Top coal fracture fractal analysis method based on CT imaging

Abstract

The invention discloses a top coal fracture fractal analysis method based on CT imaging, and belongs to the field of fracture fractal detection.The method comprises the following steps that a three-dimensional digital model containing fracture information is obtained through three-dimensional reconstruction based on CT scanning data of a top coal sample; an image processing method is adopted to process the three-dimensional digital model, a fracture network is extracted, and topological parameters are calculated; processing the fracture network by adopting a fractal analysis method, and calculating to obtain global fractal dimension and local dimension statistical characteristics; statistical features of the topological parameters, the global fractal dimensions and the local dimensions are fused to form a multi-scale feature vector; and based on the multi-scale feature vector, processing through a machine learning model to obtain a top coal fracture classification result. According to the method, the limitation that a traditional method depends on a single index and is high in subjectivity is overcome, and objective and accurate recognition of the top coal fracture development grade is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fracture fractal detection, and particularly relates to a top coal fracture fractal analysis method based on CT imaging. BACKGROUND

[0002] The development of top coal fractures directly affects the mining efficiency of coal, the safety of mines, and the resource recovery rate. Accurate identification and classification of top coal fractures play an irreplaceable role in optimizing mining plans, preventing roof accidents, and improving resource recovery rates. However, traditional top coal fracture analysis methods have many limitations and cannot meet the high requirements of modern coal mining for fracture analysis accuracy and efficiency.

[0003] Traditional top coal fracture research mainly relies on field observation, drilling sampling, and some simple physical experiment methods. Field observation can only obtain limited and local fracture information, which is difficult to fully reflect the complex fracture network structure inside the top coal. Although drilling sampling can obtain rock core samples at a certain depth, the sampling process may damage the fracture structure, and the limited number of samples cannot guarantee the representativeness of the samples. In addition, the data obtained by these traditional methods are mostly two-dimensional, which cannot accurately reflect the distribution and connectivity characteristics of top coal fractures in three-dimensional space, and there are great difficulties in quantitative analysis and classification of fractures.

[0004] In terms of fracture analysis, traditional methods mainly focus on describing the macroscopic morphological characteristics of fractures, such as length, width, and inclination, and lack in-depth analysis of the overall structural characteristics of fracture networks. For quantitative analysis of fractures, some simple statistical methods such as fracture density calculation are usually used, which cannot fully consider the mutual relationship between fractures and the complexity of fracture networks. In terms of fracture classification, traditional methods are mostly based on experience and lack scientific classification criteria and effective classification algorithms, making it difficult to guarantee the accuracy and reliability of the classification results.

[0005] Therefore, the present application proposes a top coal fracture fractal analysis method based on CT imaging. SUMMARY

[0006] To solve the above technical problems, the present application proposes a top coal fracture fractal analysis method based on CT imaging to solve the problems existing in the prior art.

[0007] To achieve the above purpose, the present application provides a top coal fracture fractal analysis method based on CT imaging, which comprises: Based on the CT scanning data of the top coal sample, a three-dimensional digital model containing fracture information is obtained through three-dimensional reconstruction; An image processing method is used to process the three-dimensional digital model, extract the fracture network, and calculate the topological parameters; The fracture network is processed by using a fractal analysis method, and global fractal dimension and local dimension statistical characteristics are calculated; The topological parameters, global fractal dimension and local dimension statistical characteristics are fused to form a multi-scale feature vector; Based on the multi-scale feature vector, a machine learning model is used to obtain a top coal fracture classification result.

[0008] Optionally, the process of obtaining a three-dimensional digital model containing fracture information includes: Based on the CT scan data of the top coal sample, a high-resolution X-ray scanning system is used to obtain a two-dimensional tomographic image sequence; The two-dimensional tomographic image sequence is adaptively filtered and denoised to obtain a denoised image; The denoised image is processed by using a multi-threshold segmentation algorithm to obtain a segmented image; Based on a region growing algorithm, the segmented image is three-dimensionally reconstructed to generate a preliminary three-dimensional model; The preliminary three-dimensional model is subjected to anisotropic diffusion smoothing processing to obtain an optimized three-dimensional digital model.

