Coal rock scanning electron microscope image pore type classification method and classification model

Abstract

The invention discloses a coal rock scanning electron microscope image pore type classification method and classification model. The method comprises the following steps: training a generative adversarial network by using a first sample set and adopting a dual manifold optimization mechanism to obtain an image generation model; image expansion is carried out on the second sample set by using an image generation model, a third sample set is established, and the first sample set and the second sample set both comprise scanning electron microscope images of coal rock primary pores, metamorphic pores, mineral pores and strain pores; based on the third sample set, adopting a dynamic boundary smoothing mechanism to train an extreme learning machine to obtain a classification model; and determining the pore type of the coal rock scanning electron microscope image based on the classification model. According to the method, training data of an adversarial network expansion classification model is introduced, the problems of insufficient training data and poor model generalization ability are solved, and a dual manifold optimization mechanism maintains a topological structure of image data; and the extreme learning machine of the dynamic boundary smoothing mechanism is adopted to perform pore type identification, so that the classification accuracy is further improved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a coal rock scanning electron microscope image pore type classification method and classification model. Background Art

[0002] The identification of different genetic types of pores in coal-rock gas reservoirs is of great significance. At present, many scholars tend to classify pore types by pore size. Although this method can provide a certain basis for classification, it ignores the structural characteristics of pore formation and is difficult to reveal the geological information of the coal-forming environment and its evolutionary history. The genetic type of pores in coal reservoirs is not only closely related to the coal-forming environment, but also directly affects the storage and migration performance of coalbed methane. However, due to the complexity and diversity of pore types, there is still a technical bottleneck in the efficient and automatic identification of pores of different genesis. If the genetic type of pores can be accurately classified through advanced automatic identification technology, it can not only improve the identification efficiency, but also provide a more comprehensive and accurate geological basis for the exploration and evaluation of coal-rock gas reservoirs, thereby significantly improving the scientificity and economy of reservoir modeling and gas reservoir development. Therefore, accurate identification of pore types and characteristics is of great significance for the development of coalbed methane and reservoir reconstruction. In traditional coal reservoir pore analysis, scanning electron microscope images, as a high-resolution characterization method, can provide detailed pore morphological characteristics. However, due to the wide variety of pore types in coal reservoirs, the huge amount of SEM image data and the difficulty in labeling, the workload of manually labeling pore types is extremely large and prone to errors. Therefore, how to efficiently and accurately automatically identify pore types has become a technical problem that needs to be solved in this field.

[0003] The Chinese invention patent with publication number CN119021675A proposes a method for automatically identifying hidden karst pores based on borehole television photography. It takes the results of borehole high-definition television photography as the main body, combines multi-parameter hydrological logging curves, core and logging data intersection diagrams, and uses image processing and computer vision technology that integrates multiple algorithms such as edge detection and corner detection to extract key features in the image (three-dimensional spatial features such as the width, height, and distribution of caves or cracks). This method can perform statistical analysis on the karst development in different directions at different depths of the borehole, significantly improving the accuracy and work efficiency of karst development feature characterization and statistics.

[0004] The Chinese invention patent with the publication number CN118097220A proposes a method and system for identifying voids in a rock particle accumulation system, which relates to the field of image recognition technology, and includes: using CT to scan the rock particle accumulation system to obtain an original slice image and performing preprocessing; selecting different data parameters such as quantity, size, and content to prepare four types of data sets for training a neural network; building an Attention R2U-net and training the network with the data sets to obtain a model; qualitatively and quantitatively evaluating the recognition effect of the model under the four types of data sets based on the label image and the recognition image, and further determining the optimal data set parameters by comparing the binary classification confusion matrix indicators; finally, preparing an optimal data set with reference to the optimal data set parameters and training the neural network with this data set, repeatedly training the neural network and adjusting the hyperparameters to obtain an optimal model, and testing the convergence and accuracy of the model through qualitative and quantitative methods. This method improves the efficiency and accuracy of the training and recognition processes of the neural network in the field of void recognition in the rock particle accumulation system. Summary of the Invention

[0005] The inventors found in their work that there are many deficiencies in the prior art in the identification of pore types in coal-rock gas reservoirs and image type automatic recognition technology:

[0006] 1. The main problem in the prior art in the field of pore identification in coal-rock gas reservoirs is the lack of an efficient automatic identification technology for multi-category genetic pores. Current research mainly focuses on pore classification based on pore size or morphology, but these methods cannot accurately reveal the genetic characteristics of pores and are difficult to provide key information related to the coal-forming environment and reservoir gas migration ability. In addition, traditional identification methods rely on manual annotation or simple algorithm processing, which is not only inefficient but also easily affected by subjective biases.

