Deep learning mineral resource classification prediction method and system based on sample imbalance

Abstract

The present invention proposes a deep learning method and system for mineral resource classification prediction based on sample imbalance. The present invention divides the study area into a mineral deposit area, a mineralization prospect area and a background area. If there is an area with abnormal geochemical data in the mineralization prospect area, it can indirectly indicate that the area has a higher probability of mineralization and shows a higher similarity with the geochemical data characteristics of the known mineral deposit. The deep learning method is used to analyze and determine the area with abnormal geochemical data in the mineralization prospect area, and then the spatial data characteristics of both the known mineral deposit area and the obtained geochemical anomaly area are regarded as learning objects of the neural network to learn the objective laws between geochemical data and mineralization. In order to prevent the neural network from overfitting the spatial data characteristics of the background area, the loss weight and penalty loss are introduced to improve the traditional cross entropy loss function. The present invention can find the mineralization law in the geochemical data of a small number of mineral target areas to predict the probability of mineralization in new geochemical data.

Description

Technical Field

[0001] The present invention relates to a deep learning mineral resource classification prediction method and system based on sample imbalance, belonging to the fields of geoinformation science and computer vision. Background Art

[0002] With the rapid development of my country's economy, the demand for various mineral resources in the industrial sector has gradually increased. How to accurately and efficiently detect mineral resources has become a hot topic in today's academic circles. With the rapid development of artificial intelligence theory and technology, artificial intelligence has gradually entered the field of earth science. With the help of computer vision processing methods, the prediction of mineral target areas has also changed from the original traditional method to intelligent prediction based on artificial intelligence, making the prediction results more accurate and reliable.

[0003] Although the method of predicting mineral targets with the help of artificial intelligence can achieve certain success, key issues such as the small number of mineral areas, small areas, and a relatively large proportion of background areas have a direct impact on the accuracy of intelligent prediction of mineral targets. In addition, traditional geochemical data anomaly analysis methods are difficult to deeply analyze weak anomalies hidden in complex geological environments. Summary of the invention

[0004] Purpose of the invention: In view of the shortcomings of the prior art, the purpose of the present invention is to propose a deep learning mineral resource classification prediction method and system based on sample imbalance, so as to find the relationship between geochemical exploration data and mineralization in a more comprehensive and objective way, find out the mineralization rules in less geochemical data of the mining target area and predict the probability of mineralization in the new geochemical data.

[0005] Technical solution: To achieve the above-mentioned purpose of the invention, the deep learning mineral resource classification prediction method based on sample imbalance described in the present invention comprises the following steps:

[0006] Step 1: Generate geological image data of the target area through geographic location information and geochemical element information;

[0007] Step 2: Divide the study area into ore deposit area, mineralization prospect area and background area;

[0008] Step 3: Analyze the characteristic distribution of geochemical data in the mineralization prospecting area through variational autoencoder and determine the area with abnormal geochemical data in the mineralization prospecting area. If there is an area with abnormal geochemical data in the mineralization prospecting area, mark the category of the area as the same as the category of the nearest mineral deposit area;

[0009] Step 4: Data enhancement of samples of the ore deposit area and the obtained geochemical anomaly area;

[0010] Step 5: Construct a convolutional neural network model to learn the law between geochemical data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and to optimize the network model to learn the mineralization characteristics of geochemical data, the traditional cross entropy loss function is improved by introducing loss weight and penalty loss;

[0011] Step 6: Input the training sample data into the neural network model for iteration, and then update the parameters of the neural network;

[0012] Step 7: Use the trained network model to predict the probability of mineralization of the geochemical exploration data in the predicted area and generate a probability prediction distribution map of mineral resources in the area.

[0013] Furthermore, the geographic location information in step 1 includes the latitude and longitude coordinates of the known mineral target area, and the geochemical element information includes the content of various geochemical elements in the known mineral target area; the geographic location information and geochemical element information are input into ArcGIS software, and the geological image of the known mineral target area is finally obtained through the Kriging interpolation method based on geo-statistics.