[0009] Optionally, the process of processing the three-dimensional digital model by using an image processing method, extracting a fracture network and calculating topological parameters includes: The three-dimensional digital model is processed by using a three-dimensional segmentation network based on deep learning to obtain a binary fracture segmentation result; The segmentation result is subjected to skeleton extraction and medial axis transformation processing to obtain a simplified fracture network skeleton; The fracture network skeleton is abstracted into a topological graph structure based on graph theory, wherein a vertex represents a fracture intersection point or an end point, and an edge represents a fracture branch; Based on the topological graph structure, topological parameters including fracture rate, fracture density, average opening, orientation anisotropy index and connectivity coefficient are calculated.

[0010] Optionally, in the process of processing the three-dimensional digital model by using a three-dimensional segmentation network based on deep learning, a fracture segmentation neural network based on a 3D U-Net architecture is constructed to segment the fractures of the three-dimensional digital model; the fracture segmentation neural network is an encoder-decoder structure, the encoder includes a plurality of down-sampling modules, and the decoder includes a plurality of up-sampling modules; The fracture segmentation neural network based on the 3D U-Net architecture segments the fractures of the three-dimensional digital model, and the process includes: inputting the three-dimensional digital model containing fissure information into a down-sampling module of an encoder to obtain a low-resolution high-dimensional feature map; wherein a plurality of down-sampling modules are connected in series, and the three-dimensional digital model sequentially undergoes two 3*3*3 convolution operations every time it passes through a down-sampling module, and then a non-linear transformation is performed on the feature map after each convolution using a ReLU activation function, and then a 2*2*2 maximum pooling operation is performed to obtain a feature map with a size reduced by half; inputting the low-resolution high-dimensional feature map into an up-sampling module of a decoder to obtain a decoder output feature map; wherein a plurality of up-sampling modules are connected in series, and the low-resolution high-dimensional feature map sequentially undergoes a transposed convolution operation every time it passes through an up-sampling module, and the size of the low-resolution high-dimensional feature map is doubled to obtain a size-expanded feature map, and the size-expanded feature map is spliced with a feature map transmitted through a skip connection in an encoder path to obtain a spliced feature map; processing the decoder output feature map based on a 1*1*1 convolution layer to obtain a probability map with the same size as the input three-dimensional digital model.

[0011] Optionally, the process of obtaining the global fractal dimension and the local dimension statistical features comprises: calculating the global fractal dimension of the fissure network using a three-dimensional box counting method; traversing the three-dimensional model of the fissure network using a sliding window method, and calculating the local box dimension at each window position to form a local dimension distribution matrix; calculating statistical feature quantities based on the local dimension distribution matrix; combining the global fractal dimension and the statistical feature quantities to form a fractal feature vector.

[0012] Optionally, the statistical feature quantities include: mean, standard deviation, skewness and peak value.

[0013] Optionally, the process of fusing the topological parameters and the fractal feature vector to form a multi-scale feature vector comprises: performing standardization preprocessing on the topological parameter feature set and the fractal feature set respectively to obtain a preprocessed topological parameter feature set and a preprocessed fractal feature set; splicing the preprocessed topological parameter feature set and the preprocessed fractal feature set into an initial feature vector; based on the initial feature vector, calculating the weight coefficient of each feature using an attention mechanism and generating a weighted feature vector; constructing a feature interaction network and extracting feature interaction terms to obtain interaction features; splicing the weighted feature vector and the interaction features to form an enhanced feature vector; performing dimension reduction processing on the enhanced feature vector to generate a final multi-scale fusion feature vector.

