CN115511214A - Multi-scale sample unevenness-based mineral product prediction method and system - Google Patents
Multi-scale sample unevenness-based mineral product prediction method and system Download PDFInfo
- Publication number
- CN115511214A CN115511214A CN202211308574.6A CN202211308574A CN115511214A CN 115511214 A CN115511214 A CN 115511214A CN 202211308574 A CN202211308574 A CN 202211308574A CN 115511214 A CN115511214 A CN 115511214A
- Authority
- CN
- China
- Prior art keywords
- data
- sample
- image
- geochemical
- encoder
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 229910052500 inorganic mineral Inorganic materials 0.000 title claims abstract description 33
- 239000011707 mineral Substances 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 29
- 230000002159 abnormal effect Effects 0.000 claims abstract description 23
- 239000000126 substance Substances 0.000 claims abstract description 15
- 230000007246 mechanism Effects 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 230000033558 biomineral tissue development Effects 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 20
- 238000005065 mining Methods 0.000 claims description 20
- 238000001514 detection method Methods 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 230000005856 abnormality Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000013145 classification model Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims 1
- 239000000523 sample Substances 0.000 description 48
- 230000008569 process Effects 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Agronomy & Crop Science (AREA)
- Molecular Biology (AREA)
- Primary Health Care (AREA)
- Marine Sciences & Fisheries (AREA)
- Animal Husbandry (AREA)
- Remote Sensing (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Mining & Mineral Resources (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-scale sample unevenness-based mineral prediction method and a multi-scale sample unevenness-based mineral prediction system, wherein the method comprises the following steps: converting the collected geological data into image data to obtain geochemical maps with different scales; dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; sample data enhancement is carried out; establishing a self-encoder and generation confrontation network joint model, and finding a chemical exploration abnormal area by using a small-scale geochemical map; and constructing a convolutional neural network model with an attention mechanism, and classifying the mineral products in the abnormal area of the large-scale data. The combined model can learn the internal relation and characteristics among the multi-element geochemical data, and can effectively avoid the influence of noise in the geochemical data by carrying out the extraction of the chemical exploration abnormity on the small-scale data. And an attention mechanism is utilized on large-scale data to enable the model to pay more attention to key information in the image, so that the image classification accuracy can be effectively improved.
Description
Technical Field
The invention relates to a method and a system for predicting mineral products based on multi-scale sample unevenness, and belongs to the field of earth information science and the field of computer vision.
Background
With the rapid development of the economy of China, mineral resources become more important material bases for the development of the economy and the society, and how to better develop and utilize the mineral resources becomes an important topic in the field of geology. With the development of artificial intelligence technology becoming more sophisticated, computer technology and artificial intelligence technology have gradually penetrated the field of geoscience as compared with the prior art that only expert knowledge and related geological software are used for mining prediction. By using machine learning and deep learning techniques, the mineralization prediction is also combined with artificial intelligence from the traditional method to the traditional method at present, so that the result is more accurate and reliable.
In recent years, the artificial intelligence technology is applied to mineral resource prediction work to gradually obtain certain effect, but due to the scarcity of samples and the characteristic that a mineral forming area is small relative to a background area, sample data of different scales is not utilized in the existing processing method, and the mineral forming area cannot be predicted more accurately.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the defects of the prior art, the invention aims to provide a method and a system for predicting mineral products based on multi-scale sample heterogeneity, so as to more comprehensively and objectively find out the relation between chemical exploration data and mineralization and find out the mineralization rule in geochemical data so as to predict the mineralization probability in new geochemical data.
