CN111460947B - BP neural network-based method and system for identifying metal minerals under microscope - Google Patents

Abstract

The invention discloses a method and system for identifying metal minerals under a microscope based on a BP neural network, and relates to the technical field of metal mineral identification. Obtain the preliminary characteristic data of the unknown metal mineral image; determine the environmental difference characteristics; calculate the final characteristic data of the unknown metal mineral image according to the preliminary characteristic data of the unknown metal mineral image and the environmental difference characteristics; input the final characteristic data of the unknown metal mineral image into the trained In the BP neural network model, the types of unknown metal minerals are determined. The invention automatically recognizes images through computer software and a reflective polarizing microscope, eliminates the interference of personal factors and environmental factors on the basis of ensuring the low cost of metal mineral recognition, and improves the recognition accuracy.

Description

Recognition method and system of metal minerals under microscope based on BP neural network

technical field

The invention relates to the technical field of metal mineral identification, in particular to a method and system for identifying metal minerals under a microscope based on a BP neural network.

Background technique

With the development of economy and society, the demand for metals is also increasing, and higher requirements are also put forward for the mining and identification of metal ores. There are many identification methods for metal ores, and electron probe analysis technology and reflection polarized light microscopy technology are the common identification methods.

With the development of science and technology and information, the electronic probe analysis technology is also constantly reformed and innovated, and has been continuously upgraded and improved from the previous multi-element analysis technology. However, the electronic probe analysis technology has the following shortcomings: the cost of the electronic probe analysis instrument is expensive, the unit price of an electronic probe analysis instrument is as high as one million yuan, and special maintenance and maintenance are required in the later stage, which is not suitable for ordinary college study and work. Use; the electronic probe analysis instrument is cumbersome and unfavorable to carry, limiting the scope of use; it needs to be operated by people with certain experience.

Reflection polarized light microscopy technology is to identify metal minerals using reflected polarized light microscopy, mainly using the optical properties and physical properties of metal minerals to identify metal minerals. Generally speaking, some minerals with similar reflectance are difficult to measure the reflectance level by simple comparison method, and it is difficult to describe the minerals with similar reflection color in precise language. In addition, due to individual differences among people, especially those with color blindness or color weakness, it is even more difficult to observe the color of minerals under the microscope, such as milky yellow, light yellow, brown, brown, tan, and sometimes it is difficult to distinguish, even experienced Practitioners cannot guarantee 100% accuracy, and the error is large.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method and system for identifying metal minerals under a microscope based on BP neural network, using machine learning means to solve the problems of high cost and large errors in identifying some metal minerals in daily geological activities.

For achieving the above object, the present invention provides following scheme:

A method for identifying metal minerals under a microscope based on BP neural network, including:

Obtain images of unknown metal minerals captured by reflection polarized light microscopy;

Use MATLAB software to process the unknown metal mineral image to obtain the preliminary characteristic data of the unknown metal mineral image;

Determine the environmental difference feature; the environmental difference feature is the difference between the first feature data and the second feature data; the first feature data is the light intensity of the first metal mineral image under the light intensity of the unknown metal mineral image. Feature data, the second feature is the feature data of the second metal mineral image under the light intensity used when training the BP neural network model; the light intensity of the first metal mineral image and the second metal mineral image Different, the metal minerals are of the same type;

According to the preliminary feature data of the unknown metal mineral image and the environmental difference feature, calculate the final feature data of the unknown metal mineral image;

Input the final feature data of the unknown metal mineral image into the trained BP neural network model to determine the type of the unknown metal mineral; wherein, the input of the trained BP neural network model is the final feature data of the unknown metal mineral image, The output of the trained BP neural network model is the type of metal mineral.

Optionally, using MATLAB software to process the unknown metal mineral image to obtain preliminary feature data of the unknown metal mineral image, specifically including:

The unknown metal mineral image is read into MATLAB software by using the imread function; the storage form of the unknown metal mineral image in the MATLAB software is a three-dimensional array form; the three-dimensional array form contains the RGB color information of the unknown metal mineral image;

Perform mean filtering and normalization processing on the unknown metal mineral image sequentially;

Feature extraction is performed on the normalized unknown metal mineral image to obtain preliminary feature data of the unknown metal mineral image.