[0014] Optionally, based on the multi-scale feature vector, the process of obtaining the top coal fracture classification result by processing through the machine learning model comprises: inputting the multi-scale feature vector into a pre-trained hybrid classification model for processing, the hybrid classification model comprising a gradient boosting decision tree module and a deep neural network module connected in parallel; weighting and fusing the output results of the gradient boosting decision tree module and the deep neural network module to generate a final probability distribution output; determining the development grade classification result of the top coal fracture based on the probability distribution output.

[0015] The application further provides a computer comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method when executing the computer program.

[0016] The application further provides a storage medium having a computer program stored thereon, wherein the program is executable by a processor to implement the method.

[0017] Compared with the prior art, the application has the following advantages and technical effects: The application realizes comprehensive quantitative characterization of the top coal fracture structure from macroscopic statistics to microscopic heterogeneity by fusing multi-scale fractal features and topological parameters, and significantly improves the accuracy and depth of fracture evaluation. The intelligent classification of the fused features by the machine learning model overcomes the limitations of the traditional method relying on a single index and strong subjectivity, and realizes objective and accurate identification of the development grade of the top coal fracture. The method provides a reliable quantitative basis for the top coal caving and drawing prediction, and has important application value for improving the efficiency of fully mechanized caving and the recovery rate of coal. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application and their explanations are used to explain this application and do not constitute an improper limitation on this application. In the drawings: Figure 1 The method flowchart of the embodiments of the application. DETAILED DESCRIPTION

[0019] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0021] As shown in the embodiment, a CT imaging-based top coal fracture fractal analysis method is provided, comprising the following steps: Figure 1 S1, based on the CT scanning data of the top coal sample, a three-dimensional digital model containing fracture information is obtained through three-dimensional reconstruction; the process includes: based on the CT scanning data of the top coal sample, a high-resolution X-ray scanning system is used to obtain a two-dimensional fault image sequence; the two-dimensional fault image sequence is subjected to adaptive filtering and noise reduction processing to obtain a denoised image; the denoised image is processed using a multi-threshold segmentation algorithm to obtain a segmented image; the segmented image is subjected to three-dimensional reconstruction based on a region growing algorithm to generate a preliminary three-dimensional model; the preliminary three-dimensional model is subjected to anisotropic diffusion smoothing processing to obtain an optimized three-dimensional digital model. S2, the three-dimensional digital model is processed using an image processing method to extract the fracture network and calculate the topological parameters; the process includes: the three-dimensional digital model is processed using a three-dimensional segmentation network based on deep learning to obtain a binary fracture segmentation result; the segmentation result is subjected to skeleton extraction and medial axis transformation processing to obtain a simplified fracture network skeleton; the fracture network skeleton is abstracted into a topological graph structure based on graph theory, wherein the vertices represent the fracture intersection points or end points, and the edges represent the fracture branches; the topological parameters including the fracture rate, the fracture density, the average opening, the orientation anisotropy index and the connectivity coefficient are calculated based on the topological graph structure.

[0022] Further, as a specific embodiment of the present embodiment, the three-dimensional digital model is processed using an image processing method to accurately extract the fracture network and quantitatively calculate its multi-dimensional topological parameters by combining deep learning and mathematical morphology algorithm; specifically including constructing a fracture segmentation network to realize voxel-level accurate extraction, and abstracting the fracture network into a node and edge relationship based on a graph theory model for topological parameter calculation.

[0023] S2.1 Precise fracture segmentation based on deep learning.

[0024]

[0025] ​A crack segmentation neural network based on a 3D U-Net architecture is constructed, which adopts an encoder-decoder structure. The encoder part is composed of four down-sampling modules, each of which contains two 3x3x3 convolution layers and a ReLU activation function, followed by a 2x2x2 max-pooling layer for down-sampling. The decoder part uses transposed convolution for up-sampling and fuses the feature maps of the corresponding scale of the encoder through a skip connection. The network finally outputs the probability prediction value of each voxel belonging to the crack through a 1x1x1 convolution layer and a Sigmoid activation function. The network training adopts a combined loss function: L = a·L dice + b·L focal , where L dice is the Dice loss function for handling the foreground-background imbalance problem, and the weight coefficient a is 0.6; L focal is the Focal loss function, which focuses on difficult classification samples, and the weight coefficient b is 0.4, and the modulation factor g is set to 2.0.