The technical scheme is as follows: in order to achieve the above object, the method for predicting mineral products based on multi-scale sample inequality according to the present invention comprises the following steps:
step 1: converting the collected geological data into image data to obtain geochemistry graphs of different scales, namely 1: p 1 Geochemical map ofAnd 1: p is 2 Wherein 1: p is 1 And 1: p 2 Represents the ratio of the distance on the graph to the actual distance, and P 1 Greater than P 2 ；；
Step 2: dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; dividing each prediction graph into grids, if the grids contain ore beds, selecting the grids as a mining area, and otherwise, selecting the grids as a background area;
and step 3: performing data enhancement on the obtained training sample;
and 4, step 4: introducing a generation countermeasure network into the deep convolution self-encoder, wherein the generator in the generation countermeasure network is used as a decoder in the self-encoder, and the sample size is 1: p 1 Mapping the grid data to a potential space from an image space, training to generate a confrontation network (WGAN) to obtain output, and calculating a Euclidean distance between input data and output data to serve as a geophysical anomaly score;
and 5: based on the following steps of 1: p 1 The geochemical anomaly score map of (1): p 2 Carrying out sample marking in the grid data, marking the grid data with the detection abnormality score higher than a set threshold value as a positive sample, and otherwise, carrying out concrete detection data classification for the negative sample on the basis; constructing a convolutional neural network model, and adding an attention mechanism into the model to enable the model to focus information of an interested region so as to learn a rule between mineralization data and geophysical data;
and 6: mixing 1: p 2 Inputting training sample data into a convolutional neural network model for iteration, and updating parameters of the neural network;
and 7: and carrying out mineralization probability prediction on the chemolithology data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region.
Further, the geological data collected in step 1 includes geographical location information and geochemical element information, the geographical location information includes longitude and latitude coordinates of a known mine area, and the geochemical element information includes contents of various geochemical elements of the known mine area; geographic position information and geochemical element information are input into ArcGIS software, and a chemometric factor is converted into geochemical spatial data through an inverse distance weighted interpolation method to obtain geochemical maps with different scales.
Further, in the step 3, data enhancement is performed by adding noise and generating new samples by using a generation countermeasure network.
Further, in the step 4
Assume that 1: p 1 The generator model G and the discriminator model D in the WGAN are generated by learning a potential representation of the background region data using the WAGN network while satisfying a probability distribution of a certain hidden space. The method specifically comprises the following steps:
step 41: constructing an encoder E by using a convolutional neural network, and enabling the data-enhanced 1: p 1 Inputting the background region data samples into an encoder, performing feature extraction through the encoder, and mapping to a potential space z;
step 42: using a generative confrontation network after the encoder network, wherein a generator G in the generative confrontation network, as a decoder in the self-encoder, performs mapping from the potential space to the image space; in order to be able to obtain the potential spatial feature distribution of the location of the background region, the loss function of the input image x and the reconstructed image G (E (x)) is minimized:
where n is the number of pixels in the image;
step 43: in training the encoder, statistical information of the computed true images and the generated images into the discriminator is used to guide the training of the encoder E. The loss function for discriminator D to guide encoder training is as follows:
wherein n is d Is the dimension of the intermediate feature representation, k is the weighting factor, f (-) is the feature generated by discriminator D during the operation, in order to count the information of a given input;
step 44: using background region data to minimize loss function L 2 Training the encoder E for the purpose to generate an image with a more similar characteristic distribution to the original image;
step 45: after training is finished, calculating Euclidean distance between an original input image x x and a trained output image y generated by the countermeasure network to serve as a chemodetective abnormal score, and setting an abnormal value threshold, namely, a score higher than the threshold serves as an abnormal part, and a score lower than the threshold serves as a non-abnormal part; the Euclidean distance formula is as follows:
further, the step 5 comprises the following steps:
step 51: obtaining a current 1 according to the abnormal score: p is 1 Obtaining a region with high mineralization probability at the gathering position of the abnormal part of the geochemistry map; and obtaining 1 of the corresponding position according to the existing data: p 2 The probe data;
step 52: constructing a convolutional neural network model, wherein the input data of the model is a c multiplied by h multiplied by w matrix, c represents the number of image channels of geochemical data, each channel contains information of one type of chemodetection element, c pieces of chemodetection information are totally contained, h and w represent the height and width of an image, and the data output format of the convolutional neural network is a probability vector representing each ore type or background area;
step 53: in the construction of a convolution model, a channel attention module is fused, a weight is applied to a feature map on each channel, namely attention is focused on mineral areas to be identified through the following formula, feature extraction processing is carried out on the areas, and correlation in channel data in geophysical prospecting data is captured;
g(x)＝Sigmoid(W 2 ReLU(W 1 x))
where Sigmoid and ReLU are activation functions,W 1 and W 2 Is a weight parameter of the fully connected layer. Through the calculation of the channel attention, an attention weight matrix g (x) can be obtained, and the obtained weight matrix g (x) is multiplied by the original feature map x to obtain feature information required by the operation of the next-layer network, namely the input of the next-layer network.