[0014] Furthermore, in step 2, ArcGIS software is used to iteratively self-organize clustering of geochemical exploration data, and known ore deposit areas are selected as cluster centers, and the study area is divided into mineralization prospective areas and background areas. The mineralization prospective areas account for A% of the study area, and the known ore deposit areas are included in the mineralization prospective areas; the background area accounts for (100-A)% of the study area, where A is a preset proportion.

[0015] Furthermore, the step 3 specifically includes the following steps:

[0016] Step 31: Divide the prospective mineralization area into several square areas, and expand the geochemical data X of each area from the matrix format into an n-dimensional vector [x1, x2, …, x n ], x a ∈(0,1), a=1,2,…,n, assuming that the various attribute characteristics of geochemical exploration data in the prospecting area of ​​ore formation satisfy the probability distribution z of the hidden space;

[0017] Step 32: Construct an encoder q using a neural network Φ , Φ is a training parameter, and the encoder generates an i-dimensional mean vector M=[μ1,μ2,…,μ i ] and the i-dimensional standard deviation vector N = [σ1,σ2,…,σ i ], these two vectors can represent a mixed Gaussian distribution q Φ (z|X) is used to approximate the probability distribution z of the hidden space; then in the distribution q Φ Random sampling on (z|X) generates implicit feature vectors of geochemical exploration data eps is an i-dimensional random number vector with a mean of 0 and a variance of 1;

[0018] Step 33: Use a neural network to construct a decoder p θ , θ is a training parameter, and the decoder is used to generate data using the implicit feature vector γ And make Be as similar to X as possible;

[0019] Step 34: To make the encoder q Φ The resulting distribution q Φ (z|X) approximates the probability distribution z and improves the decoder p θ Reconstruct the probability that the implicit feature vector γ is the geochemical data X and construct the loss function

[0020]

[0021] Step 35: Use the data of the mineralization prospect to train the encoder q with the goal of minimizing the loss function Φ and decoder p θ ;

[0022] Step 36: After the training is completed, the geochemical exploration data of each area in the mineralization prospect area are checked in turn. If the reconstructed cross entropy of a certain area The value of is higher than the average value, indicating that the probability of data in this area being reconstructed is small, and the spatial data characteristics are also different from those of the surrounding areas, which is regarded as a geochemical anomaly area;

[0023] Step 37: The geochemical anomaly area and the nearest known mineral deposit area are considered to have highly similar spatial data characteristics.

[0024] Furthermore, the step 4 specifically includes the following steps:

[0025] Step 41: Convert the geological image into a digital matrix x0;

[0026] Step 42: Create a digital matrix x1 with the same width and height dimensions, number of channels and data type as the geological image, where each element of the matrix x1 is a random value and obeys a normal distribution;

[0027] Step 43: Let y1 = x0 + x1, and convert the digital matrix y1 into an image data format to obtain a new geological image.

[0028] Furthermore, the step 5 specifically includes the following steps:

[0029] Step 51: construct a convolutional neural network model, the data input format of the model is a matrix of c×h×w, c represents the number of image channels of geochemical data, each channel contains information of one geochemical element, a total of c geochemical information, h and w represent the height and width of the image, and the data output format of the convolutional neural network is a probability vector representing each ore type or background area;

[0030] Step 52: The labels of each type of sample in the data set are i, there are I types of samples, and the total number of each type of training samples is n i The total number of training samples is The loss weight of each class of samples during training The weight corresponding to the category with a large number of samples is smaller, so that the network will not overfit to this category of samples during inference;

[0031] Step 53: Sample x i , i is its corresponding label, and the vector obtained by the neural network output is P = [p1, p2, ..., p I ], take out the probability p corresponding to its label i i , p i ∈(0,1], let This term represents the cross entropy, the cross entropy The smaller it is, the closer the output of the neural network is to the distribution of the corresponding label i. This term represents information entropy. The smaller it is, the lower the distribution uncertainty of the data P is. This term represents the sample x i penalty losses;

[0032] Step 54: The loss function of the neural network is the average penalty loss of m samples, where m is the number of training samples in the same batch. Training the neural network with the goal of minimizing the loss function can effectively prevent the network model from overfitting the spatial data characteristics of the background area and optimize the network model to learn the mineralization characteristics of complex geochemical exploration data.