[0026] Further, in the process of processing a three-dimensional digital model using a three-dimensional segmentation network based on deep learning, a crack segmentation neural network based on a 3D U-Net architecture is constructed to segment cracks in the three-dimensional digital model. The crack segmentation neural network is an encoder-decoder structure, and the encoder includes a plurality of down-sampling modules, and the decoder includes a plurality of up-sampling modules.

[0027] The process of segmenting cracks in the three-dimensional digital model by the crack segmentation neural network based on the 3D U-Net architecture includes: inputting the three-dimensional digital model containing crack information into the down-sampling module of the encoder to obtain a low-resolution high-dimensional feature map; wherein the plurality of down-sampling modules are connected in series, and the three-dimensional digital model is sequentially subjected to two 3x3x3 convolution operations every time it passes through a down-sampling module, and after each convolution, a ReLU activation function is used for nonlinear transformation, and then a 2x2x2 max-pooling operation is performed to obtain a feature map with half the size; inputting the low-resolution high-dimensional feature map into the up-sampling module of the decoder to obtain a decoder output feature map; wherein the plurality of up-sampling modules are connected in series, and the low-resolution high-dimensional feature map is sequentially subjected to transposed convolution operation every time it passes through an up-sampling module, which enlarges the size of the low-resolution high-dimensional feature map by a factor of two to obtain a size-enlarged feature map, and the size-enlarged feature map is spliced with the feature map transmitted from the encoder path through the skip connection to obtain a spliced feature map; and the decoder output feature map is processed based on a 1x1x1 convolution layer to obtain a probability map with the same size as the input three-dimensional digital model.

[0028] S2.2 Crack network skeleton extraction and simplification.

[0029] The binarized segmentation result of the neural network output is subjected to a three-dimensional thinning algorithm for skeleton extraction, and through iteration, boundary voxels satisfying certain conditions are removed, and a single-pixel-width crack center line is retained. Then, the medial axis transform algorithm is applied to calculate the shortest distance of each point in the skeleton to the crack boundary, and the crack opening information w(x, y, z) = 2 d(x, y, z) is obtained, where d(x, y, z) represents the shortest distance of the skeleton point (x, y, z) to the boundary.

[0030] S2.3 Topological modeling based on graph theory.

[0031] The skeleton extraction result is modeled as an undirected graph G = (V, E, A), where the vertex set V represents the crack intersection points or end points, the edge set E represents the crack branches, and the attribute set A includes the geometric characteristics of each edge: length l i , average opening w i , and strike vector θ i ; the length l i is calculated by the Euclidean distance accumulation of the skeleton point sequence; the average opening w i is the arithmetic mean of the opening values of all skeleton points on the branch; and the strike vector θ i is obtained by fitting the spatial direction of the skeleton points through principal component analysis.

[0032] S2.4 Topological parameter calculation.

[0033] Based on the topological graph G, the following parameters are calculated: crack rate φ = V pores / V total × 100%, where V pores is the total volume of the crack, and V total is the total volume of the sample; crack density ρ = N / V total , where N is the total number of crack branches; average opening μ w = (Σw i ) / N; orientation anisotropy index η = λ max / λ min , where λ max and λ min are the maximum and minimum eigenvalues of the strike vector covariance matrix, respectively; connectivity coefficient λ = (S max / S total ) × C global , where S_max is the number of edges in the largest connected component, S total is the total number of edges, and C global is the global clustering coefficient, reflecting the clustering degree of network connection.