Further, the loss function of the training network model in step 6 is:
wherein L is CE (p i ,y i ) Cross entropy, y, representing multiple classes i Is a true tag, p i Is the predicted label classification probability, N is the total number of samples; GD (g) i ) Is the gradient density;
g i denotes the gradient, gm i Representing the gradient mode length, epsilon is a hyperparameter.
The invention provides a mineral product prediction system based on multi-scale sample unevenness, which comprises:
a preprocessing module for converting the collected geological data into image data, to obtain 1: p 1 And 1: p is 2 The geochemistry of (a), wherein 1: p 1 And 1: p 2 Represents the ratio of the distance on the graph to the actual distance, and P 1 Greater than P 2 (ii) a (ii) a Dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; dividing each prediction graph into grids, if the grids contain ore beds, selecting the grids as mining areas, and if not, selecting the grids as background areas;
an anomaly identification module for identifying an anomalyGenerating a countermeasure network is introduced into the deep convolutional auto-encoder, wherein the generator in the countermeasure network is generated as a decoder in the auto-encoder, and the sample size is 1: p 1 The grid data are mapped to a potential space from an image space, then a confrontation network is generated through training to obtain output, and the Euclidean distance between input data and output data is calculated to serve as a geophysical anomaly score;
the data enhancement module is used for enhancing data of the mining area and the obtained chemical exploration abnormal area;
a classification model module to classify the model based on a probability distribution between 1: p 1 The geochemical map of (1): p 2 Carrying out sample marking on the grid data, marking the grid data with the detection abnormality score higher than a set threshold value as a positive sample, otherwise, carrying out concrete detection data classification on the basis of the positive sample and the negative sample; constructing a convolutional neural network model, and adding an attention mechanism into the model to enable the model to focus information of the region of interest so as to learn a rule between mineralization data and geophysical prospecting data;
a model training module to compare 1: p 2 Inputting training sample data into a convolutional neural network model for iteration, and updating parameters of the neural network;
and the classification prediction module is used for carrying out mineralization probability prediction on the chemolithologic data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region.
The invention also provides a computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the computer program when loaded into the processor implements the method for multi-scale sample heterogeneity-based mineral prediction.
Has the advantages that: the invention provides a method and a system for predicting mineral products based on multi-scale sample nonuniformity. In order to more accurately and objectively find out the relation between the chemical exploration data and the mineralization probability, the grid data under a smaller scale is firstly analyzed, and the chemical exploration abnormity is extracted. Then, after the abnormal region is found in the small-scale buffer map, the large-scale data of the abnormal region is used for carrying out specific mining data classification. The method utilizes the two data with different scales to carry out mineralization prediction, and uses a loss function based on gradient aiming at the problem of non-uniformity of mineral product data samples, so that the final result is more accurate and reliable.
Drawings
FIG. 1 is an overall flow chart of an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating an anomaly analysis process for chemical exploration data according to an embodiment of the present disclosure;
FIG. 3 is a CA _ Block diagram of a channel attention network structure according to an embodiment of the present invention;
fig. 4 is a structure diagram of a ResNet34CA _ Block network incorporating a channel attention mechanism in an embodiment of the present invention.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.
As shown in fig. 1, the method for predicting mineral products based on multi-scale sample unevenness disclosed in the embodiments of the present invention mainly includes the following steps:
step 1: and converting the collected geological data into image data to obtain two different scales of geochemical maps.
Collecting relevant geological exploration data, wherein the geological exploration data comprises geographical position information and geochemical element information, the geographical position information comprises longitude and latitude coordinates of a known mining area, and the geochemical element information comprises the content of various geochemical elements of the known mining area; geographic position information and geochemical element information are input into ArcGIS software, and a chemometric factor is converted into geochemical spatial data through an inverse distance weighted interpolation method, so that geochemical maps with different scales are obtained finally. Two different rulersThe geochemical map of degrees is 1: p 1 And 1: p 2 The geochemistry of (a), wherein 1: p1 and 1: p2 represents the ratio of the distance on the graph to the actual distance, and P1 is greater than P2. In this embodiment, 1:200000 geochemistry and 1:50000 geochemistry.