[0033] Furthermore, after the trained network model is obtained in step 7, the geochemical exploration data of the mining area to be predicted is input into the system, and the mineralization probability of each window area is obtained through the sliding window algorithm, and finally the mineralization probability prediction distribution map of the entire mining area is formed.

[0034] Based on the same inventive concept, the present invention provides a deep learning mineral resource classification prediction system based on sample imbalance, comprising:

[0035] A preprocessing module, used to form geological image data of the target area of ​​the mine through geographical location information and geochemical element information;

[0036] Zoning module, used to divide the study area into ore deposit area, mineralization prospect area and background area;

[0037] The module for identifying geochemical anomaly areas is used to analyze the characteristic distribution of geochemical data in the mineralization prospect area through a variational autoencoder and determine the area with geochemical anomaly in the mineralization prospect area. If there is an area with geochemical anomaly in the mineralization prospect area, the category of the area is marked as the same as the category of the nearest ore deposit area.

[0038] A data enhancement module is used to enhance the data of samples in the ore deposit area and the obtained geochemical anomaly area;

[0039] The classification model module is used to construct a convolutional neural network model to learn the laws between geochemical exploration data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and to optimize the network model to learn the mineralization characteristics of geochemical exploration data, loss weights and penalty losses are introduced to improve the traditional cross entropy loss function;

[0040] The model training module is used to input the training sample data into the neural network model for iteration and then update the parameters of the neural network;

[0041] The classification prediction module is used to use the trained network model to predict the mineralization probability of the geochemical exploration data in the predicted area and generate a probability prediction distribution map of the mineral resources in the area.

[0042] Based on the same inventive concept, the present invention provides a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the steps of the deep learning mineral resource classification prediction method based on sample imbalance are implemented.

[0043] Based on the same inventive concept, the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of a deep learning mineral resource classification prediction method based on sample imbalance.

[0044] Beneficial effects: The present invention provides a deep learning method and system for mineral resource classification prediction based on sample imbalance. In order to find the relationship between geochemical data and mineralization more comprehensively and objectively, the present invention analyzes and obtains the data of geochemical anomalies in the prospective mineralization area. Compared with the traditional geochemical anomaly extraction method, the method proposed by the present invention can identify weak anomalies hidden in complex geological environments and can more effectively obtain outliers in geochemical data; in addition, the problem of small number of samples in the mining area is overcome by adding noise to the geoscience data; by introducing loss weights and penalty loss technology, the network model is effectively prevented from overfitting the spatial data characteristics of the background area and the network model is optimized to learn the mineralization characteristics of complex geochemical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is an overall flow chart of an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the geochemical exploration data anomaly analysis process of the present invention;

[0047] Figure 3 It is a model structure diagram of the convolutional neural network of the present invention;

[0048] Figure 4 It is a schematic diagram of the mineral resource classification prediction method based on convolutional neural network of the present invention. DETAILED DESCRIPTION

[0049] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0050] like Figure 1 As shown, a deep learning mineral resource classification prediction method based on sample imbalance disclosed in an embodiment of the present invention mainly includes the following steps:

[0051] Step 1: Generate geological image data of the target area through geographic location information and geochemical element information.

[0052] The step 1 specifically includes:

[0053] The geographic location information includes the longitude and latitude coordinates of the known mineral target area, and the geochemical element information includes the contents of 10 geochemical elements such as Ag, As, Au, Bi, Cu, Hg, Pb, Sb, Sn and Zn in the known mineral target area. The geoscientific data such as geographic location information and geochemical element information are input into the ArcGIS software, and the geological image of the known mineral target area is finally obtained through the Kriging interpolation method based on geo-statistics.

[0054] Step 2: Divide the study area into ore deposit area, mineralization prospect area and background area.

[0055] The step 2 specifically includes:

[0056] ArcGIS software was used to perform iterative self-organizing clustering on geochemical exploration data, and the known ore deposit area was selected as the cluster center. The study area was divided into a prospective mineralization area and a background area. The prospective mineralization area accounted for 10% of the study area, and the known ore deposit area was included in the prospective mineralization area; the background area accounted for 90% of the study area.