[0007] 2. Traditional image data augmentation methods usually cannot fully preserve the topological structure of the image, and the generated images may lack diversity and cannot effectively cover various morphologies and sizes of pores, resulting in over-reliance on limited original data during model training and prone to overfitting or underfitting.

[0008] 3. Existing dimensionality reduction methods often fail to fully consider the local structure and boundary effects of the data, and may lose important features or introduce outliers during the dimensionality reduction process, resulting in unstable features after dimensionality reduction and affecting subsequent classification effects.

[0009] 4. Traditional classification algorithms may not be able to effectively adjust the decision boundary between classes when dealing with high-dimensional and imbalanced data sets, resulting in low recognition accuracy for some classes, and the classification performance is easily affected in the face of uneven data distribution.

[0010] To at least partially solve the above problems, an embodiment of the present invention provides a method and a classification model for classifying pore types in coal-rock scanning electron microscope images. The method uses a generative adversarial network based on dual manifold optimization to augment training data, solving the problems of insufficient data and poor model generalization ability. At the same time, a extreme learning machine classification algorithm based on a dynamic boundary smoothing mechanism is used to identify pore types, further improving the model's processing ability and classification accuracy for complex geological data.

[0011] In a first aspect, an embodiment of the present invention provides a method for classifying pore types in coal-rock scanning electron microscope images, including:

[0012] Determining the pore type of a coal-rock scanning electron microscope image based on a classification model, where the classification model is established through the following steps:

[0013] Using a first sample set, training a generator and a discriminator of a generative adversarial network using a dual manifold optimization mechanism to obtain an image generation model;

[0014] Obtaining a second sample set, using the image generation model for image augmentation, and establishing a third sample set. Both the first sample set and the second sample set include scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in a coal-rock reservoir;

[0015] Based on the third sample set, training an extreme learning machine using a dynamic boundary smoothing mechanism to obtain a classification model.

[0016] Optionally, the step of using a first sample set to train a generator and a discriminator of a generative adversarial network using a dual manifold optimization mechanism includes:

[0017] Initializing the weight parameters of the generator and the discriminator of the generative adversarial network respectively;

[0018] Inputting a random noise vector into the generator, inputting the image generated by the generator and the first sample set into the discriminator, and obtaining a discrimination result of the authenticity of the input image;

[0019] Using a dual manifold optimization mechanism to determine the objective function of the generator, determining the generator loss function by maximizing the misclassification probability of the discriminator and minimizing the objective function, and determining the discriminator loss function by maximizing the sum of the correct classification of real images and the misclassification of generated images;

[0020] Updating the weight parameters of the generator based on the gradient of the generator loss function, and updating the weight parameters of the discriminator based on the gradient of the discriminator loss function;

[0021] If the iteration termination condition is not satisfied, return to execute the step of inputting a random noise vector into the generator.

[0022] Optionally, determining the objective function of the generator by using the dual manifold optimization mechanism includes:

[0023] Determine the difference between the random noise vector and the generated image, and take the square of the L2 norm of this difference as the first objective term;

[0024] Take the square of the F norm of the product of this difference and the gradient of the random noise vector, multiplied by the coefficient controlling the importance of the high-order derivative term, as the second objective term;

[0025] Take the result of multiplying the sum of the first objective term and the second objective term by the weight adjustment coefficient of the dual manifold as the objective function of the generator.

