Mine time shift resistivity monitoring data screening method, storage medium and software

Abstract

The invention discloses a mine time shift resistivity monitoring data screening method, a storage medium and software. The problems that in the prior art, data screening mainly depends on artificial experience, the intelligent degree is insufficient, and efficiency is not high are solved. The method comprises the following steps: firstly, performing dimension reduction on high-dimensional data to obtain essential features of the data, and removing data redundancy to realize feature extraction; feature selection is carried out in the feature space after dimension reduction; and finally, screening the monitoring data in the feature space by using a spectral clustering method. The method can be realized by means of a software technology.

Description

Technical Field

[0001] The invention belongs to the technical field of mine water prevention and control, and relates to a method for intelligent screening of mine time-lapse resistivity monitoring data inversion results based on artificial intelligence. Background Art

[0002] Time-lapse resistivity monitoring in mines is a scientific and effective technical means for coal mine geological transparency and water hazard monitoring and early warning. Time-lapse resistivity monitoring belongs to continuous electrical monitoring, and multiple sets of monitoring results are obtained every day. However, during the mining process of the coal mining face, the destruction of the roof and floor plates and the development of water channels are dynamic and slowly changing processes, and the monitoring results at different times of the day are highly similar; at the same time, affected by mining disturbances, the monitoring data inevitably contains noise interference, so it is necessary to scientifically select a set of reliable data from the daily monitoring data. At present, the mine time-lapse resistivity monitoring data are all displayed in the form of resistivity cloud maps, which mainly rely on manual experience for data screening, and the degree of intelligence is insufficient. Summary of the invention

[0003] In view of the defects or shortcomings of the prior art, the present invention provides a method for screening mine time-lapse resistivity monitoring data, wherein the mine time-lapse resistivity monitoring data includes data obtained from M monitorings of the same mine area, M≥2; wherein the mth monitoring performs several data collections, m=1, 2, 3,…, M; and each data collection obtains the resistivity value at the spatial coordinates (x, y, z).

[0004] The mine time-lapse resistivity monitoring data screening method provided by the present invention comprises:

[0005] Step 1: extract the resistivity values ​​at different x spatial positions in the mth monitoring data, and normalize the resistivity values ​​at the same x spatial coordinate position. n All resistivity values ​​at the spatial coordinate position are normalized to obtain x = x n Normalized resistivity sequence R corresponding to the spatial coordinate position scale，n ; n=1,2,3,…,N,N≥2, and x n With x n+1 The x-direction spatial positions are adjacent; N normalized resistivity sequences corresponding to the m-th monitoring data are obtained; m = 1, 2, 3, ..., M;

[0006] Step 2: Visualize the N normalized resistivity series corresponding to the mth monitoring data to obtain the corresponding atlas Img m ={img1,img2,img3,…img n ,…,img N}, img n For x = xn Resistivity plane contour map corresponding to the spatial coordinate position; n = 1, 2, 3, ..., N; m = 1, 2, 3, ..., M;

[0007] Step 3: Use the image classification network model to obtain the atlas Img m The high-dimensional eigenvectors of the resistivity plane contour maps in ; find I n with I n+1 The cosine similarity of n with I n+1 Respectively, they are the resistivity plane contour map img n and img n+1 The high-dimensional feature vector of n = 1, 2, 3, ..., N-1; sum up the N-1 cosine similarities to obtain the structural similarity S corresponding to the m-th monitoring data m ; m = 1, 2, 3, ..., M;

[0008] Step 4: Use the Transformer neural network to extract the atlas Img m The multidimensional feature vector F m ; m = 1, 2, 3, ..., M;

[0009] Step 5, screening the M multi-dimensional feature vectors, including:

[0010] Step 5.1, selecting a multidimensional feature vector according to the size of the structural similarity, and selecting the multidimensional feature vectors corresponding to the corresponding sub-monitorings with multiple structural similarities having the largest size;

[0011] Step 5.2, using the K-means clustering (KMeans) algorithm to divide the selected multiple multidimensional feature vectors into two categories; calculating the cumulative sum of the structural similarities corresponding to each multidimensional feature vector in each category, and obtaining the sum of the structural similarities of each category;