[0057] Step 3: Analyze the characteristic distribution of geochemical data in the mineralization prospecting area through variational autoencoder and determine the area with abnormal geochemical data in the mineralization prospecting area. Since the spatial data characteristics of mineralized areas and non-mineralized areas are quite different, if there are areas with abnormal geochemical data in the mineralization prospecting area, it can indirectly indicate that the area has a higher probability of mineralization and shows a higher similarity with the geochemical data characteristics of known deposits. The category of the area is marked as the same as the category of the nearest deposit area.

[0058] like Figure 2 As shown, step 3 specifically includes the following steps:

[0059] Step 31: Divide the prospective mineralization area into several square areas, which cover all prospective mineralization areas. The geochemical data X of each area is expanded from the matrix format into an n-dimensional vector [x1, x2, …, x n ], x a ∈(0,1), a=1,2,…,n, assuming that the various attribute characteristics of geochemical exploration data in the prospecting area of ​​ore formation satisfy the probability distribution z of the hidden space;

[0060] Step 32: Construct an encoder q using a neural network Φ , Φ is a training parameter, and the encoder generates an i-dimensional mean vector M=[μ1,μ2,…,μ i ] and the i-dimensional standard deviation vector N = [σ1,σ2,…,σ i ], these two vectors can represent a mixed Gaussian distribution q Φ (z|X) is used to approximate the probability distribution z of the hidden space; then in the distribution q Φ Random sampling on (z|X) generates implicit feature vectors of geochemical exploration data eps is an i-dimensional random number vector with a mean of 0 and a variance of 1;

[0061] Step 33: Use a neural network to construct a decoder p θ , θ is a training parameter, and the decoder is used to generate data using the implicit feature vector γ And make Be as similar to X as possible;

[0062] Step 34: To make the encoder q Φ The resulting distribution q Φ (z|X) approximates the probability distribution z and improves the decoder p θ Reconstruct the probability that the implicit feature vector γ is the geochemical data X and construct the loss function

[0063]

[0064] Step 35: Use the data of the mineralization prospect to train the encoder q with the goal of minimizing the loss function Ф and decoder p θ ;

[0065] Step 36: After the training is completed, the geochemical exploration data of each area in the mineralization prospect area are checked in turn. If the reconstructed cross entropy of a certain area The value of is higher than the average value, indicating that the probability of data in this area being reconstructed is small, and the spatial data characteristics are also different from those of the surrounding areas, which is regarded as a geochemical anomaly area;

[0066] Step 37: The geochemical anomaly area and the nearest known mineral deposit area are considered to have highly similar spatial data characteristics.

[0067] In the embodiment of the present invention, the mineralization prospecting area is divided into several square areas of 10×10 pixels, corresponding to a 10×10×10 digital matrix X (the first two 10s are width and height, respectively, and the third 10 is the number of geochemical elements), which is expanded into a 1000-dimensional vector and input into the encoder. The encoder is composed of three linear layers and adopts the Relu activation function. According to the number of geochemical elements, it finally outputs a mean vector M and a standard deviation vector N with two dimensions of 10, which are used to approximate the feature probability distribution z of the hidden space; the decoder is composed of two linear layers and adopts the Relu activation function. Φ The geochemical data X is reconstructed by random uniform sampling on (z|X); the learning rate of the encoder and decoder is kept at 0.001 during the process of learning to extract geochemical anomalies.

[0068] Step 4: Data enhancement is performed on samples of the ore deposit area and the obtained geochemical anomaly area.

[0069] The step 4 specifically includes the following steps:

[0070] Step 41: Convert the geological image into a digital matrix x0;

[0071] Step 42: Create a digital matrix x1 with the same width and height dimensions, number of channels and data type as the geological image. Each element of the matrix x1 is a random value and follows the distribution Where μ = 0.1, σ = 0.1;

[0072] Step 43: Let y1 = x0 + x1, and convert the digital matrix y1 into an image data format to obtain a new geological image.

[0073] After data augmentation, all samples are randomly divided into training set and test set according to a certain ratio (such as 7:3), and all images after data augmentation are classified as training set.