[0026] Optionally, training the extreme learning machine by using the dynamic boundary smoothing mechanism based on the third sample set includes:

[0027] Input the third sample set into the extreme learning machine, and initialize the weight parameters and biases of the extreme learning machine;

[0028] Determine the average within-class variance and the average between-class distance according to the initialized classification result to obtain the initial smoothing parameter;

[0029] Determine the boundary smoothing function according to the current smoothing parameter and the classification result, and obtain a new classification result from the boundary smoothing function, the weight parameters and biases of the extreme learning machine;

[0030] Determine the optimization objective function. If the optimization objective function does not meet the iteration termination condition, update the weight parameters and biases through backpropagation, update the smoothing parameter based on the current iteration number, and return to execute determining the boundary smoothing function according to the current smoothing parameter and the classification result.

[0031] Optionally, obtaining the initial smoothing parameter includes:

[0032] Determine the ratio of the average within-class variance to the average between-class distance, and take the product of this ratio and the smoothing adjustment factor as the initial smoothing parameter.

[0033] Optionally, determining the boundary smoothing function according to the current smoothing parameter and the classification result includes:

[0034] Determine the boundary smoothing function according to the current smoothing parameter and the classification result through the following formula:

[0035]

[0036] In the formula, is the boundary smoothing function, is the input data of the extreme learning machine, is the smoothing parameter after the th iteration, The mean vector of the class, is the mean vector of the class, is the number of classes, is the L2 norm.

[0037] Optionally, the determining of the optimization objective function includes:

[0038]

[0039]

[0040] In the formula, is the optimization objective function, is the output data of the extreme learning machine, is the target output of the extreme learning machine, is the regularization parameter of the extreme learning machine, is the weight parameter of the extreme learning machine, is the multi-objective optimization term, and are the weight factors of the within-class variance and the between-class distance respectively, is the variance of the class, is the mean vector of the class, is the mean vector of the class, is the number of classes, is the L2 norm.

[0041] Optionally, the updating of the smoothing parameter based on the current iteration number includes:

[0042] Updating the smoothing parameter based on the current iteration number using the following formula:

[0043]

[0044] In the formula, is the smoothing parameter after the th iteration, is the initial smoothing parameter,

[0045] Optionally, after establishing the third sample set, it further includes:

[0046] Training an autoencoder based on boundary smoothing using the third sample set to obtain a feature extraction model; correspondingly,

[0047] The training of the extreme learning machine using the dynamic boundary smoothing mechanism based on the third sample set includes:

[0048] The fourth sample set is obtained from the output data of the feature extraction model, and an extreme learning machine is trained using a dynamic boundary smoothing mechanism.

[0049] Optionally, the autoencoder includes an encoder and a decoder. The encoder is used to perform boundary smoothing processing on the output data of the decoder through the following smoothing function:

[0050]

[0051] where is the smoothing function, is the output data of the decoder, is the smoothing intensity parameter; is the smoothing degree threshold parameter.

[0052] Optionally, the smoothing intensity parameter is determined by an adaptive method based on the local density of the data.

[0053] Optionally, training the boundary smoothing-based autoencoder using the third sample set includes:

[0054] Inputting the third sample set into the autoencoder, and respectively initializing the weight parameters and biases of the encoder and decoder of the autoencoder;

[0055] Based on the current weight parameters and biases of the encoder, the output data of the encoder is obtained through the activation function. Based on the current weight parameters and biases of the decoder, the output data of the decoder is obtained through the activation function. After the encoder performs boundary smoothing processing on the output data of the decoder, the final smoothed output is obtained;

[0056] Determining the autoencoder loss function based on the final smoothed output and the weight coefficients of the autoencoder;

[0057] Updating the weight parameters and biases of the encoder and decoder through the backpropagation of the loss function;

[0058] If the iteration termination condition is not satisfied, return to execute obtaining the output data of the encoder through the activation function based on the current weight parameters and biases of the encoder.

[0059] Optionally, the original pore scanning electron microscope images include the scanning electron microscope images of intercellular pores and inter-chip pores; the metamorphic pore scanning electron microscope images include the scanning electron microscope images of gas pores, shrinkage pores, and inter-chain pores; the mineral pore scanning electron microscope images include the scanning electron microscope images of dissolution pores, intercrystalline pores, and mold pores.

[0060] In a second aspect, an embodiment of the present invention provides a method for establishing a pore type classification model for coal-rock scanning electron microscope images, including:

[0061] Using the first sample set, the generator and discriminator of the generative adversarial network are trained by adopting a dual manifold optimization mechanism to obtain an image generation model;

[0062] Obtain a second sample set, use the image generation model for image augmentation, and establish a third sample set. Both the first sample set and the second sample set contain scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in coal-rock reservoirs;

[0063] Based on the third sample set, an extreme learning machine is trained by adopting a dynamic boundary smoothing mechanism to obtain a classification model for determining the pore type of the input coal-rock scanning electron microscope image.

[0064] In a third aspect, an embodiment of the present invention provides a classification model for the pore type of a coal-rock scanning electron microscope image, which is used to determine the pore type of the input coal-rock scanning electron microscope image. The classification model is established through the following steps:

[0065] Using the first sample set, the generator and discriminator of the generative adversarial network are trained by adopting a dual manifold optimization mechanism to obtain an image generation model;

[0066] Obtain a second sample set, use the image generation model for image augmentation, and establish a third sample set. Both the first sample set and the second sample set contain scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in coal-rock reservoirs;

[0067] Based on the third sample set, an extreme learning machine is trained by adopting a dynamic boundary smoothing mechanism to obtain a classification model.

[0068] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, any of the above methods is implemented.

[0069] In a fifth aspect, an embodiment of the present disclosure provides a server, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, any of the above methods is implemented.

[0070] The beneficial effects of the above technical solutions provided by the embodiments of the present invention at least include:

[0071] (1) The classification method for the pore type of the coal-rock scanning electron microscope image provided by the embodiment of the present invention greatly reduces the time and error required for manual analysis through an automated classification method, improves the working efficiency of deep coal-rock reservoir exploration, and provides more scientific and efficient technical support for coal-rock reservoir exploration and the analysis of reservoir gas migration characteristics.

[0072] (2) The pore type classification method for coal-rock scanning electron microscope images provided by the embodiments of the present invention uses a generative adversarial network optimized by dual manifolds to augment the samples of coal-rock reservoir scanning electron microscope images. Through the dual manifold optimization mechanism, the topological structure of the image data is maintained during the generation process, making the generated pore images more realistic in visual effects, and the generated data is closer to the original data in distribution. The dual manifold optimization adjusts the training processes of the generator and discriminator, making the generated images more diverse and consistent in terms of category distribution and morphology.

[0073] (3) The pore type classification method for coal-rock scanning electron microscope images provided by the embodiments of the present invention uses an extreme learning machine classification algorithm based on a dynamic boundary smoothing mechanism to identify pore categories. By dynamically adjusting the classification boundary, it solves the sensitivity of traditional classification methods to uneven distribution between categories when dealing with imbalanced data, can adaptively adjust the classification boundary according to the characteristics of the sample distribution, effectively avoids the imbalance problem of the inter-class boundary, and thus improves the accuracy of pore category recognition.

[0074] (4) The pore type classification method for coal-rock scanning electron microscope images provided by the embodiments of the present invention uses an autoencoder based on boundary smoothing to extract pore features. The autoencoder compresses high-dimensional data into a low-dimensional space through the structures of the encoder and decoder, and avoids outliers and boundary effects during the dimensionality reduction process through a boundary smoothing strategy. By adaptively adjusting the smoothing intensity, it enhances the adaptability of the model to the local structure of the data and improves the stability and accuracy of feature extraction.

[0075] (5) The pore type classification method for coal-rock scanning electron microscope images provided by the embodiments of the present invention can not only accurately distinguish various genetic type pores such as primary pores, metamorphic pores, mineral pores, and strain pores, but also further realize the identification of 9 sub-type pores of the genetic type, providing a data basis for the correlation analysis between pore types and coal-forming environments.

[0076] Other features and advantages of the present invention will be described in the following specification, and, in part, will become apparent from the specification or be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification, claims, and drawings.

[0077] The technical solution of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0078] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0079] Figure 1 This is the flowchart of the pore type classification method for coal rock scanning electron microscope images in the first embodiment of the present invention;

[0080] Figure 2 It is Figure 1 the specific implementation flowchart of step S11 in;

[0081] Figure 3 It is Figure 1 the specific implementation flowchart of step S13 in;

[0082] Figure 4 This is the training flowchart of the autoencoder in the first embodiment of the present invention;

[0083] Figures 5a - 5i This is the typical pore type diagram of coal rock scanning electron microscope images in the second embodiment of the present invention. Detailed implementation manners

[0084] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0085] It should be understood that the terms used in the present invention are only for describing specific embodiments and are not used to limit the present invention. Additionally, for the numerical ranges in the present invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Each intermediate value within any stated value or stated range, as well as each smaller range between any other stated value or intermediate value within the stated range, is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded from the range.

[0086] Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention pertains. Although the present invention only describes preferred methods and materials, any methods and materials similar or equivalent to those described herein can also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In case of conflict with any incorporated document, the content of this specification shall prevail.

[0087] Embodiment 1

[0088] The first embodiment of the present invention provides a pore type classification method for coal rock scanning electron microscope images, and its process is referred to Figure 1 as shown, and includes the following steps:

[0089] Step S11: Using the first sample set, train the generator and discriminator of the generative adversarial network with the dual manifold optimization mechanism to obtain an image generation model.

[0090] Use the generative adversarial network with dual manifold optimization for sample augmentation of coal-rock reservoir scanning electron microscope images. Through the dual manifold optimization mechanism, the topological structure of the image data is maintained during the generation process, making the generated pore images more realistic in visual effect, and the generated data is closer to the original data in distribution. Dual manifold optimization makes the generated images more diverse and consistent in class distribution and morphology by adjusting the training processes of the generator and discriminator.

[0091] Specifically, as shown in Figure 2 The training process of the generative adversarial network is as follows:

[0092] Step S111: Initialize the weight parameters of the generator and discriminator of the generative adversarial network respectively.

[0093] The way to initialize the weight parameters of the generator and discriminator is expressed as:

[0094]

[0095]

[0096] In the formula, and respectively represent the weight parameters of the generator and discriminator, represents a normal distribution with a mean of 0 and a variance of 0.01.

[0097] Step S112: Input a random noise vector into the generator, and input the image generated by the generator and the first sample set into the discriminator to obtain the discrimination result of the authenticity of the input image.

[0098] The initial input for generator training is a random noise vector generated by a uniform distribution or a Gaussian distribution. The image generation process is expressed as:

[0099]

[0100] In the formula, is the input random noise vector, and let the dimension of the input noise vector be ; is the generator function, which maps the noise to the image space based on the weight ; is the generated image.

[0101] Input the image generated by the generator and the real image in the first sample set into the discriminator together to obtain the discrimination result of the authenticity of the input image.

[0102] Step S113: Determine the objective function of the generator using the dual manifold optimization mechanism. By maximizing the misclassification probability of the discriminator and minimizing the objective function, determine the generator loss function. Determine the discriminator loss function by maximizing the sum of the correct classification of real images and the misclassification of generated images.

[0103] To maintain the intrinsic topological relationship of image data during the generation process, the dual manifold optimization mechanism is adopted, and the objective function is set to maintain the intrinsic structure and topological relationship of the generated data.

[0104] The process of determining the objective function includes: determining the difference between the random noise vector and the generated image, and taking the square of the L2 norm of this difference as the first objective term; taking the square of the Frobenius norm of the product of this difference and the gradient of the random noise vector, multiplied by the coefficient controlling the importance of the high-order derivative term, as the second objective term; taking the result of multiplying the sum of the first objective term and the second objective term by the weight adjustment coefficient of the dual manifold as the objective function of the generator. It is expressed by the formula:

[0105]

[0106] In the formula, is the objective function, is the weight adjustment coefficient of the dual manifold, is the random noise vector, is the generator function; represents the L2 norm, which is used to measure the difference between the generated data and the original noise; is the coefficient controlling the importance of the high-order derivative term, represents the Frobenius norm, ensuring that during the generation process of the data, not only the positions of the original data points are retained, but also their local structures on the manifold are protected; represents the gradient operation on the random noise vector .