[0012] Step 5.3, select the category that meets the screening condition and execute step 5.5, otherwise execute step 5.4; the screening condition is that the category with the largest number of multidimensional feature vectors and the largest sum of structural similarities in the two categories is selected as the screening object;

[0013] Step 5.4, remove the multidimensional feature vector with the smallest corresponding structural similarity among the multiple multidimensional feature vectors classified by the K-means clustering (KMeans) algorithm last time, and execute steps 5.2 and 5.3 on the remaining multidimensional feature vectors in sequence, until the category with the largest number of multidimensional feature vectors and the largest sum of structural similarities is selected according to the screening conditions of step 5.3 to execute step 5.5;

[0014] Step 5.5, for the category selected according to step 5.3, calculate the Euclidean distance between each multidimensional feature vector in the category and the class center vector; select the monitoring data corresponding to the multidimensional feature vector with the smallest Euclidean distance as the screened data.

[0015] An optional solution is that the sampling space coordinate points of the M monitorings are the same or identical.

[0016] An optional solution is that the normalization of the resistivity values ​​at the same x spatial coordinate position in step 1 includes: first performing mean-variance normalization on multiple resistivity values, and then performing maximum normalization on the mean-variance normalized values; x = x n All the normalized values ​​corresponding to the spatial coordinate positions constitute x=x n Normalized resistivity series R corresponding to the spatial coordinate position scale，n .

[0017] An optional solution is that the visualization of the N normalized resistivity series corresponding to the m-th monitoring data in step 2 includes: for any x=x n Normalized resistivity series R corresponding to the spatial coordinate position scale，n Gray value conversion, interpolation processing, edge expansion to adjust image size and colorization are performed in sequence.

[0018] An optional solution is that the image classification network model is a ResNet10 network trained using the ImageNet dataset. Further, the ResNet10 network includes a two-dimensional convolutional layer conv Layer, a maximum pooling layer Maxpoo, a basic residual module, a residual module and a linear layer FC connected in sequence, wherein the residual module and the linear layer FC are connected by a flattening operation.

[0019] An optional solution is that the high-dimensional feature vector is a 1024-dimensional feature vector.

[0020] An optional solution is that the Transformer neural network includes two three-dimensional convolutional layers conv3D Layer, a BatchNorm layer, a non-linear activation function ReLU, a Flattening operation, a first linear layer LinearLayer, a YZ-Transformer encoder, an X-Transformer encoder and a second linear layer Linear Layer connected in sequence.

[0021] An optional solution is that the multi-dimensional feature vector is a 128-dimensional feature vector.

[0022] An optional solution is that step 5.1 includes: arranging the M structural similarities from large to small, and taking the multi-dimensional feature vectors corresponding to the corresponding monitoring of the first 0.6 to 0.8M structural similarities with large structure.

[0023] The present invention also provides a related storage medium on which a computer program / instruction is stored, characterized in that the steps of the above method are implemented when the computer program / instruction is executed by a processor.

[0024] The present invention also provides a related software product, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0025] The present invention solves the problem that the existing technology mainly relies on manual experience for data screening, which is not intelligent enough and inefficient. The technical solution adopted by the present invention is that the time-lapse resistivity monitoring data contains five-dimensional attributes such as spatial three-dimensional coordinate information, time information and resistivity value, which belongs to high-dimensional discrete data, and there will be dimensionality disasters such as sparse data samples, increased computational complexity, and visualization difficulties. In order to overcome the dimensionality disaster, the present invention first reduces the dimensionality of high-dimensional data, obtains the essential characteristics of the data, removes data redundancy, and realizes feature extraction; then performs feature selection in the feature space after dimensionality reduction; finally, uses the spectral clustering method to screen the monitoring data in the feature space. The present invention can be implemented with the help of software technology, such as python language programming. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The lightweight residual network ResNet 10 network structure adopted in the embodiment of the present invention;

[0027] Figure 2 The Transformer encoder network structure used in the embodiment of the present invention;

[0028] Figure 3 1 is an example of a resistivity plane contour map obtained in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Unless otherwise specified, the science and technology in this article are based on common knowledge in the relevant fields.

[0030] For the data coordinates selected by the present invention, in physical space, x generally represents the excavation direction of the mine tunnel, y coordinate represents the width direction of the tunnel, and z coordinate represents the height direction of the tunnel.