[0074] Step 5: Construct a convolutional neural network model to learn the rules between geochemical data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and to optimize the network model to learn the mineralization characteristics of geochemical data, the loss weight and penalty loss are introduced to improve the traditional cross entropy loss function.

[0075] The step 5 specifically includes the following steps:

[0076] Step 51: construct a convolutional neural network model, the data input format of the model is a matrix of c×h×w, c represents the number of image channels of geochemical data, each channel contains information of one geochemical element, a total of c geochemical information, h and w represent the height and width of the image, and the data output format of the convolutional neural network is a probability vector representing each ore type or background area;

[0077] Step 52: The labels of each type of sample in the data set are i, there are I types of samples, and the total number of each type of training samples is n i The total number of training samples is The loss weight of each class of samples during training The weight corresponding to the category with a large number of samples is smaller, so that the network will not overfit to this category of samples during inference;

[0078] Step 53: Sample x i , i is its corresponding label, and the vector obtained by the neural network output is P = [p1, p2, ..., p I ], take out the probability p corresponding to its label i i , p i ∈(0,1], let This term represents the cross entropy, the cross entropy The smaller it is, the closer the output of the neural network is to the distribution of the corresponding label i. This term represents information entropy. The smaller it is, the lower the distribution uncertainty of the data P is. This term represents the sample x i penalty losses;

[0079] Step 54: The loss function of the neural network is the average penalty loss of m samples, where m is the number of training samples in the same batch. Training the neural network with the goal of minimizing the loss function can effectively prevent the network model from overfitting the spatial data characteristics of the background area and optimize the network model to learn the mineralization characteristics of complex geochemical exploration data.

[0080] Step 6: Input the training sample data into the neural network model for iteration, and then update the parameters of the neural network.

[0081] The step 6 specifically includes:

[0082] The training sample data includes the training data after sample enhancement. After each parameter update, the test sample data is tested to obtain the accuracy, and the set of model parameters with the highest accuracy is saved.

[0083] Step 7: Use the trained network model to predict the probability of mineralization of the geochemical exploration data in the predicted area and generate a probability prediction distribution map of mineral resources in the area.

[0084] The step 7 specifically includes:

[0085] After obtaining the trained network model, the geochemical exploration data of the mining area to be predicted is input into the system. Through the sliding window algorithm, the mineralization probability of each window area is obtained, and finally the mineralization probability prediction distribution map of the entire mining area is formed.

[0086] In order to prevent the characteristics of mining samples from gradually disappearing as the number of convolutional neural network convolutional layers increases, which results in the inability to optimize parameters and ultimately affects the performance of the classifier, the embodiment of the present invention introduces the structure of the residual network and combines convolution kernels of different sizes to extract the characteristics of the mining area from different angles, making full and effective use of precious sample resources. Figure 3 As shown, the model of the convolutional neural network in the embodiment of the present invention is composed of 29 convolutional layers. The size of the convolution kernel and the feature dimension have been indicated in the figure. In order to prevent the features of the samples from gradually disappearing as the number of convolutional layers increases, 14 residual blocks are introduced; the number of iterations of the model is determined to be 100 times, the amount of training batch data is 16, and the learning rate is linearly decayed from the initial 0.01 to 0.0001 through iterations.

[0087] After the neural network model is trained, the geochemical data of the area to be predicted is input into the system. The system traverses the geochemical data of the entire area through a sliding window algorithm. In each window, the model will output the mineralization probability. Finally, the system will predict and output the mineral probability distribution map of the area. Figure 4 .

[0088] Based on the same inventive concept, an embodiment of the present invention discloses a deep learning mineral resource classification prediction system based on sample imbalance, including: a preprocessing module, used to form geological image data of a mineral target area through geographic location information and geochemical element information; a partitioning module, used to divide the study area into a mineral deposit area, a mineralization prospect area and a background area; a geochemical anomaly area identification module, used to analyze the characteristic distribution of geochemical data in the mineralization prospect area through a variational autoencoder and determine the area with abnormal geochemical data in the mineralization prospect area. If there is an area with abnormal geochemical data in the mineralization prospect area, the category of the area is marked as the same as the category of the mineralization deposit area closest to it; a data enhancement module, used to identify the mineral deposit area. The samples of the bed area and the obtained geochemical anomaly area are used for data enhancement; the classification model module is used to construct a convolutional neural network model to learn the rules between geochemical data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and optimize the network model to learn the mineralization characteristics of the geochemical data, the loss weight and penalty loss are introduced to improve the traditional cross entropy loss function; the model training module is used to input the training sample data into the neural network model for iteration, and then update the parameters of the neural network; the classification prediction module is used to use the trained network model to predict the probability of mineralization of the geochemical data in the predicted area, and generate a probability prediction distribution map of mineral resources in the area.