Step 2: dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; and dividing each prediction graph into grids, and selecting the grids as mining areas if the grids contain ore beds, or selecting the grids as background areas if the grids do not contain ore beds. In the embodiment, the grid is divided into grids according to the equal proportion, and the data of the divided grids are divided into the mining area and the background area according to expert knowledge.
And step 3: and performing data enhancement on the obtained training sample. In the embodiment, data enhancement is performed by adding noise and generating new samples by using a generation countermeasure network. Adding noise to original data, setting random noise to be 0.03,0.04,0.05, and setting the number of times of adding noise to each sample to be 20; and then the number of samples is increased by using the picture reconstruction function of the depth convolution self-encoder.
And 4, step 4: for 1:200000 mesh data for anomaly identification. Introducing a generation countermeasure network into a deep convolution self-encoder, and enabling the sample size to be 1:200000 grid data are mapped to potential space from image space, then training is carried out to generate a confrontation network to obtain output, and Euclidean distance between input data and output data is calculated to serve as a detection anomaly score.
In the present embodiment, assume that 1:200000 background region data satisfies a probability distribution of a hidden space, and potential representation of the background region data is learned by the WAGN network, thereby generating a generator model G and a discriminator model D in WGAN. As shown in fig. 2, the method specifically includes the following steps:
step 41: constructing an encoder E by using a convolutional neural network, inputting the data-enhanced sample into the encoder, and performing feature extraction by using the encoder to obtain the feature distribution of the background region data, namely a potential space z;
step 42: using a generative confrontation network after the network of encoders, wherein the generator G in the generative confrontation network can be regarded as a decoder in the self-encoder, the last step maps the samples from the real image to the latent space z, the decoder performs the mapping from the latent space to the image space; in order to obtain the potential spatial feature distribution of the background region, the loss function of the input image x and the reconstructed image G (E (x)) needs to be minimized:
where n is the number of pixels in the image;
step 43: using only the difference of the minimized pixel size may result in the resulting image not being fully compliant with the distribution of the original image, since the true target position of the hidden space of the given original image data is unknown, i.e. the position in each sample that ultimately affects the mineralization probability is uncertain. This may result in the discriminator D not giving the abnormal image a high residual value. So in training the encoder, the statistical information of the computed true images and the generated images into the discriminator is used to guide the training of the encoder E. The loss function for discriminator D to guide encoder training is as follows:
wherein n is d Is the dimension of the intermediate feature representation, k is the weighting factor, f (-) is the feature generated by discriminator D during the operation, in order to count the information of a given input;
step 44: using background region data to minimize loss function L 2 The encoder E is trained for this purpose to generate an image that more closely resembles the original image feature distribution.
Step 45: after training is finished, calculating the Euclidean distance between an original input image x and a trained output image y generated by the countermeasure network as an abnormal score, setting an abnormal value threshold, namely, a score higher than the threshold is used as an abnormal part, and a score lower than the threshold is used as a non-abnormal part; the Euclidean distance formula is as follows:
and 5: aiming at the problems that in 1: anomaly region at 200000, 1: the mineralization prediction is based on the data on 50000 specifically, 1:200000 geochemical mapping to obtain mapping between 1:50000, carrying out sample marking in the grid data, marking the mark with the chemical detection abnormal score higher than a set threshold value as a positive sample (the mineralization probability is higher), otherwise, marking as a negative sample (the mineralization probability is lower), and carrying out concrete chemical detection data classification on the basis; a convolutional neural network model is constructed, and an attention mechanism is added into the model, and the mechanism can enable the model to focus on information of an interested area, so that the rule between mineralization and geophysical data is learned. The method specifically comprises the following steps:
step 51: obtaining a current 1 according to the abnormal score: 200000 geochemical map abnormal part gathering place, area with high mineralization probability can be obtained; according to the existing data, 1: 50000's sounding data;
step 52: constructing a convolutional neural network model, wherein input data of the model is a c multiplied by h multiplied by w matrix, c represents the number of image channels of geochemical data, each channel contains information of one type of chemical exploration element and c pieces of chemical exploration information in total, h and w represent the height and width of an image, and the data output format of the convolutional neural network is probability vectors representing all ore types or background areas;
step 53: in the process of constructing the convolution model, the fusion channel attention module is used for fusing the information contained in different channels because each channel of the geophysical prospecting data represents one element, and the final mineralization probability corresponding to the different channels is also greatly influenced. The channel attention network structure is shown in fig. 3, and a weight is applied to the feature map of each channel to represent the similarity between the channel and the key information. The method comprises the following steps of focusing attention on mineral areas to be identified through the following formula, rapidly carrying out feature extraction processing on the areas and capturing correlation among channel data in the geophysical prospecting data;
g(x)＝Sigmoid(W 2 ReLU(W 1 x))
wherein Sigmoid and ReLU are activation functions, W 1 And W 2 Is a weight parameter of the fully connected layer. Through the above channel attention calculation, an attention weight matrix g (x) including the weight of each channel during the operation of the network can be obtained, and then the obtained weight matrix g (x) is multiplied by the original feature map x to obtain the feature information required by the operation of the next network, i.e., the input of the next network, as shown in fig. 4.