[0031] The present invention normalizes the resistivity values ​​at the same x spatial coordinate position. In a specific solution, the existing normalization method can be used to normalize the original resistivity value to a value of [0,1]. In a specific example, the original data can be firstly normalized by mean variance, and then the mean variance normalized value can be normalized by the maximum value. The specific calculation process is as follows:

[0032] Step 1.1: Extract x=x from the mth monitoring data n All resistivity values ​​at spatial locations; n = 1, 2, 3, ..., N;

[0033] Step 1.2, find x=x n The mean value of all resistivity values ​​at a spatial position is R mean，n and variance R σ,n , n=1,2,3,…,N;

[0034] Step 1.3: According to formula (1), x = x n Each resistivity value at the spatial position is normalized by mean variance; n = 1, 2, 3, ..., N;

[0035]

[0036] In the formula, r xn For x = x n Any resistivity value at any spatial position, R xn For r xn The mean-variance normalized value of ;

[0037] Step 1.4: Count the x=x obtained in step 1.3 n The maximum value R among the normalized values ​​of the resistivity values ​​at the spatial position xn,max and the minimum value R xn,min , according to formula (2), the mean variance normalized values ​​obtained in step 1.3 are normalized to the maximum value;

[0038]

[0039] R xn, For R xn The maximum normalized value of

[0040] x=x n All the normalized values ​​corresponding to the spatial positions constitute x=x n Normalized resistivity series R corresponding to the spatial position scale，n .

[0041] The image classification network model of the present invention is used to obtain the atlas Img mThe high-dimensional feature vector of each resistivity plane contour map in the image. For example, a ResNet10 network trained with a public image classification dataset (such as an ImageNe t image classification dataset) can be used. The ResNet10 network is preferably a pruned ResNet10 network. A specific ResNet10 network includes a two-dimensional convolutional layer conv Layer, a maximum pooling layer Maxpoo, a basic residual module, a residual module, and a linear layer FC connected in sequence, wherein the residual module and the linear layer FC are connected by a flattening operation; the input of the network is a resistivity plane contour map; after the convolution operation and the maximum pooling layer processing, the first intermediate feature map is output; then the second intermediate feature map is processed by the basic residual module; then the third intermediate feature map is processed by the residual module and output; finally, the third intermediate feature map is flattened and input into the linear layer FC, and the output is a high-dimensional feature vector.

[0042] The present invention uses Transformer neural network to extract the atlas Img m The multidimensional feature vector F m The Transformer neural network is preferably a pruned Transformer neural network. A specific Transformer neural network structure includes two layers of three-dimensional convolutional layers conv3D Layer, BatchNorm layer, nonlinear activation function ReLU, Flattening operation, first linear layer Linear Layer, YZ-Transformer encoder, X-Transformer encoder and second linear layer Linear Layer connected in sequence. The input of the network is the atlas Img m , for the atlas Img m Reshape is performed to obtain the input data input_image with n channels and a width and height of 224. The convolution kernel is used to perform sliding window convolution on the input_image to obtain the feature representation of the image; then enter the second three-dimensional convolution layer to obtain the features of the target size; the obtained features are flattened into vectors and input into the linear layer to obtain multi-dimensional tokens; then the multi-dimensional tokens are divided into multiple groups, and a position code is added to each token, which are integrated into new feature vectors, and then enter multiple independent YZ-Transformer encoders respectively to obtain purified multi-dimensional features; then the multi-dimensional features output by multiple groups of YZ-Transformer encoders are spliced ​​to obtain YZ plane space features; and the features are input into the X-Transformer encoder for feature integration to obtain a new multi-dimensional feature; the new multi-dimensional feature is input into the last linear layer to obtain the final multi-dimensional feature vector.

[0043] The present invention is further explained in detail below with reference to the examples, but the present invention is not limited thereto.