[0089] Based on the same inventive concept, an embodiment of the present invention discloses a computer system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the steps of the deep learning method for mineral resource classification prediction based on sample imbalance are implemented.

[0090] Based on the same inventive concept, an embodiment of the present invention discloses a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of a deep learning mineral resource classification prediction method based on sample imbalance.

Claims

1. A deep learning mineral resource classification prediction method based on sample imbalance, characterized in that: The steps include: Step 1: Generate geological image data of the target area through geographic location information and geochemical element information; Step 2: Divide the study area into ore deposit area, mineralization prospect area and background area; Step 3: Analyze the characteristic distribution of geochemical data in the mineralization prospecting area through variational autoencoder and determine the area with abnormal geochemical data in the mineralization prospecting area. If there is an area with abnormal geochemical data in the mineralization prospecting area, mark the category of the area as the same as the category of the nearest mineral deposit area; Step 4: Data enhancement of samples of the ore deposit area and the obtained geochemical anomaly area; Step 5: Construct a convolutional neural network model to learn the law between geochemical data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and to optimize the network model to learn the mineralization characteristics of geochemical data, the traditional cross entropy loss function is improved by introducing loss weight and penalty loss; Step 6: Input the training sample data into the neural network model for iteration, and then update the parameters of the neural network; Step 7: Use the trained network model to predict the probability of mineralization for the geochemical exploration data of the predicted area, and generate a probability prediction distribution map of mineral resources in the area; The step 3 comprises the following steps: Step 31: Divide the prospective mineralization area into several square areas, and expand the geochemical data X of each area from the matrix format into an n-dimensional vector [x1, x2, …, x n ], x a ∈(0,1), a=1,2,…,n, assuming that the various attribute characteristics of geochemical exploration data in the prospecting area of ​​ore formation satisfy the probability distribution z of the hidden space; Step 32: Construct an encoder q using a neural network Φ , Φ is a training parameter, and the encoder generates an i-dimensional mean vector M=[μ1,μ2,…,μ i ] and the i-dimensional standard deviation vector N = [σ1,σ2,…,σ i ], these two vectors can represent a mixed Gaussian distribution q Φ (z|X) is used to approximate the probability distribution z of the hidden space; then in the distribution q Φ Random sampling on (z|X) generates implicit feature vectors of geochemical exploration data eps is an i-dimensional random number vector with a mean of 0 and a variance of 1; Step 33: Use a neural network to construct a decoder p θ , θ is a training parameter, and the decoder is used to generate data using the implicit feature vector γ And make Be as similar to X as possible; Step 34: To make the encoder q Φ The resulting distribution q Φ (z|X) approximates the probability distribution z and improves the decoder p θ Reconstruct the probability that the implicit feature vector γ is the geochemical data X and construct the loss function Step 35: Use the data of the mineralization prospect to train the encoder q with the goal of minimizing the loss function Φ and decoder p θ ; Step 36: After the training is completed, the geochemical exploration data of each area in the mineralization prospect area are checked in turn. If the reconstructed cross entropy of a certain area The value is higher than the average value, which is regarded as an abnormal area of ​​geochemical exploration; Step 37: The geochemical anomaly area and the nearest known mineral deposit area are considered to have highly similar spatial data characteristics; The step 4 comprises the following steps: Step 41: Convert the geological image into a digital matrix x0; Step 42: Create a digital matrix x1 with the same width and height dimensions, number of channels and data type as the geological image, where each element of the matrix x1 is a random value and obeys a normal distribution; Step 43: Let y1 = x0 + x1, and convert the digital matrix y1 into an image data format to obtain a new geological image; The step 5 comprises the following steps: Step 51: construct a convolutional neural network model, the data input format of the model is a matrix of c×h×w, c represents the number of image channels of geochemical data, each channel contains information of one geochemical element, a total of c geochemical information, h and w represent the height and width of the image, and the data output format of the convolutional neural network is a probability vector representing each ore type or background area; Step 52: The labels of each type of sample in the data set are i, there are I types of samples, and the total number of each type of training samples is n i The total number of training samples is The loss weight of each class of samples during training Step 53: Sample x i , i is its corresponding label, and the vector obtained by the neural network output is P = [p1, p2, ..., p I ], take out the probability p corresponding to its label i i , p i ∈(0,1], let Cross Entropy The smaller it is, the closer the output of the neural network is to the distribution of the corresponding label i. Information Entropy The smaller it is, the lower the distribution uncertainty of the data P is. Represents sample x i penalty losses; Step 54: The loss function of the neural network is the average penalty loss of m samples, where m is the number of training samples in the same batch. Training the neural network with the goal of minimizing the loss function can effectively prevent the network model from overfitting the spatial data characteristics of the background area and optimize the network model to learn the mineralization characteristics of complex geochemical exploration data.