[0034] S3, the fractal analysis method is used for processing the fracture network, and global fractal dimension and local dimension statistical characteristics are calculated. The process of obtaining the global fractal dimension and the local dimension statistical characteristics includes: the three-dimensional box counting method is used to calculate the global fractal dimension of the fracture network; the sliding window method is used to traverse the three-dimensional model of the fracture network, the local box dimension is calculated at each window position, and a local dimension distribution matrix is formed; statistical characteristic quantities are calculated based on the local dimension distribution matrix; and the global fractal dimension and the statistical characteristic quantities are combined to form a fractal feature vector.

[0035] Further, as a specific embodiment of the present embodiment, the fractal analysis method is used for processing the fracture network, the global fractal dimension representing the overall complexity of the fracture system and the local dimension statistical characteristics reflecting the spatial heterogeneity are calculated by introducing the multifractal theory and the adaptive sliding window technology; specifically, the global dimension is calculated based on the improved three-dimensional box counting method, the local dimension distribution is calculated by constructing a scale-adaptive sliding window matrix, and then the multifractal spectrum characteristic parameters are extracted, and finally a high-dimensional fractal feature vector is formed by fusion.

[0036] S3.1 Improved three-dimensional box counting method for global dimension calculation.

[0037] The weighted box counting method considering the anisotropy of the fracture direction is used to calculate the global fractal dimension D global . A cubic grid with a side length of l i is used to cover the binary fracture model, where l i =2 k (k=1,2,…,n); for each scale l i , the number of boxes containing fractures N(l i ) is counted, and the fracture direction entropy value of each box is calculated: ; where p j represents the proportion of the fracture voxel in the jth direction interval in the box, and M is the total number of direction intervals. The weighted box number N w (l i )=∑(W i ⋅N i ), and D global is obtained by linear fitting of the slope.

[0038] S3.2 Scale-adaptive sliding window local dimension calculation.

[0039] The scale-adaptive sliding window algorithm is constructed. First, the optimal window size is calculated according to the fracture density distribution: ; ​Where ρ(x,y,z) is the local fracture density. A variable-size sliding window (size range L) is used. opt (±ΔL) Traverse the 3D model and calculate the local box dimension D at each window position. local (x,y,z) generates a three-dimensional local dimension distribution matrix D. map The missing areas are filled using Kriging interpolation.

[0040] S3.3 Multifractal Spectrum Feature Extraction.

[0041] Based on the local dimension distribution matrix D map Calculate the multifractal spectrum. First, calculate the partition function: .in, is the weighting factor, and q is the order (a sequence of integers from -10 to 10). Through linear fitting: Obtain the quality index Finally, the multifractal spectrum is obtained through the Legendre transform.

[0042] S3.4 Construction of fractal eigenvectors.

[0043] The global fractal dimension D global With multifractal spectrum characteristic parameters Δα, B, f max and local dimensional statistical features (mean μ) d Standard deviation σ d skewness S d kurtosis K d The combination of these elements forms a nine-dimensional fractal eigenvector: .

[0044] S4. The topological parameters, global fractal dimension, and local dimensionality statistical features are fused to form a multi-scale feature vector. The process of forming the multi-scale feature vector includes: standardizing and preprocessing the topological parameter feature set and the fractal feature set to obtain preprocessed topological parameter feature set and preprocessed fractal feature set, respectively; concatenating the preprocessed topological parameter feature set and preprocessed fractal feature set to form an initial feature vector; based on the initial feature vector, using an attention mechanism to calculate the weight coefficient of each feature and generate a weighted feature vector; constructing a feature interaction network and extracting feature interaction terms to obtain interaction features; concatenating the weighted feature vector with the interaction features to form an enhanced feature vector; and performing dimensionality reduction on the enhanced feature vector to generate the final multi-scale fused feature vector.

[0045] Further, as a specific embodiment of the present embodiment, the topological parameters, global fractal dimension and local dimension statistical features are fused to form a multi-scale feature vector, and through the construction of a multi-modal feature fusion network based on a gating mechanism and a structured interaction modeling, adaptive weighting and deep interaction of different physical meanings and scale features are realized; specifically including using a modal-specific standardization method, feature weighting based on a gating attention mechanism, high-order interaction term extraction and manifold learning dimension reduction, and finally generating a low-dimensional fusion feature vector with stronger representation ability.