Step 6: mixing the following components in parts by weight: 50000 the training sample data is input into the convolutional neural network model for iteration, and the parameters of the neural network are updated.
The number of samples in the mining area is far less than that of samples in the background area, namely the number of samples has the characteristic of unbalance; assume that the sample is x i The output P = [ P ] is obtained after being processed through a neural network 1 ,p 2 ,…,p N ]The classification probability p corresponding to the sample i is taken out i Let us order
This term represents the cross entropy of multiple classifications. The smaller the cross entropy is, the more accurate the neural network classification is; in order to solve the problem of unbalanced sample difficult to classify, a gradient module length gm is introduced i And gradient Density GD (g) i ) Wherein the gradient mode length gm i ＝|p i -y i L, wherein y i Is a true tag, p i Is the classification probability of the predicted ith sample label;
wherein g is i The gradient of the ith sample is represented,
ε is a hyperparameter, which yields the following loss function:
n is the total number of samples. In the training process, the weights of simple negative samples (namely most background area pictures in the mineral data) and very difficult samples (namely excessively abnormal mining area samples) in the samples are reduced, namely loss is reduced, and the influence degree on the model is greatly reduced. The weight of normal difficult samples (namely common mineralization area samples) is improved, so that the model can be more focused on effective normal difficult samples, namely common mineralization samples;
the neural network is trained by taking the minimum loss function as a target, so that overfitting of the network model to the spatial data characteristics of the background area can be effectively prevented, and the mineralization characteristics of the network model learning complex exploration data are optimized.
And 7: and carrying out mineralization probability prediction on the chemolithology data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region. Specifically, after a trained network model is obtained, mining area chemical exploration data to be predicted are input into the system, the mining probability of each window area is obtained through a sliding window algorithm, and finally a mining probability prediction distribution graph of the whole mining area is formed.
Based on the same inventive concept, the embodiment of the invention discloses a multi-scale sample unevenness-based mineral prediction system, which comprises a preprocessing module, a data acquisition module, a data processing module and a data processing module, wherein the preprocessing module is used for converting collected geological data into image data to obtain a data set 1: p 1 And 1: p 2 Dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; dividing each prediction graph into grids, if the grids contain ore beds, selecting the grids as mining areas, and otherwise, selecting the grids as background areas;
an anomaly identification module for introducing a generative countermeasure network to depthIn a convolutional auto-encoder in which a generator in a countermeasure network is generated as a decoder in the auto-encoder, a sample size of 1: p 1 The grid data are mapped to a potential space from an image space, then a confrontation network is generated through training to obtain output, and the Euclidean distance between input data and output data is calculated to serve as a geophysical anomaly score;
the data enhancement module is used for enhancing data of the mining area and the obtained chemical exploration abnormal area;
a classification model module to classify the model based on a probability distribution between 1: p 1 The geochemical map of (1): p 2 Carrying out sample marking on the grid data, marking the grid data with the detection abnormality score higher than a set threshold value as a positive sample, otherwise, carrying out concrete detection data classification on the basis of the positive sample and the negative sample; constructing a convolutional neural network model, and adding an attention mechanism into the model to enable the model to focus information of an interested region so as to learn a rule between mineralization data and geophysical prospecting data;
a model training module to compare 1: p 2 Inputting training sample data into a convolutional neural network model for iteration, and updating parameters of the neural network;
and the classification prediction module is used for carrying out mineralization probability prediction on the chemolithologic data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region.