Optionally, performing feature extraction on the normalized unknown metal mineral image to obtain preliminary feature data of the unknown metal mineral image, specifically including:

Dimensional separation is performed on the normalized image of the unknown metal mineral to obtain three one-dimensional arrays; the three one-dimensional arrays are respectively, the first one-dimensional array, the second one-dimensional array and the third one-dimensional array;

Using the mean function, calculate the mean of each of the one-dimensional arrays;

According to the average value of all the one-dimensional arrays, the preliminary feature data of the unknown metal mineral image is determined; the preliminary feature data of the unknown metal mineral image includes four elements, the first element is the average value of the first one-dimensional array, and the second element is the average value of the first one-dimensional array. The element is the average value of the second one-dimensional array, the third element is the average value of the third one-dimensional array, and the fourth element is the type of the unknown metal mineral.

Optionally, calculating the final feature data of the unknown metal mineral image according to the preliminary feature data of the unknown metal mineral image and the environmental difference feature, specifically including:

The environmental difference feature is superimposed on the preliminary feature data of the unknown metal mineral image to obtain final feature data of the unknown metal mineral image.

Optionally, the steps of training the BP neural network model include:

Obtain different types of historical metal mineral images under the same light intensity;

Use MATLAB software to process all the historical metal mineral images;

Group the processed historical metal mineral images to obtain a training set and a test set;

Use the data in the training set to train the BP neural network model, and when the set training stop condition is met, stop training the BP neural network model to obtain a preliminary BP neural network model;

Use the data in the test set to test the preliminary BP neural network model, and when the set test conditions are met, stop testing the preliminary BP neural network model, and save the preliminary BP neural network model that meets the test conditions; wherein , the preliminary BP neural network model that meets the test conditions is a trained BP neural network model.

A recognition system for metal minerals under microscope based on BP neural network, including:

The unknown metal mineral image acquisition module is used to obtain the unknown metal mineral image captured by the reflection polarized light microscope;

The unknown metal mineral image preliminary feature data calculation module is used to process the unknown metal mineral image by using MATLAB software to obtain the unknown metal mineral image preliminary feature data;

The environmental difference feature determination module is used to determine the environmental difference feature; the environmental difference feature is the difference between the first feature data and the second feature data; the first feature data is the illumination of the image of the unknown metal mineral. Intensity, the characteristic data of the first metal mineral image, and the second characteristic is the characteristic data of the second metal mineral image under the light intensity used when training the BP neural network model; the first metal mineral image and the The light intensity of the second metal mineral image is different, and the metal mineral type is the same;

The final feature data calculation module of the unknown metal mineral image is used to calculate the final feature data of the unknown metal mineral image according to the preliminary feature data of the unknown metal mineral image and the environmental difference feature;

The unknown metal mineral type determination module is used to input the final characteristic data of the unknown metal mineral image into the trained BP neural network model to determine the type of the unknown metal mineral; wherein, the input of the trained BP neural network model is the final feature data of the unknown metal mineral image, and the output of the trained BP neural network model is the type of metal mineral.

Optionally, the unknown metal mineral image preliminary feature data calculation module specifically includes:

The read-in unit is used to read the unknown metal mineral image into the MATLAB software by using the imread function; the storage form of the unknown metal mineral image in the MATLAB software is in the form of a three-dimensional array; the three-dimensional array form contains the unknown metal mineral image RGB color information;

a processing unit, configured to sequentially perform mean filtering and normalization processing on the unknown metal mineral image;

The unit for determining the preliminary feature data of the unknown metal mineral image is used for extracting the features of the normalized unknown metal mineral image to obtain the preliminary feature data of the unknown metal mineral image.

Optionally, the unit for determining the preliminary feature data of the unknown metal mineral image specifically includes:

The dimension separation subunit is used to separate the dimensions of the normalized unknown metal mineral image to obtain three one-dimensional arrays; the three one-dimensional arrays are, respectively, the first one-dimensional array and the second one-dimensional array and the third one-dimensional array;

an average value calculation subunit, used for calculating the average value of each of the one-dimensional arrays using the mean function;

The subunit for determining the preliminary feature data of the unknown metal mineral image is used to determine the preliminary feature data of the unknown metal mineral image according to the average value of all the one-dimensional arrays; the preliminary feature data of the unknown metal mineral image includes four elements, the first The element is the average value of the first one-dimensional array, the second element is the average value of the second one-dimensional array, the third element is the average value of the third one-dimensional array, and the fourth element is the type of unknown metal minerals.

Optionally, the final feature data calculation module of the unknown metal mineral image specifically includes:

The final feature data calculation unit of the unknown metal mineral image is configured to superimpose the environmental difference feature into the preliminary feature data of the unknown metal mineral image to obtain the final feature data of the unknown metal mineral image.