[0044] This embodiment intelligently screens the inversion results of M=16 data obtained by the underground time-lapse resistivity monitoring system of a mining area working face in one day. This embodiment sets 12772 data collections for each monitoring, and the intervals of adjacent data collections are 10m in the x direction, 10m in the y direction, and 20m in the z direction; there are 43 sampling points in the x direction (equivalent to 43 YZ slices), that is, N=43; the number of sampling points in the y direction is 27, and the number of sampling points in the z direction is 11;

[0045] The method of the present invention is used to screen M=16 monitoring data:

[0046] (1) The resistivity values ​​of the YZ slices corresponding to the N = 43 sampling points in the x direction corresponding to the m-th monitoring data are normalized (using the above example of first mean variance normalization and then maximum normalization) and visualized, and the resistivity plane contour lines are randomly selected, for example Figure 3 As shown; the visualization method of this embodiment is as follows:

[0047] (1.1) Map the normalized resistivity data from the interval [0,1] to the grayscale value interval [0,255], and round down the data to obtain the corresponding gray value result;

[0048] (1.2) Perform bicubic spline interpolation on the gray value result, and the interpolation target size (w, h) is:

[0049]

[0050] where y n 、z n are the number of sampling points in the y and z directions respectively, and Δy and Δz are the sampling intervals in the y and z directions respectively.

[0051] (1.3) The interpolated data is expanded to become a square, the size of the square is dsize = max(w, h), the size of the left and right sides that need to be padded is left = (dsize-w) / / 2, the size of the top and bottom sides that need to be padded is top = (dsize-h) / / 2, and the constant 0 is used to fill the padded area;

[0052] (1.4) Perform bicubic spline interpolation again on the square data after edge filling, and the interpolation target size is (224, 224);

[0053] (1.5) OpenCV is used to visualize the slice data with an interpolated size of (224, 224) and a value in the interval [0, 255] to obtain an 8-bit grayscale image of each slice; the grayscale image is then mapped to an RGB three-channel color image to obtain the corresponding resistivity plane contour map.

[0054] (2) The mth monitoring data image Img m The 43 images are input into the ResNet10 network trained with the ImageNet dataset, and 43 1024-dimensional feature vectors I are obtained. m = {I1, I2, I3, ... I 43}. And then (I1,I2), (I2,I3),…, (I 42 ,I 43 ) calculate the cosine similarity, get 42 cosine similarities, and then add them up to get the structural similarity S corresponding to the mth monitoring data m ; A total of 16 structural similarities were obtained: {41.58, 41.16, 41.23, 41.82, 41.67, 41.21, 41.72, 41.39, 41.51, 40.92, 41.76, 41.07, 40.74, 40.98, 41.80, 41.71};

[0055] The image classification network of this embodiment adopts Figure 1 The lightweight residual network ResNet10 network shown in the figure has a resistivity plane contour map as its input. Its structure is as follows: the first layer is a two-dimensional convolution layer convLayer with a convolution kernel of (5*5), followed by a maximum pooling layer Maxpool (the first intermediate feature map is output after convolution operation and maximum pooling layer processing); followed by a basic residual module composed of two (3*3) two-dimensional convolution layers conv Layer and a jump connection (the basic residual module outputs the second intermediate feature map); followed by a 1*1 residual module composed of two (3*3) two-dimensional convolution layers conv Layer and a (1*1) convolution (the residual module outputs the third intermediate feature map); finally, the convolution layer features are flattened and input into the linear layer FC, and the output is a 1024-dimensional logit. This embodiment uses the classification task to pre-train the lightweight residual network ResNet10 network on the ImageNet dataset. During pre-training, a linear layer is added after the last layer, the number of output categories is 1000, the pre-training learning rate is 0.01, and the learning rate is reduced by 10 times at 30, 60, and 90 epochs, and the final learning rate is 10 -5 , and trained for a total of 100 epochs.

[0056] (3) Img corresponding to the mth monitoring data m={img1, img2, img3, ...img 43 As a whole, it is input into the Transformer neural network to extract the features of the three-dimensional data volume; 16 128-dimensional feature vectors are obtained;

[0057] This embodiment adopts the Figure 3 The Transformer neural network shown extracts the 128-dimensional feature vector of each monitored atlas, and obtains a total of 16 128-dimensional feature vectors; Figure 3 The Transformer neural network structure shown in the figure includes two 3D convolutional layers (conv3D Layer) (each convolution kernel is 3*3*3), the two convolutional layers are connected to BatchNorm layer and nonlinear activation function ReLU respectively, followed by Flattening operation, a linear layer (the linear layer outputs 128-dimensional tokens), 4 independent YZ-Transformer encoders (each YZ-Transformer encoder outputs 64-dimensional features), an X-Transformer encoder with 128-dimensional output and a linear layer with 128-dimensional output. The process of extracting 128-dimensional feature vector is as follows:

[0058] Img m Reshape to obtain input data input_image(1,224,224,n) with n channels and 224 width and height. Use a three-dimensional convolution with a convolution kernel of (3*3*3) and a step size of (2*2*2) to perform sliding window convolution on input_image to obtain the feature representation of an image with a size of (128,128,n / 2). Then enter the second three-dimensional convolution layer to obtain features with a size of (64,64,n / 4). Flatten the obtained features into 1 vector and input them into the linear layer to obtain a 128-dimensional token.

[0059] The 128-dimensional tokens are divided into 4 groups, each with 32 dimensions. After adding a position code to each token, they are integrated into a new 64-dimensional feature vector, which then enters 4 independent YZ-Transformer encoders to obtain the purified 64-dimensional features.

[0060] The 4*64-dimensional features output by the four groups of YZ-Transformer encoders are concatenated to obtain a 256-dimensional YZ plane spatial feature; and then input into the X-Transformer encoder for feature integration to obtain a 128-dimensional feature;

[0061] The above 128-dimensional features are input into the last linear layer to obtain the final 128-dimensional feature vector.

[0062] (4) Sort the 16 structural similarities from large to small, and take the 128-dimensional feature vectors corresponding to the first 16*0.7=11 structural similarities {41.82, 41.80, 41.76, 41.72, 41.71, 41.67, 41.58, 41.51, 41.39, 41.23, 41.21} for clustering and divide them into two categories; and obtain the class center vectors of the two categories and the 11 cluster feature vectors of the 11 monitoring data {J} l1 ;

[0063] The total structural similarity of the two classes was calculated. The total structural similarity of the 7 samples in class 1 was 291.99, and the average structural similarity was 41.713; the total structural similarity of the 4 samples in class 2 was 165.41, and the average structural similarity was 41.353. Class 1 was selected as the class with higher data quality.

[0064] For the 7 samples in category 1, the Euclidean distance between their cluster feature vectors and the class center vector C1 is calculated respectively, and the distance d = {0.0766, 0.0772, 0.0769, 0.0772, 0.0771, 0.0767, 0.5379} is obtained. The monitoring data with the smallest Euclidean distance is taken as the final screening result.

[0065] In this embodiment, 16 monitoring data of the same day are screened 10 times using the method of the present invention, and the fourth monitoring result is the best, and each screening takes less than 30 seconds, indicating that the method is objective, stable, reliable and fast.

[0066] Similarly, for the data of this embodiment, manual screening (the rule of manual screening is generally to remove data with large differences relative to other data, retain data with high consistency, and combine the subjective professional experience of technicians), it takes 30 minutes, and multiple screenings often give completely different screening results, which are not repeatable, have no fixed evaluation criteria, and the screening results are too subjective.