[0046] S4.1 Modal adaptive standardization processing. Modal-specific standardization processing is performed on the topological parameter feature set and the fractal feature set. The topological parameter feature set includes five parameters: fissure ratio, fissure density, average opening, orientation anisotropy index and connectivity coefficient; the fractal feature set includes eight parameters: global fractal dimension, multi-fractal spectrum width, spectrum symmetry parameter, maximum dimension, and mean, standard deviation, skewness and kurtosis of local dimension. The mean and standard deviation of each modal are used for standardization processing to eliminate the dimension effect.

[0047] Let the topological parameter feature set be T and the fractal feature set be F, and use the modal-specific standardization method: . Wherein, , and , are the mean and standard deviation of the topological parameters and fractal features on the training set, respectively.

[0048] S4.2 Feature weighting based on gating attention. The gating attention mechanism is used to calculate the feature importance weight. A gating network containing a learnable weight matrix and a bias vector is constructed, and the attention score of each feature is calculated through a sigmoid activation function. After softmax normalization, the weight coefficient is obtained. These weight coefficients are used to weight the standardized features to generate a weighted feature vector.

[0049] S4.3 Structured feature interaction modeling. The decomposition machine model is used to extract the high-order interaction between features, and the inner product of the hidden vectors between each pair of features is calculated as the interaction strength. At the same time, based on the correlation coefficient and mutual information, the significant interaction feature items are screened, and the interaction items with a correlation coefficient greater than zero point three and a mutual information greater than zero point two are retained. The high-order interaction features output by the decomposition machine and the significant interaction items screened are spliced into the weighted feature vector to form an enhanced feature vector.

[0050] S4.4 Manifold learning dimension reduction. Isometric mapping manifold learning algorithm is adopted for nonlinear dimension reduction. The original manifold structure is maintained by calculating the geodesic distance between feature vectors, the final dimension is determined according to the cumulative contribution rate of the characteristic value, the principal components with a cumulative contribution rate of more than 95% are retained, and the final multi-scale fusion feature vector is generated. The feature vector not only retains the main information of the original feature, but also enhances the representation ability of the feature.

[0051] S5, based on the multi-scale feature vector, the top coal fracture classification result is obtained by processing through the machine learning model. The process of obtaining the top coal fracture classification result by processing through the machine learning model includes: inputting the multi-scale feature vector into the pre-trained hybrid classification model for processing, the hybrid classification model includes parallel gradient boosting decision tree module and deep neural network module; the output results of the gradient boosting decision tree module and the deep neural network module are weighted and fused to generate the final probability distribution output; and the development level classification result of the top coal fracture is determined based on the probability distribution output.

[0052] Further, as a specific embodiment of the present embodiment, based on the multi-scale feature vector, the accurate classification result of the development level of the top coal fracture is obtained by processing through the integrated multi-modal machine learning model and the adaptive decision fusion mechanism; specifically, a hybrid architecture of gradient boosting decision tree and deep neural network is adopted, a dynamic weighted fusion strategy based on confidence is combined, and transfer learning and uncertainty quantification are introduced, to construct an intelligent classification system that can fully mine feature information and ensure classification reliability.

[0053] First, the hybrid classification model architecture is constructed. The model is composed of gradient boosting decision tree module and deep neural network module in parallel. The gradient boosting decision tree module adopts multiple regression trees for ensemble learning, iteratively optimizes the prediction results through gradient boosting algorithm, sets the depth of the tree to 6-10 layers, the learning rate to 0.05, and the number of sub-trees to 100. The deep neural network module adopts a three-layer fully connected network structure, the input layer dimension is the same as that of the multi-scale feature vector, the dimensions of the two hidden layers are 64 and 32 respectively, the output layer dimension is the same as the number of crack categories, the ReLU activation function is used, and the Dropout regularization with a dropout rate of 0.2 is used to prevent overfitting.