2. According to claim 1, a deep learning mineral resource classification prediction method based on sample imbalance is characterized in that: The geographic location information in step 1 includes the latitude and longitude coordinates of the known mineral target area, and the geochemical element information includes the content of various geochemical elements in the known mineral target area; the geographic location information and geochemical element information are input into ArcGIS software, and the geological image of the known mineral target area is finally obtained through the Kriging interpolation method based on geo-statistics.

3. The method for classification and prediction of mineral resources based on deep learning and sample imbalance according to claim 1 is characterized in that: In step 2, ArcGIS software is used to iteratively self-organize clustering of geochemical exploration data, and known ore deposit areas are selected as cluster centers. The study area is divided into mineralization prospective areas and background areas. The mineralization prospective areas account for A% of the study area, and the known ore deposit areas are included in the mineralization prospective areas; the background area accounts for (100-A)% of the study area, where A is a preset proportion.

4. The method for classification and prediction of mineral resources based on deep learning and sample imbalance according to claim 1 is characterized in that: After the trained network model is obtained in step 7, the geochemical exploration data of the mining area to be predicted is input into the system, and the mineralization probability of each window area is obtained through the sliding window algorithm, and finally the mineralization probability prediction distribution map of the entire mining area is formed.

5. A deep learning mineral resource classification prediction system based on sample imbalance, characterized in that: The method for implementing the deep learning mineral resource classification prediction method based on sample imbalance according to any one of claims 1 to 4 comprises: A preprocessing module, used to form geological image data of the target area of ​​the mine through geographical location information and geochemical element information; Zoning module, used to divide the study area into ore deposit area, mineralization prospect area and background area; The module for identifying geochemical anomaly areas is used to analyze the characteristic distribution of geochemical data in the mineralization prospect area through a variational autoencoder and determine the area with geochemical anomaly in the mineralization prospect area. If there is an area with geochemical anomaly in the mineralization prospect area, the category of the area is marked as the same as the category of the nearest ore deposit area. A data enhancement module is used to enhance the data of samples in the ore deposit area and the obtained geochemical anomaly area; The classification model module is used to construct a convolutional neural network model to learn the laws between geochemical exploration data and mineralization. In order to prevent the neural network model from overfitting the spatial data characteristics of the background area and to optimize the network model to learn the mineralization characteristics of geochemical exploration data, loss weights and penalty losses are introduced to improve the traditional cross entropy loss function; The model training module is used to input the training sample data into the neural network model for iteration and then update the parameters of the neural network; The classification prediction module is used to use the trained network model to predict the probability of mineralization of the geochemical exploration data in the predicted area and generate a probability prediction distribution map of the mineral resources in the area.

6. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into the processor, the steps of the deep learning mineral resource classification prediction method based on sample imbalance according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the deep learning mineral resource classification prediction method based on sample imbalance according to any one of claims 1 to 4 are implemented.

Images