The specific working process of each module described above may refer to the corresponding process in the foregoing method embodiment, and is not described herein again. The division of the modules is only one logical functional division, and other division modes can be realized in practice, for example, a plurality of modules can be combined or can be integrated into another system.
Based on the same inventive concept, the embodiment of the present invention discloses a computer system, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is loaded into the processor, the computer system implements the method for predicting mineral products based on multi-scale sample inequality.
It will be understood by those skilled in the art that the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer system (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes: various media capable of storing computer programs, such as a U disk, a removable hard disk, a read only memory ROM, a random access memory RAM, a magnetic disk, or an optical disk.
Claims (10)
1. A mineral product prediction method based on multi-scale sample nonuniformity is characterized by comprising the following steps:
step 1: converting the collected geological data into image data, resulting in 1: p 1 And 1: p 2 The geochemistry of (a), wherein 1: p 1 And 1: p 2 Represents the ratio of the distance on the graph to the actual distance, and P 1 Greater than P 2 ；
And 2, step: dividing a geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; dividing each prediction graph into grids, if the grids contain ore beds, selecting the grids as a mining area, and otherwise, selecting the grids as a background area;
and step 3: performing data enhancement on the obtained training sample;
and 4, step 4: introducing a generation countermeasure network into the deep convolution self-encoder, wherein the generator in the generation countermeasure network is used as a decoder in the self-encoder, and the sample size is 1: p is 1 The grid data are mapped to a potential space from an image space, then a confrontation network is generated through training to obtain output, and the Euclidean distance between input data and output data is calculated to serve as a geophysical anomaly score;
and 5: based on the following steps of 1: p is 1 The geochemical map of (1): p 2 The grid data of (2) is marked with a sample, the mark with the abnormal detection score higher than a set threshold value is a positive sample, and whether the mark is a positive sample or notClassifying the sample as a negative sample on the basis of the negative sample; constructing a convolutional neural network model, and adding an attention mechanism into the model to enable the model to focus information of an interested region so as to learn a rule between mineralization data and geophysical prospecting data;
and 6: mixing 1: p 2 Inputting training sample data into a convolutional neural network model for iteration, and updating parameters of the neural network;
and 7: and carrying out mineralization probability prediction on the chemolithological data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region.
2. The method according to claim 1, wherein the geological data collected in step 1 comprises geographical location information and geochemical element information, the geographical location information comprises longitude and latitude coordinates of a known mine area, and the geochemical element information comprises contents of various geochemical elements of the known mine area; geographic position information and geochemical element information are input into ArcGIS software, and a chemometric factor is converted into geochemical spatial data through an inverse distance weighted interpolation method to obtain geochemical maps with different scales.
3. The method of claim 1, wherein the step 3 is data enhancement by adding noise and using a generation countermeasure network to generate new samples.
4. The method of claim 1, wherein the step 4 comprises the steps of:
step 41: constructing an encoder E by using a convolutional neural network, and performing data enhancement on the data of 1: p 1 Inputting the background region data samples into an encoder, performing feature extraction through the encoder, and mapping to a potential space;
step 42: using a generative confrontation network after the encoder network, wherein a generator G in the generative confrontation network, as a decoder in the self-encoder, performs mapping from the potential space to the image space; in order to be able to obtain the potential spatial feature distribution of the location of the background region, the loss function of the input image x and the reconstructed image G (E (x)) is minimized:
where n is the number of pixels in the image;
step 43: in training the encoder, the statistical information of the generated confrontation network discriminator for calculating the real image and the generated image is used for guiding the training of the encoder E, and the loss function of the discriminator D for guiding the training of the encoder is as follows:
wherein n is d Is the dimension of the intermediate feature representation, k is the weighting factor, | · | | non-calculation 2 Representing the square of the pixel difference, f (-) is a feature generated by discriminator D during the operation, with the aim of counting the information given the input;
step 44: using background region data to minimize loss function L 2 Training the encoder E for the purpose to generate an image with a more similar characteristic distribution to the original image;
step 45: after the training is finished, the Euclidean distance between the input image x and the output image y generated by the trained generated confrontation network is calculated to be used as the detection anomaly score.