[0054] Then the dynamic weighted fusion strategy is implemented. The outputs of the two modules are fused through a dynamic weighting method based on confidence. First, the probability distribution of the outputs of the gradient boosting decision tree module and the deep neural network module is calculated respectively, and then the weights are dynamically allocated according to the classification confidence of each module on the validation set. The final fusion result is the weighted sum of the outputs of the two modules, and the weights are dynamically adjusted according to the classification accuracy of the last k samples of each module, the initial weight is set to 0.5, and the sum of the two weights is ensured to be 1.

[0055] Then the transfer learning and model optimization methods are adopted. First, the fracture data of other geological conditions are used for pre-training, and then the data of the target mine area are used for fine-tuning to improve the model generalization ability. The training process adopts a phased strategy: first, fix the gradient to improve the decision tree parameter training deep neural network module, use the Adam optimizer, and set the learning rate to 0.001; then, jointly train the entire hybrid model, and use the cosine annealing algorithm to dynamically adjust the learning rate. The model loss function is the cross-entropy loss plus the L2 regularization term.

[0056] Subsequently, uncertainty quantification and classification decision are carried out. Based on the fused probability distribution, the classification uncertainty is calculated, the prediction confidence is measured by entropy value, and the threshold is set to 0.3. When the entropy value exceeds the threshold, the expert review mechanism is started to ensure the reliability of classification. The final classification result is determined according to the maximum probability principle, and the corresponding confidence score is output.

[0057] Finally, the classification result explanation and visualization are provided. Based on the SHAP value, the contribution of each feature to the classification result is calculated, and the feature importance map is generated. At the same time, the t-SNE algorithm is used to reduce the high-dimensional features to two-dimensional or three-dimensional space, realizing the visualization display of the classification result.

[0058] The embodiment also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method when executing the computer program.

[0059] The embodiment also provides a storage medium, which stores a computer program, and the program implements the method when executed by a processor.

[0060] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A CT imaging-based top coal fracture fractal analysis method, characterized in that, The method comprises the following steps: Based on the CT scanning data of the top coal sample, a three-dimensional digital model containing fracture information is obtained through three-dimensional reconstruction; An image processing method is used to process the three-dimensional digital model, extract the fracture network and calculate the topological parameters; A fractal analysis method is used to process the fracture network, and global fractal dimension and local dimension statistical characteristics are calculated; The topological parameters, global fractal dimension and local dimension statistical characteristics are fused to form a multi-scale feature vector; Based on the multi-scale feature vector, a machine learning model is used to process to obtain a top coal fracture classification result.

2. The CT imaging based top coal fracture fractal analysis method according to claim 1, characterized in that, The process of obtaining the three-dimensional digital model containing fracture information comprises: Based on the CT scanning data of the top coal sample, a two-dimensional fault image sequence is obtained by using a high-resolution X-ray scanning system; The two-dimensional fault image sequence is subjected to adaptive filtering and denoising processing to obtain a denoised image; A multi-threshold segmentation algorithm is used to process the denoised image to obtain a segmented image; Based on a region growing algorithm, the segmented image is subjected to three-dimensional reconstruction to generate a preliminary three-dimensional model; The preliminary three-dimensional model is subjected to anisotropic diffusion smoothing processing to obtain an optimized three-dimensional digital model.

3. The CT imaging based top coal fracture fractal analysis method according to claim 1, characterized in that, The process of using an image processing method to process the three-dimensional digital model, extract the fracture network and calculate the topological parameters comprises: A three-dimensional segmentation network based on deep learning is used to process the three-dimensional digital model to obtain a binary fracture segmentation result; The segmentation result is subjected to skeleton extraction and medial axis transformation processing to obtain a simplified fracture network skeleton; Based on a graph theory method, the fracture network skeleton is abstracted into a topological graph structure, wherein the vertices represent fracture intersection points or end points, and the edges represent fracture branches; Based on the topological graph structure, topological parameters including fracture rate, fracture density, average opening, orientation anisotropy index and connectivity coefficient are calculated.