Claims

1. A method for screening mine time-lapse resistivity monitoring data, wherein the mine time-lapse resistivity monitoring data includes data obtained from M monitorings of the same mine area, M≥2; wherein the mth monitoring performs several data collections, m=1, 2, 3, ..., M; Each data acquisition obtains the resistivity value at the spatial coordinate (x, y, z); it is characterized in that, Methods include: Step 1: extract the resistivity values ​​at different x spatial positions in the mth monitoring data, and normalize the resistivity values ​​at the same x spatial coordinate position. n All resistivity values ​​at the spatial coordinate position are normalized to obtain x = x n Normalized resistivity sequence R corresponding to the spatial coordinate position scale，n ; n=1,2,3,…,N,N≥2, and x n With x n+1 The x-direction spatial positions are adjacent; N normalized resistivity sequences corresponding to the m-th monitoring data are obtained; m=1,2,3,…,M; Step 2: Visualize the N normalized resistivity series corresponding to the mth monitoring data to obtain the corresponding atlas Img m ={img1,img2,img3,…img n ,…,img N }, img n For x=x n Resistivity plane contour map corresponding to the spatial coordinate position; n = 1, 2, 3, ..., N; m = 1, 2, 3, ..., M; Step 3: Use the image classification network model to obtain the atlas Img m The high-dimensional eigenvectors of the resistivity plane contour maps in ; find I n with I n+1 The cosine similarity of n with I n+1 Respectively, they are the resistivity plane contour map img n and img n+1 The high-dimensional feature vector of n = 1, 2, 3, ..., N-1; sum up the N-1 cosine similarities to obtain the structural similarity S corresponding to the m-th monitoring data m ; m = 1, 2, 3, ..., M; Step 4: Use the Transformer neural network to extract the atlas Img m The multidimensional feature vector F m ; m = 1, 2, 3, ..., M; Step 5, screening the M multi-dimensional feature vectors, including: Step 5.1, selecting a multidimensional feature vector according to the size of the structural similarity, and selecting the multidimensional feature vectors corresponding to the corresponding sub-monitorings with multiple structural similarities having the largest size; Step 5.2, using the K-means clustering (KMeans) algorithm to divide the selected multiple multidimensional feature vectors into two categories; calculating the cumulative sum of the structural similarities corresponding to each multidimensional feature vector in each category, and obtaining the sum of the structural similarities of each category; Step 5.3, select the category that meets the screening condition and execute step 5.5, otherwise execute step 5.4; the screening condition is that the category with the largest number of multidimensional feature vectors and the largest sum of structural similarities in the two categories is selected as the screening object; Step 5.4, remove the multidimensional feature vector with the smallest corresponding structural similarity among the multiple multidimensional feature vectors classified by the K-means clustering (KMeans) algorithm last time, and execute steps 5.2 and 5.3 on the remaining multidimensional feature vectors in sequence, until the category with the largest number of multidimensional feature vectors and the largest sum of structural similarities is selected according to the screening conditions of step 5.3 to execute step 5.5; Step 5.5, for the category selected according to step 5.3, calculate the Euclidean distance between each multidimensional feature vector in the category and the class center vector; select the monitoring data corresponding to the multidimensional feature vector with the smallest Euclidean distance as the screened data.

2. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The sampling space coordinate points of the M monitorings are the same or different.

3. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The step 1 of normalizing the resistivity values ​​at the same x spatial coordinate position includes: first performing mean-variance normalization on multiple resistivity values, and then performing maximum normalization on the mean-variance normalized values; x = x n All the normalized values ​​corresponding to the spatial coordinate positions constitute x=x n Normalized resistivity series R corresponding to the spatial coordinate position scale，n .

4. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: Step 2 of respectively visualizing the N normalized resistivity sequences corresponding to the m-th monitoring data includes: for any x=x n Normalized resistivity series R corresponding to the spatial coordinate position scale，n Gray value conversion, interpolation processing, edge expansion to adjust image size and colorization are performed in sequence.

5. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The image classification network model is a ResNet10 network trained using the ImageNet dataset.

6. The method for screening mine time-lapse resistivity monitoring data according to claim 5, characterized in that: The ResNet10 network includes a two-dimensional convolutional layer conv Layer, a maximum pooling layer Maxpoo, a basic residual module, a residual module and a linear layer FC which are connected in sequence, wherein the residual module and the linear layer FC are connected through a flattening operation.

7. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The high-dimensional feature vector is a 1024-dimensional feature vector.

8. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The Transformer neural network includes two three-dimensional convolutional layers conv3D Layer, a BatchNorm layer, a nonlinear activation function ReLU, a Flattening operation, a first linear layer Linear Layer, a YZ-Transformer encoder, an X-Transformer encoder and a second linear layer Linear Layer which are connected in sequence.

9. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The multi-dimensional feature vector is a 128-dimensional feature vector.

10. The method for screening mine time-lapse resistivity monitoring data according to claim 1, characterized in that: The step 5.1 includes: arranging the M structural similarities from large to small, and taking the multi-dimensional feature vectors corresponding to the corresponding monitoring of the first 0.6 to 0.8M structural similarities with large structure.

11. A storage medium, characterized in that: A computer program / instruction is stored thereon, wherein the computer program / instruction, when executed by a processor, implements the steps of the method described in any one of claims 1 to 10.

12. A software product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method described in any one of claims 1 to 10 are implemented.

Images