5. The method of claim 1, wherein the step 5 comprises the steps of:
step 51: obtaining 1 according to the abnormality score: p 1 And (3) obtaining 1: p 2 To change intoDetecting data;
step 52: constructing a convolutional neural network model, wherein the input data of the model is a c multiplied by h multiplied by w matrix, c represents the number of image channels of geochemical data, each channel contains information of a type of chemical detection element, h and w represent the height and width of an image, and the data output format of the convolutional neural network is a probability vector representing each ore type or background area;
step 53: in the construction of a convolutional neural network model, an attention mechanism is fused, a weight is applied to a feature map on each channel, namely attention is focused on mineral regions to be identified through the following formula, feature extraction processing is carried out on the regions, and the correlation among channel data in geophysical prospecting data is captured;
g(x)＝Sigmoid(W 2 ReLU(W 1 x))
wherein Sigmoid and ReLU are activation functions, W 1 And W 2 Is the weight parameter of the full connection layer; and obtaining an attention weight matrix g (x) through the calculation of the attention of the channel, and multiplying the obtained weight matrix g (x) by the original characteristic diagram x to obtain characteristic information required by the operation of the next layer of network.
6. The method according to claim 1, wherein the loss function for training the network model in step 6 is:
wherein L is CE (p i ,y i ) Cross entropy, y, representing multiple classes i Is a true tag, p i Is the predicted label classification probability, N is the total number of samples; GD (g) i ) Is the gradient density;
g i denotes the gradient, gm i Representing the gradient mode length, epsilon is a hyperparameter.
7. A multi-scale sample heterogeneity based mineral prediction system comprising:
a preprocessing module for converting the collected geological data into image data and obtaining 1: p 1 And 1: p 2 The geochemistry of (a), wherein 1: p 1 And 1: p 2 Represents the ratio of the distance on the graph to the actual distance, and P 1 Greater than P 2 (ii) a (ii) a Dividing the geochemical map into grids according to a certain pixel proportion, and forming a training sample according to expert knowledge; dividing each prediction graph into grids, if the grids contain ore beds, selecting the grids as mining areas, and otherwise, selecting the grids as background areas;
and the anomaly identification module is used for introducing the generation countermeasure network into the deep convolution self-encoder, wherein the generator in the generation countermeasure network is used as a decoder in the self-encoder, and the sample size is 1: p 1 The grid data are mapped to a potential space from an image space, then a confrontation network is generated through training to obtain output, and the Euclidean distance between input data and output data is calculated to serve as a geophysical anomaly score;
the data enhancement module is used for enhancing data of the mining area and the obtained chemical exploration abnormal area;
a classification model module to classify the model based on a probability distribution between 1: p 1 The geochemical map of (1): p is 2 Carrying out sample marking on the grid data, marking the grid data with the detection abnormality score higher than a set threshold value as a positive sample, otherwise, carrying out concrete detection data classification on the basis of the positive sample and the negative sample; and construction rollIntegrating a neural network model, and adding an attention mechanism into the model to enable the model to focus information of an interested region so as to learn a rule between mineralization data and chemolithography data;
a model training module to compare 1: p 2 Inputting training sample data into a convolutional neural network model for iteration, and updating parameters of the neural network;
and the classification prediction module is used for carrying out mineralization probability prediction on the chemolithologic data of the region to be predicted by the trained convolutional neural network model to generate a probability prediction distribution map of mineral resources in the region.
8. The system of claim 7, wherein the anomaly identification module trains the encoder with background region data, and the loss function of the encoder training is as follows:
where n is the number of pixels in the image, n d Is the dimension of the intermediate feature representation, k is the weighting factor, | · | | cals | 2 Representing the square of the pixel difference, f (-) is a feature generated by discriminator D during the operation, with the aim of accounting for information given the input.
9. The system of claim 7, wherein the model training module, the convolutional neural network model training loss function is:
wherein L is CE (p i ,y i ) Cross entropy, y, representing multiple classes i Is a true tag, p i Is the predicted label classification probability, N is the total number of samples; GD (g) i ) Is the gradient density;
g i denotes the gradient, gm i Representing the gradient mode length, epsilon is a hyperparameter.