4. The CT imaging based top coal fracture fractal analysis method according to claim 3, characterized in that, In the process of using a three-dimensional segmentation network based on deep learning to process the three-dimensional digital model, a fracture segmentation neural network based on a 3D U-Net architecture is constructed to segment the three-dimensional digital model; the fracture segmentation neural network is an encoder-decoder structure, the encoder comprises a plurality of down-sampling modules, and the decoder comprises a plurality of up-sampling modules; The process of using the fracture segmentation neural network based on the 3D U-Net architecture to segment the three-dimensional digital model comprises: The three-dimensional digital model containing fracture information is input into the down-sampling module of the encoder to obtain a low-resolution high-dimensional feature map; a plurality of down-sampling modules are connected in series, and the three-dimensional digital model passes through each down-sampling module twice to perform 3*3*3 convolution operation, then the ReLU activation function is used for nonlinear transformation, and then 2*2*2 maximum pooling operation is performed to obtain a feature map with a size reduced by half; Input the low-resolution high-dimensional feature map into an up-sampling module of a decoder to obtain a decoder output feature map; wherein a plurality of up-sampling modules are connected in series, and the low-resolution high-dimensional feature map is sequentially subjected to a transposed convolution operation every time it passes through an up-sampling module to enlarge the size of the low-resolution high-dimensional feature map by one time to obtain a size-enlarged feature map, and the size-enlarged feature map is spliced with a feature map transmitted through a skip connection in an encoder path to obtain a spliced feature map; Process the decoder output feature map based on a 1x1x1 convolution layer to obtain a probability map with the same size as the input three-dimensional digital model.

5. The CT imaging based top coal fracturing fractal analysis method according to claim 1, characterized in that, The process of obtaining the global fractal dimension and the local dimension statistical features includes: Adopting a three-dimensional box counting method to calculate the global fractal dimension of the fracture network; Adopting a sliding window method to traverse the three-dimensional model of the fracture network, and calculating the local box dimension at each window position to form a local dimension distribution matrix; Based on the local dimension distribution matrix, calculating statistical feature quantities; Combining the global fractal dimension and the statistical feature quantities to form a fractal feature vector.

6. The CT imaging based top coal fracture fractal analysis method according to claim 5, characterized in that, The statistical feature quantities include: mean, standard deviation, skewness and peak value.

7. The CT imaging based top coal fracturing fractal analysis method according to claim 5, characterized in that, The process of fusing the topological parameters and the fractal feature vector to form a multi-scale feature vector includes: Respectively performing standardization preprocessing on the topological parameter feature set and the fractal feature set to obtain a preprocessed topological parameter feature set and a preprocessed fractal feature set; Splicing the preprocessed topological parameter feature set and the preprocessed fractal feature set into an initial feature vector; Based on the initial feature vector, adopting an attention mechanism to calculate the weight coefficient of each feature and generating a weighted feature vector; Constructing a feature interaction network and extracting feature interaction terms to obtain interaction features; Splicing the weighted feature vector and the interaction features to form an enhanced feature vector; Performing dimension reduction processing on the enhanced feature vector to generate a final multi-scale fusion feature vector.

8. The CT imaging based top coal fracturing fractal analysis method according to claim 1, characterized in that, Based on the multi-scale feature vector, processing by a machine learning model to obtain a top coal fracture classification result includes: Inputting the multi-scale feature vector into a pre-trained hybrid classification model for processing, wherein the hybrid classification model includes a gradient boosting decision tree module and a deep neural network module connected in parallel; Weighted fusion of the output results of the gradient boosting decision tree module and the deep neural network module to generate a final probability distribution output; Based on the probability distribution output, determining the development grade classification result of the top coal fracture.

9. A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the method of claim 1.

10. A storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the method of claim 1.

Images