Optionally, the identification system further includes a BP neural network model training module; wherein, the BP neural network model training module specifically includes:

The historical metal mineral image acquisition unit is used to obtain different types of historical metal mineral images under the same light intensity;

A historical metal mineral image processing unit for processing all the historical metal mineral images by using MATLAB software;

The training set and the test set determine the unit, which is used to group the processed historical metal mineral images to obtain the training set and the test set;

The preliminary BP neural network model determination unit is used to train the BP neural network model by using the data in the training set, and when the set training stop condition is satisfied, stop training the BP neural network model, and obtain a preliminary BP neural network Model;

The trained BP neural network model determination unit is used to test the preliminary BP neural network model using the data in the test set, and when the set test conditions are met, stop testing the preliminary BP neural network model, Preliminary BP neural network models that meet the test conditions are saved; wherein, the preliminary BP neural network models that meet the test conditions are trained BP neural network models.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The invention is a method and system for identifying metal minerals under a microscope based on a BP neural network, relying on computer software and a reflection polarizing microscope to automatically identify images, and on the basis of ensuring low identification costs for metal minerals, eliminating the interference of personal factors and environmental factors , improve the recognition accuracy, and expand the scope of use.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings required in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. In the embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative labor.

1 is a flow chart of a method for identifying metal minerals under a microscope based on a BP neural network according to an embodiment of the present invention;

2 is a structural diagram of a recognition system for metal minerals under a microscope based on a BP neural network according to an embodiment of the present invention;

Fig. 3 is the galena map under the 40 times mirror of the embodiment of the present invention;

Fig. 4 is the luminance distribution diagram of the galena picture surface of the embodiment of the present invention;

Fig. 5 is the boron mafic ore picture under the 40 times mirror of the embodiment of the present invention;

6 is a diagram of sphalerite under a 40-fold mirror according to an embodiment of the present invention;

FIG. 7 is a structural diagram of a BP neural network model according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

Example 1

As shown in FIG. 1 , a method for identifying metal minerals under a microscope based on a BP neural network provided in this embodiment specifically includes the following steps.

Step 101: Acquire an image of an unknown metal mineral photographed by a reflection polarizing microscope.

Step 102: Use MATLAB software to process the image of the unknown metal mineral to obtain preliminary characteristic data of the image of the unknown metal mineral.

Step 103: Determine the environmental difference characteristic; the environmental difference characteristic is the difference between the first characteristic data and the second characteristic data; Feature data of the mineral image, the second feature is the feature data of the second metal mineral image under the light intensity used when training the BP neural network model; the first metal mineral image and the second metal mineral image The light intensity is different, and the types of metal minerals are the same.

Step 104: Calculate the final feature data of the unknown metal mineral image according to the preliminary feature data of the unknown metal mineral image and the environmental difference feature.

Step 105: Input the final feature data of the unknown metal mineral image into the trained BP neural network model to determine the type of the unknown metal mineral; wherein, the input of the trained BP neural network model is the final unknown metal mineral image. Feature data, the output of the trained BP neural network model is the type of metal mineral.

Step 102 specifically includes:

The unknown metal mineral image is read into MATLAB software by using the imread function; the storage form of the unknown metal mineral image in the MATLAB software is a three-dimensional array form; the three-dimensional array form contains the RGB color information of the unknown metal mineral image.

Perform mean filtering and normalization processing on the unknown metal mineral image sequentially.

Feature extraction is performed on the normalized unknown metal mineral image to obtain preliminary feature data of the unknown metal mineral image. Specifically: the first step is to separate the dimensions of the normalized unknown metal mineral image to obtain three one-dimensional arrays; the three one-dimensional arrays are, respectively, the first one-dimensional array and the second one-dimensional array and the third one-dimensional array; the second step, using the mean function, to calculate the average value of each of the one-dimensional arrays; the third step, according to the average value of all the one-dimensional arrays, determine the preliminary characteristics of the unknown metal mineral image data; the preliminary feature data of the unknown metal mineral image includes four elements, the first element is the average value of the first one-dimensional array, the second element is the average value of the second one-dimensional array, and the third element is the third one-dimensional array The average value of the array, the fourth element is the type of the unknown metal mineral.

Step 104 specifically includes:

The environmental difference feature is superimposed on the preliminary feature data of the unknown metal mineral image to obtain final feature data of the unknown metal mineral image.

Preferably, the present invention also includes the step of training the BP neural network model, and the step is specifically:

Obtain historical metal mineral images of different species under the same light intensity.

All the historical metal mineral images were processed using MATLAB software.

The processed historical metal mineral images are grouped to obtain a training set and a test set.

Use the data in the training set to train the BP neural network model, and stop training the BP neural network model when the set training stop condition is satisfied to obtain a preliminary BP neural network model.

Use the data in the test set to test the preliminary BP neural network model, and when the set test conditions are met, stop testing the preliminary BP neural network model, and save the preliminary BP neural network model that meets the test conditions; wherein , the preliminary BP neural network model that meets the test conditions is a trained BP neural network model.

Embodiment 2

As shown in Figure 2, this embodiment also provides a recognition system for metal minerals under a microscope based on BP neural network, including:

The unknown metal mineral image acquisition module 201 is used to obtain an unknown metal mineral image captured by a reflection polarized light microscope.

The unknown metal mineral image preliminary feature data calculation module 202 is used to process the unknown metal mineral image by using MATLAB software to obtain the unknown metal mineral image preliminary feature data.

The environmental difference feature determination module 203 is used to determine the environmental difference feature; the environmental difference feature is the difference between the first feature data and the second feature data; the first feature data is the image of the unknown metal mineral when the image is taken. Under the light intensity, the characteristic data of the first metal mineral image, and the second feature is the characteristic data of the second metal mineral image under the light intensity used when training the BP neural network model; the first metal mineral image and The light intensity of the second metal mineral image is different, and the metal mineral type is the same.

The final feature data calculation module 204 of the unknown metal mineral image is configured to calculate the final feature data of the unknown metal mineral image according to the preliminary feature data of the unknown metal mineral image and the environmental difference feature.

The unknown metal mineral type determination module 205 is used to input the final feature data of the unknown metal mineral image into the trained BP neural network model to determine the type of the unknown metal mineral; wherein, the trained BP neural network model The input is the final feature data of the unknown metal mineral image, and the output of the trained BP neural network model is the type of metal mineral.

The unknown metal mineral image preliminary feature data calculation module 202 specifically includes:

The read-in unit is used to read the unknown metal mineral image into the MATLAB software by using the imread function; the storage form of the unknown metal mineral image in the MATLAB software is in the form of a three-dimensional array; the three-dimensional array form contains the unknown metal mineral image RGB color information.

The processing unit is configured to sequentially perform mean filtering and normalization processing on the unknown metal mineral image.

The unit for determining the preliminary feature data of the unknown metal mineral image is used for extracting the features of the normalized unknown metal mineral image to obtain the preliminary feature data of the unknown metal mineral image.

Wherein, the unit for determining the preliminary characteristic data of the unknown metal mineral image specifically includes:

The dimension separation subunit is used to separate the dimension of the normalized image of unknown metal minerals to obtain three one-dimensional arrays; the three one-dimensional arrays are, respectively, the first one-dimensional array, the second one-dimensional array and a third one-dimensional array.

The mean value calculation subunit is used to calculate the mean value of each of the one-dimensional arrays using the mean function.

The subunit for determining the preliminary feature data of the unknown metal mineral image is used to determine the preliminary feature data of the unknown metal mineral image according to the average value of all the one-dimensional arrays; the preliminary feature data of the unknown metal mineral image includes four elements, the first The element is the average value of the first one-dimensional array, the second element is the average value of the second one-dimensional array, the third element is the average value of the third one-dimensional array, and the fourth element is the type of unknown metal minerals.

The final feature data calculation module 204 of the unknown metal mineral image specifically includes:

The final feature data calculation unit of the unknown metal mineral image is configured to superimpose the environmental difference feature into the preliminary feature data of the unknown metal mineral image to obtain the final feature data of the unknown metal mineral image.

Preferably, the identification system further includes a BP neural network model training module; wherein, the BP neural network model training module specifically includes:

The historical metal mineral image acquisition unit is used to obtain different types of historical metal mineral images under the same light intensity.

The historical metal mineral image processing unit is used to process all the historical metal mineral images by using MATLAB software.

The training set and the test set determine the unit for grouping the processed historical metal mineral images to obtain the training set and the test set.

The preliminary BP neural network model determination unit is used to train the BP neural network model by using the data in the training set, and when the set training stop condition is satisfied, stop training the BP neural network model, and obtain a preliminary BP neural network Model.

The trained BP neural network model determination unit is used to test the preliminary BP neural network model using the data in the test set, and when the set test conditions are met, stop testing the preliminary BP neural network model, Preliminary BP neural network models satisfying the test conditions are saved; wherein, the preliminary BP neural network models satisfying the test conditions are trained BP neural network models.

Embodiment 3

Introduction to Recognition Principles

The RGB color mode is a color standard in the industry. It obtains various colors by changing the three color channels of red (R), green (G), and blue (B) and superimposing them on each other. Yes, RGB represents the color of the three channels of red, green and blue. This standard includes almost all the colors that human vision can perceive, and it is one of the most widely used color systems.

Under normal circumstances, RGB has 256 levels of brightness, which are represented by numbers from 0, 1, 2... until 255, where 0 means the darkest, 255 means the brightest, and the value varies with the brightness, and this The change is linear. 256 levels of RGB colors can combine a total of about 16.78 million colors, that is, 256×256×256=16777216. Since there are no metal minerals with the same color and brightness, if all image information can be recorded when shooting, there can be about 16.78 million metal minerals with color brightness information, which lays the RGB color components as the basis for identifying features. In the computer MATLAB software, the RGB color components are stored in the form of a three-dimensional matrix, and each dimension stores a color component.

For example, read in a picture of galena (as shown in Figure 3) and study its brightness (brightness is represented by the RGB color components) distribution (as shown in Figure 4). It is found that the surface brightness of galena is relatively uniform. In fact, the reflectivity of metallic minerals such as galena (which is positively correlated with brightness) is a certain value, which means that the brightness distribution of metallic minerals is relatively uniform. in a very small range.

Introduction to Neural Network Principles

The BP (Back Propagation) network was proposed in 1986 by a group of scientists headed by Rumelhart and McCelland. It is a multi-layer feedforward network trained by the error back-propagation algorithm. It is one of the most widely used neural network models. BP network can learn and store a large number of input-output pattern mapping relationships without revealing the mathematical equations describing this mapping relationship in advance. Its learning rule is to use the steepest descent method to continuously adjust the weights and thresholds of the network through backpropagation to minimize the sum of squared errors of the network. The topology of BP neural network model includes input layer (input), hidden layer (hide layer) and output layer (output layer). Each layer is composed of units (also called neural nodes). The input layer has the characteristics of the training set, and is passed to the next layer through the weights of the connection points. The output of one layer is the input of the next layer, and the number of hidden layers is Arbitrarily, each layer is weighted and summed, and then output according to the nonlinear transformation equation. Through the BP (Backpropagation) algorithm, that is, comparing the error between the predicted value of the neural network output layer and the actual value, and minimizing the error in the opposite direction (from the output layer => hidden layer => input layer), the weight of each connection is updated. At present, it has been widely used in classification and recognition, approximation, regression, compression and other fields.

Based on the above principles, the method for identifying metal minerals under a microscope based on a BP neural network provided in this embodiment mainly includes the following steps.

Step 1: Image acquisition.

Specifically, it includes: using an OLYMPUS-BX51M reflective polarizing microscope, setting the exposure time to 750 μS, the sensitivity (ISO) of 200, and the resolution of 1600×1200. Minerals such as galena, sphalerite and boron mafic were collected by the digital camera of the reflection polarizing microscope. About 20 photos of each mineral are shown in Figures 3, 5 and 6.

Step 2: Image processing and feature extraction.

Put the image into the root path specified by the MATLAB software, and use its own function to import the image. At this time, the image is stored in the MATLAB software as a three-dimensional digital group. This three-dimensional digital group is the RGB color information of the image.

Image processing specifically includes:

1. Digitization; use the imread function to read the image into the MATLAB software, the image at this time becomes a three-dimensional matrix, and the size of each element in the matrix is between 0-256.

2. Mean filtering; use the fspecial('average') function in MATLAB software to filter the image. Mean filtering is also called linear filtering. It mainly adopts the neighborhood averaging method. The core idea of mean filtering is to treat the entire image as composed of many small blocks with constant grayscale. The correlation between adjacent pixels is strong, but the noise is statistically independent. sex. Therefore, the average value of the neighborhood can be used to replace each pixel value in the original image. This processing is mainly to remove noise.

3. Normalization; use the premnmx(P, T) function in MATLAB software for normalization, where P, T are the original input and output data, respectively. Normalization is to turn a number into a decimal between (0, 1). One is to eliminate dimensions, and the other is to facilitate data processing.

Feature extraction specifically includes:

1. Dimension separation; because the read image is a three-dimensional array, it is now necessary to divide the three-dimensional array into three one-dimensional arrays.

2. Average value: find the average value of each dimension of the three-dimensional array; directly use the mean function to find the average value of each dimension array. There are three means in three dimensions, and one picture gets three means, and these three means are stored in an array of size 4, the first three store the mean, and the fourth stores the mineral species. For example: galena: array{250, 253, 253, galena}. There are 20 images for each mineral, resulting in a total of 20 arrays.

Step 3: Training of BP neural network model.

First, the image feature data is divided into training set and test set; each mineral has 20 arrays, of which 15 groups are used as training sets and 5 groups are used as test sets. The training set of all minerals is a training set, and the test set is a test set.

Secondly, neural network training; as shown in Figure 7, the essence of neural network training is to make the error between the output value of a certain mineral and the mineral label small enough, use W _ij to represent the weight, and θ to represent the bias, and the main line of this process is realized. It is to update the weights and biases in the training process, that is, to get the most suitable weights and biases, because the values of the hidden layer and output layer in the neural network mainly depend on the weights and biases. Specifically include:

1. Input layer -> hidden layer -> output layer.

The three nodes of the input layer, x1, x2, x3, are the eigenvalues of human input, which are reflected in the MATLAB software function as net=newff(P,....), and P is the input eigenvalues x1, x2, x3 's matrix. Then it propagates backward from the input layer, specifically to the hidden layer or forward. At this time, it is necessary to calculate the value of the two nodes of the hidden layer, multiply the value of each input layer by the weight of the corresponding node, and then sum it up. Add the bias to get a value, which is then fed into the transfer function sigmoid(f(x)=1/1+ex). For example, the value of node 4 is f(x)=1/1+ex, where x=x1×w ₁₄ +x2×w ₂₄ +x3×w ₃₄ +θ ₁ , the value of node 5 can be obtained in the same way , where the value of each node is represented by O _i .

After getting the value of the hidden layer, the neural network continues to propagate backwards. Since the values of nodes 4 and 5 are already known, the value of node 6, that is, the value of the output layer, can be obtained in the same way. The next step is to calculate the error of the output layer.

2. The error of the output layer.

Because the purpose of training is to minimize the error between the output value of a certain mineral and the label of this mineral. In fact know the real label of the output layer, which is already included in the training set. For example, the eigenvalues {190, 195, 205} of galena are input, and the corresponding mark of galena is 1, then the error between the value of the output layer and the corresponding mark needs to be calculated, and the corresponding calculation formula is Err _j =O _j (1- _Oj )( _{Tj-Oj )} .

Among them, T _j is the corresponding mark, O _i is the value of the error node to be calculated, and the error between the predicted value and the actual value of the output layer can be calculated. If the error is small enough, then stop training and save the existing Neural network weights and biases. Otherwise, continue training to update weights and biases. The updated strategy is backpropagation, that is, the output layer goes to the hidden layer but the input layer.

3. Output layer -> hidden layer -> input layer.

4. Hidden layer error.

The hidden layer is the same as the output layer, and the error of each node in the hidden layer needs to be calculated. The formula is

Among them, Err _k is the error of the node in the next layer of the calculated node. w _jk is the weight of this node and the next layer of nodes. For example, to calculate the error of node 5, Err ₅ =O ₅ (1-O ₅ )Err ₆ *w ₅₆ . If there are multiple output layer nodes, the error of each output layer node can be multiplied by the weight corresponding to the previous layer node by the formula, and then multiplied by o _j (1-O _j ). Similarly, the error of node 4 can be obtained.

After the errors of nodes 4, 5, and 6 are obtained, the weights and biases need to be updated next.

5. Weight update.

The weight update formula is:

Δw _ij =(l) Err _j O _j ;

w _ij =w _ij +Δw _ij ;

Δw _ij is the variation of the weight, which is equal to the error of the node multiplied by the value of the node and then multiplied by the learning rate l (generally the default value, do not need to be set by yourself), and then add the original weight to be the new weight. Through the above formula, the weights of all nodes can be calculated.

6. Biased to update.

The weight update formula is:

Δθ _j =(l) Err _j

θ _j = θ _j +Δθ _j

The weight update formula is similar to the weight update formula, but the value of Δθ becomes the learning rate multiplied by the error, without the value of the node. All biases are updated by the above formula.

The above process is a complete training process. After the weights and biases are updated, a set of eigenvalues are input to the neural network to continue to repeat the above process.

The conditions for stopping the training process: the weights are lower than a certain threshold; the predicted error rate is lower than a certain threshold; the preset number of cycles is reached.

After stopping, the optimal weights and biases should be obtained at this time, and the network data saved at this time can be used to predict and classify minerals.

The core code is to input the training set into the neural network. Then train. This process mainly uses the characteristics of the neural network to find an optimal input (the first three items of the array) output (the fourth item of the array) mapping. The main steps are as follows:

1. Use net=newff(P, [...], {...}) to create a neural network, where P is the input data, and ellipses are other parameters.

2.net.trainParam.epochs=5000; set the number of training times.

3.net.trainParam.goal=0.0001; set the convergence error.

4. [net, tr]=train(net); train the network.

Finally, use the test set data to test the accuracy

After training, the test set is input as a parameter to the above trained neural network to study the correctness of its output results. For example, input the test data of galena, whether the output of the observation period is galena. If the accuracy is low, repeat training and testing.

Step 4: Input the images collected in real time into the trained BP neural network model to realize the identification of metal minerals.

The currently extracted three-dimensional statistical features are the brightness and color features of the image, which are related to the light intensity (also related to the instrument) when the microscope is photographed. The following method is adopted: that is, at the beginning of training the network, a standard image (marked as A1) is taken and the feature F={x1, x2, x3, x4} of the standard image is extracted. When identifying the unknown mineral image (change the light intensity when shooting), then use the light that changes the light intensity (or instrument) to collect an image of the standard mineral A1 and extract the features of the standard mineral A1 F′={Z1, Z2, Z3, Z4}, then subtract F' and F to get a difference Y={y1, y2, y3, y4} (correction number), and then add this correction number to the characteristics of the unknown mineral to be identified, thus avoiding The problem of identification errors caused by different lighting or different instruments.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method.

In this paper, specific examples are used to illustrate the principles and implementations of the present invention. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present invention; meanwhile, for those skilled in the art, according to the present invention There will be changes in the specific implementation and application scope. In conclusion, the contents of this specification should not be construed as limiting the present invention.

Claims (6)

1. A method for identifying microscopic metal minerals based on a BP neural network is characterized by comprising the following steps:

acquiring an unknown metal mineral image shot by a reflection polarization microscope;

processing the unknown metal mineral image by using MATLAB software to obtain primary characteristic data of the unknown metal mineral image;

determining the difference characteristics of the environment; the environmental difference characteristic is a difference value between the first characteristic data and the second characteristic data; the first characteristic data is characteristic data of a first metal mineral image under the illumination intensity of an unknown metal mineral image, and the second characteristic data is characteristic data of a second metal mineral image under the illumination intensity adopted in the process of training a BP neural network model; the illumination intensity of the first metal mineral image is different from that of the second metal mineral image, and the types of metal minerals are the same;

calculating final characteristic data of the unknown metal mineral image according to the primary characteristic data of the unknown metal mineral image and the environment difference characteristics;

inputting the final characteristic data of the unknown metal mineral image into a trained BP neural network model to determine the type of the unknown metal mineral; the input of the trained BP neural network model is final characteristic data of an unknown metal mineral image, and the output of the trained BP neural network model is the type of a metal mineral;

calculating final feature data of the unknown metal mineral image according to the preliminary feature data of the unknown metal mineral image and the environment difference features, wherein the calculating comprises the following steps:

superposing the environmental difference characteristics to the primary characteristic data of the unknown metal mineral image to obtain final characteristic data of the unknown metal mineral image;

the step of training the BP neural network model comprises the following steps:

acquiring historical metal mineral images of different types under the same illumination intensity;

processing all the historical metal mineral images by using MATLAB software;

grouping the processed historical metal mineral images to obtain a training set and a test set;

training a BP neural network model by using the data in the training set, and stopping training the BP neural network model when a set training stopping condition is met to obtain a preliminary BP neural network model;

testing the preliminary BP neural network model by adopting the data in the test set, stopping testing the preliminary BP neural network model when set test conditions are met, and storing the preliminary BP neural network model meeting the test conditions; and the preliminary BP neural network model meeting the test conditions is a trained BP neural network model.

2. The method for identifying the metal mineral under the microscope based on the BP neural network according to claim 1, wherein the processing the unknown metal mineral image by using MATLAB software to obtain the preliminary feature data of the unknown metal mineral image specifically comprises:

reading the unknown metal mineral image into MATLAB software by adopting an imread function; the storage form of the unknown metal mineral image in MATLAB software is a three-dimensional array form; the three-dimensional array form comprises RGB color information of unknown metal mineral images;

sequentially carrying out mean value filtering and normalization processing on the unknown metal mineral image;

and performing feature extraction on the normalized unknown metal mineral image to obtain primary feature data of the unknown metal mineral image.

3. The method for identifying the metal mineral under the microscope based on the BP neural network according to claim 2, wherein the step of performing feature extraction on the normalized unknown metal mineral image to obtain the preliminary feature data of the unknown metal mineral image specifically comprises the steps of:

carrying out dimension separation on the normalized unknown metal mineral image to obtain three one-dimensional arrays; the three one-dimensional arrays are respectively a first one-dimensional array, a second one-dimensional array and a third one-dimensional array;

calculating the average value of each one-dimensional array by using a mean function;

determining the primary characteristic data of the unknown metal mineral image according to the average value of all the one-dimensional arrays; the preliminary feature data of the unknown metal mineral image comprises four elements, wherein the first element is an average value of a first one-dimensional array, the second element is an average value of a second one-dimensional array, the third element is an average value of a third one-dimensional array, and the fourth element is the type of the unknown metal mineral.

4. A recognition system for microscopic metal minerals based on a BP neural network is characterized by comprising:

the unknown metal mineral image acquisition module is used for acquiring an unknown metal mineral image shot by a reflection polarization microscope;

the unknown metal mineral image preliminary characteristic data calculation module is used for processing the unknown metal mineral image by using MATLAB software to obtain unknown metal mineral image preliminary characteristic data;

the environment difference characteristic determining module is used for determining environment difference characteristics; the environment difference characteristic is a difference value of the first characteristic data and the second characteristic data; the first characteristic data is characteristic data of a first metal mineral image under the illumination intensity of an unknown metal mineral image, and the second characteristic data is characteristic data of a second metal mineral image under the illumination intensity adopted in the process of training a BP neural network model; the illumination intensity of the first metal mineral image is different from that of the second metal mineral image, and the types of metal minerals are the same;

the unknown metal mineral image final characteristic data calculation module is used for calculating the unknown metal mineral image final characteristic data according to the unknown metal mineral image preliminary characteristic data and the environment difference characteristics;

the unknown metal mineral type determining module is used for inputting the final characteristic data of the unknown metal mineral image into a trained BP neural network model to determine the type of the unknown metal mineral; the input of the trained BP neural network model is final characteristic data of an unknown metal mineral image, and the output of the trained BP neural network model is the type of a metal mineral;

the unknown metal mineral image final feature data calculation module specifically comprises:

the unknown metal mineral image final characteristic data calculation unit is used for superposing the environmental difference characteristics to the unknown metal mineral image preliminary characteristic data to obtain the unknown metal mineral image final characteristic data;

the recognition system also comprises a BP neural network model training module; the BP neural network model training module specifically comprises:

the historical metal mineral image acquisition unit is used for acquiring different types of historical metal mineral images under the same illumination intensity;

the historical metal mineral image processing unit is used for processing all the historical metal mineral images by using MATLAB software;

the training set and test set determining unit is used for grouping the processed historical metal mineral images to obtain a training set and a test set;

a preliminary BP neural network model determining unit, configured to train the BP neural network model by using the data in the training set, and stop training the BP neural network model when a set training stop condition is met, so as to obtain a preliminary BP neural network model;

the trained BP neural network model determining unit is used for testing the preliminary BP neural network model by adopting the data concentrated by the test, stopping testing the preliminary BP neural network model when set test conditions are met, and storing the preliminary BP neural network model meeting the test conditions; and the preliminary BP neural network model meeting the test conditions is a trained BP neural network model.

5. The system for identifying the metal mineral under the microscope based on the BP neural network as claimed in claim 4, wherein the unknown metal mineral image preliminary feature data calculation module specifically comprises:

the reading unit is used for reading the unknown metal mineral image into MATLAB software by adopting an imread function; the storage form of the unknown metal mineral image in MATLAB software is a three-dimensional array form; the three-dimensional array form comprises RGB color information of unknown metal mineral images;

the processing unit is used for sequentially carrying out mean value filtering and normalization processing on the unknown metal mineral image;

and the unknown metal mineral image preliminary characteristic data determining unit is used for extracting the characteristics of the normalized unknown metal mineral image to obtain the unknown metal mineral image preliminary characteristic data.

6. The system for identifying the metal mineral under the microscope based on the BP neural network as claimed in claim 5, wherein the unknown metal mineral image preliminary characteristic data determination unit specifically comprises:

the dimension separation subunit is used for performing dimension separation on the normalized unknown metal mineral image to obtain three one-dimensional arrays; the three one-dimensional arrays are respectively a first one-dimensional array, a second one-dimensional array and a third one-dimensional array;

the mean value operator unit is used for calculating the mean value of each one-dimensional array by using a mean function;

the unknown metal mineral image preliminary characteristic data determining subunit is used for determining the unknown metal mineral image preliminary characteristic data according to the average value of all the one-dimensional arrays; the preliminary characteristic data of the unknown metal mineral image comprises four elements, wherein the first element is the average value of a first one-dimensional array, the second element is the average value of a second one-dimensional array, the third element is the average value of a third one-dimensional array, and the fourth element is the type of the unknown metal mineral.

Images