10. A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the computer program when loaded into the processor implements a method of multi-scale sample heterogeneity-based mineral prediction method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211308574.6A CN115511214A (en) | 2022-10-25 | 2022-10-25 | Multi-scale sample unevenness-based mineral product prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211308574.6A CN115511214A (en) | 2022-10-25 | 2022-10-25 | Multi-scale sample unevenness-based mineral product prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115511214A true CN115511214A (en) | 2022-12-23 |
Family
ID=84513423
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211308574.6A Pending CN115511214A (en) | 2022-10-25 | 2022-10-25 | Multi-scale sample unevenness-based mineral product prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115511214A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117095216A (en) * | 2023-08-23 | 2023-11-21 | 湖北省地质调查院 | Model training method, system, equipment and medium based on countermeasure generation network |
| CN117576463A (en) * | 2023-11-20 | 2024-02-20 | 中国地质大学（北京） | Geological structure background prediction method, device and storage medium |
| CN117095216B (en) * | 2023-08-23 | 2024-06-04 | 湖北省地质调查院 | Model training method, system, equipment and medium based on countermeasure generation network |
-
2022
- 2022-10-25 CN CN202211308574.6A patent/CN115511214A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117095216A (en) * | 2023-08-23 | 2023-11-21 | 湖北省地质调查院 | Model training method, system, equipment and medium based on countermeasure generation network |
| CN117095216B (en) * | 2023-08-23 | 2024-06-04 | 湖北省地质调查院 | Model training method, system, equipment and medium based on countermeasure generation network |
| CN117576463A (en) * | 2023-11-20 | 2024-02-20 | 中国地质大学（北京） | Geological structure background prediction method, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110119753B (en) | Lithology recognition method by reconstructed texture | |
| CN109871875B (en) | Building change detection method based on deep learning | |
| Wang et al. | Fast subpixel mapping algorithms for subpixel resolution change detection | |
| Ge | Sub-pixel land-cover mapping with improved fraction images upon multiple-point simulation | |
| Ebrahimy et al. | Per-pixel land cover accuracy prediction: A random forest-based method with limited reference sample data | |
| Moorthi et al. | Kernel based learning approach for satellite image classification using support vector machine | |
| CN112232371B (en) | American license plate recognition method based on YOLOv3 and text recognition | |
| Keck | Machine learning algorithms for the Belle II experiment and their validation on Belle data | |
| CN116206185A (en) | Lightweight small target detection method based on improved YOLOv7 | |
| CN114997501A (en) | Deep learning mineral resource classification prediction method and system based on sample unbalance | |
| CN115511214A (en) | Multi-scale sample unevenness-based mineral product prediction method and system | |
| CN115147615A (en) | Rock image classification method and device based on metric learning network | |
| Xu et al. | An interpretable graph attention network for mineral prospectivity mapping | |
| CN114139819A (en) | Geochemical variable space prediction method based on geostatistical weighted random forest | |
| Nong et al. | Airborne LiDAR point cloud classification using PointNet++ network with full neighborhood features | |
| Wang et al. | Classification and extent determination of rock slope using deep learning | |
| Li et al. | A Markov chain geostatistical framework for land-cover classification with uncertainty assessment based on expert-interpreted pixels from remotely sensed imagery | |
| CN110321528B (en) | Hyperspectral image soil heavy metal concentration assessment method based on semi-supervised geospatial regression analysis | |
| CN117011759A (en) | Method and system for analyzing multi-element geological information of surrounding rock of tunnel face by drilling and blasting method | |
| Zhai et al. | GAN-BiLSTM network for field-road classification on imbalanced GNSS recordings | |
| CN114998719A (en) | Forest fire prediction method based on deep learning and multi-source remote sensing data | |
| Mou et al. | Detecting changes by learning no changes: Data-enclosing-ball minimizing autoencoders for one-class change detection in multispectral imagery | |
| Ung et al. | Leverage samples with single positive labels to train cnn-based models for multi-label plant species prediction | |
| Liu et al. | Peaks Fusion assisted Early-stopping Strategy for Overhead Imagery Segmentation with Noisy Labels | |
| Cevik et al. | Machine learning in the mineral resource sector: An overview |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |