectral remote sensing data and geophysical data of the working area, and pre-process the hyperspectral remote sensing data and geophysical data respectively; then carry out feature extraction of hyperspectral remote sensing data and geophysical data; then fuse the hyperspectral remote sensing data and geophysical data; finally, establish a deep mineral prediction model, train and verify the deep mineral prediction model, and carry out deep anomaly identification and target area delineation.

2. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1 is characterized in that: The specific process of preprocessing the hyperspectral remote sensing data and geophysical data is as follows: completing radiation correction, atmospheric correction and geometric correction for the hyperspectral remote sensing data to ensure data quality; while performing element content analysis on geochemical samples, and carrying out data cleaning, standardization and normalization processing.

3. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1 is characterized in that: The process of extracting features from hyperspectral remote sensing data and geophysical data is as follows: extracting mineral spectral features from hyperspectral remote sensing data and extracting geophysical features from geophysical data.

4. The deep mineral prediction method based on the fusion of hyperspectral and geophysical data as claimed in claim 3 is characterized by: The mineral spectral characteristics include absorption characteristics and reflectivity; the geophysical characteristics include velocity, density and resistivity.

5. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1 is characterized in that: The specific process of fusing the hyperspectral remote sensing data and geophysical data is as follows: (3.1) Statistical Fusion Hyperspectral remote sensing data and geophysical data are both continuous data, and the statistical characteristics of target minerals can be calculated by approximating them to Gaussian distribution; Then, multiple Gaussian distributions are unified into the Bayesian framework to construct the posterior probability distribution function: p(md)=const×p(dm)p(m); Among them, the prior information is the product of the Gaussian distribution of hyperspectral remote sensing data and the Gaussian distribution of geophysical data, which can be expressed as: p(m)=p1(m)p2(m)…; (3.2) Fusion of hyperspectral and geophysical data relationships Integrate hyperspectral remote sensing data and geophysical data to establish the relationship between mineral spectral characteristics and geophysical characteristics, which facilitates the subsequent transformation of mineral spectral characteristics and geophysical characteristics.

6. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 5 is characterized in that: The process of approximating the hyperspectral remote sensing data and geophysical data to Gaussian distribution in step (3.1) is as follows: Let d represent the data vector and m represent the model parameter vector; let d = g(m) represent the theoretical relationship between the data and the parameters of the model that describes the underground conditions with physical parameters; let d represent the observed value of the true value of d. obs Indicates that the prior information of m is expressed as m pri express; It is assumed that the errors are independently distributed and conform to the zero-mean Gaussian distribution, with standard deviations of σ d and σ m , then the posterior probability distribution can be expressed as follows: Unify various Gaussian distributions into the Bayesian framework and select the random variable expression used in geophysical inversion. Let m and d be two random variables, p(dm) is the conditional probability of d to m, p(m) is the marginal probability of m, then the conditional probability of m to d is:

7. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 5 is characterized in that: The specific process of step (3.2) is: (3.2.1) Feature extraction Extracting hyperspectral data features, that is, extracting mineral spectral features, using principal component analysis to reduce dimensionality, and extracting principal components as features; extracting geophysical data features, that is, extracting geophysical features, using wavelet transform to extract time-frequency features of geophysical data; (3.2.2) Selecting a deep learning model A combined model of a convolutional neural network and a long short-term memory network is selected, wherein the model architecture of the combined model includes an input layer, a hidden layer, a fusion layer and an output layer; the input layer inputs the hyperspectral data features and the geophysical data features respectively; the hidden layer uses the convolutional neural network to extract the local features of the hyperspectral data, and uses the long short-term memory network to extract the time series features of the geophysical data; The fusion layer fuses the outputs of the convolutional neural network and the long short-term memory network; the output layer outputs the fusion features of the mineral spectral features and the geophysical features; (3.2.3) Establishing feature relationships Use known geological label data to train a combined model of convolutional neural network and long short-term memory network, learn the mapping relationship between hyperspectral data and geophysical data, and then use the cross entropy loss function to optimize the combined model of convolutional neural network and long short-term memory network; (3.2.4) Feature Fusion Use a deep fusion strategy to weight the output features of the convolutional neural network and the long short-term memory network, and then use the attention mechanism to automatically learn the importance of different features; (3.2.5) Model training and validation First, train the data set, which is divided into training set, validation set and test set; then select hyperparameters, including learning rate, batch size and number of iterations; (3.2.6) Result analysis Taking accuracy, recall and F1 score as evaluation indicators, the performance of the combined model of convolutional neural network and long short-term memory network was first evaluated on the validation set, the hyperparameters were adjusted, and then the generalization ability of the combined model of convolutional neural network and long short-term memory network was tested on the test set.

8. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1 is characterized in that: The specific process of establishing the deep mineral prediction model is: selecting a suitable machine learning algorithm to construct a prediction model, that is, a combination model of a convolutional neural network and a long short-term memory network, and then tuning the parameters of the combination model of the convolutional neural network and the long short-term memory network to find the optimal parameter combination to improve the prediction performance of the combination model of the convolutional neural network and the long short-term memory network.

9. The deep mineral prediction method based on hyperspectral and geophysical data fusion as claimed in claim 8, characterized in that: The tuning of the model parameters is to use Bayesian optimization to tune the parameters of the combined model of the convolutional neural network and the long short-term memory network. The specific process is: (4.01) Define optimization goals Define an objective function, select the validation set loss rate or validation set accuracy of the combined model of the convolutional neural network and the long short-term memory network as the optimization target, and then determine the parameters to be optimized and their value ranges; (4.02) Select the Bayesian optimization library; (4.03) Prepare the model Define the combined model architecture of convolutional neural network and long short-term memory network, and prepare training data and validation data; (4.04) Define the hyperparameter search space Define convolutional neural network parameters, long short-term memory network parameters and training parameters; (4.05) Configuring Bayesian Optimization Set up a random exploration step to obtain initial data points, choose an acquisition function to guide subsequent exploration, and choose a Bayesian optimization algorithm; (4.06) Run Bayesian Optimization Use the Bayesian optimization algorithm to find the optimal parameters in the defined search space, set the number of iterations, and each iteration will select new parameters for trial based on the acquisition function; (4.07) Model training and evaluation Use the parameter training model recommended by Bayesian optimization and evaluate the performance of the parameter training model on the validation set; (4.08) Result Analysis Record and analyze the optimal parameters found by Bayesian optimization, and compare the performance of the combined model of convolutional neural network and long short-term memory network under the optimal parameters with the baseline model; (4.09) Model verification Use the test set to verify the performance of the combined model of convolutional neural network and long short-term memory network under the optimal parameters; (4.10) Model deployment The trained combined model of convolutional neural network and long short-term memory network is saved for subsequent use, and the combined model of convolutional neural network and long short-term memory network is applied to actual mineral prediction tasks.

10. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 8, characterized in that: The machine learning algorithms include support vector machines, random forests, neural networks and LightGBM; the methods for tuning model parameters include grid search, random search and Bayesian optimization.

11. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1, characterized in that: The model training and verification process is: using known geological and mineral data to train the prediction model, and using geochemical methods to verify unknown points to evaluate the prediction accuracy of the prediction model.

12. The deep mineral prediction method based on hyperspectral and geophysical data fusion according to claim 1, characterized in that: The process of carrying out deep anomaly identification and target area delineation is as follows: by fusing hyperspectral remote sensing data and geophysical data, establishing a deep mineral prediction model, and carrying out training and verification of the deep mineral prediction model, hyperspectral anomalies, geophysical anomalies and deep mineralization probability maps are obtained; and the prediction model is used to identify the areas of geophysical anomalies and hyperspectral anomalies; A constraint condition must be added in the deep anomaly identification link: it is assumed that the distribution of solid minerals within a certain scale has a certain continuity, and a Markov random field neighborhood system is used to constrain it. When determining whether a certain point is an anomaly, it is necessary to consider the relationship between hyperspectral remote sensing data and geophysical data established by machine learning in the neighborhood system to make a comprehensive judgment; the prospecting target area is delineated by fusing hyperspectral remote sensing data and geophysical data, establishing a deep mineral prediction model, and training and verifying the deep mineral prediction model.

Images