[0107] The training objective of the generator is to minimize its loss function, and the generator loss function is expressed as:

[0108]

[0109] In the formula, represents the discriminator function, indicating that the generator generates realistic images by maximizing the misjudgment probability of the discriminator, and at the same time minimizes to maintain the topological structure of the image.

[0110] Moreover, the training objective of the discriminator is to minimize its loss function, and the discriminator loss function is expressed as:

[0111]

[0112] In the formula, represents the real image data.

[0113] Step S114: Update the weight parameters of the generator based on the gradient of the generator loss function, and update the weight parameters of the discriminator based on the gradient of the discriminator loss function.

[0114] Dynamically adjusting the learning rates of the generator and the discriminator according to the training progress helps to converge quickly in the early stage of training and fine-tune the generation effect in the later stage of training, which is expressed as:

[0115]

[0116]

[0117] In the formula, and are the learning rates of the generator and the discriminator respectively, is the initial learning rate, and are the learning rate decay coefficients of the generator and the discriminator respectively, is the number of iterations of the current training; and are the feedback adjustment coefficients in the learning rate adjustment of the generator and the discriminator respectively; is to evaluate the average generation quality of the generator in the most recent batch of samples, and this quality is measured by the distance between the generated samples and the real samples, so that the model can adaptively adjust the learning rate during training and optimize according to the quality of the generated samples.

[0118] Based on the gradients calculated by backpropagation and combined with the updated learning rates above, optimize the weights of the generator and the discriminator respectively, so that the network weights gradually converge after multiple rounds of training, the generator can generate more real images, and the discriminator can more accurately distinguish between real images and generated images.

[0119] Step S115: Determine whether the iteration termination condition is satisfied.

[0120] The preset iteration termination condition can be reaching the preset maximum number of iterations; preferably, the preset maximum number of iterations is set to 1000 times.

[0121] If the judgment in step S115 is no, return to step S112; if the judgment in step S115 is yes, end the training of the generative adversarial network, and use the trained generator as the image generation model.

[0122] Step S12: Obtain the second sample set, use the image generation model for image augmentation, and establish the third sample set.

[0123] Both the above-mentioned first sample set and the second sample set contain SEM images of the primary pores, metamorphic pores, mineral pores, and strain pores in the coal-rock reservoir, which will be introduced in detail later.

[0124] After the image generation model is trained, the trained image generation model is used to expand the number of samples.

[0125] In one embodiment, assume that the number of samples in the originally collected second sample set is 800. After image expansion using the image generation model, 200 samples are obtained. Then, the expanded third data set contains 1000 samples.

[0126] Step S13: Based on the third sample set, train an extreme learning machine using the dynamic boundary smoothing mechanism to obtain a classification model.

[0127] Adopt the extreme learning machine classification algorithm based on the dynamic boundary smoothing mechanism to identify the pore types. By dynamically adjusting the classification boundary, it solves the sensitivity of traditional classification methods to the uneven distribution between classes when dealing with imbalanced data, can adaptively adjust the classification boundary according to the characteristics of the sample distribution, effectively avoids the imbalance problem of the inter-class boundary, and thus improves the accuracy of pore type identification.

[0128] Specifically, refer to Figure 3 As shown, the training process of the extreme learning machine classification algorithm based on the dynamic boundary smoothing mechanism is as follows:

[0129] Step S131: Input the third sample set into the extreme learning machine, and initialize the weight parameters and bias of the extreme learning machine.

[0130] Initialize the weights and biases of the extreme learning machine according to the dimension of the input data and the complexity of the classification task:

[0131]

[0132]

[0133] In the formula, is the weight of the extreme learning machine, is the bias of the extreme learning machine, is a normal distribution with a mean of 0 and a standard deviation of , is the number of input units, and represents a uniform distribution in the interval , represents being subject to a specific distribution.

[0134] Step S132: Determine the average within-class variance and the average between-class distance based on the initialized classification results to obtain the initial smoothing parameter.

[0135] Determine the ratio of the average within-class variance and the average between-class distance, and take the product of this ratio and the smoothing adjustment factor as the initial smoothing parameter.

[0136] That is, the initial smoothing parameter takes into account the dynamic changes of the within-class variance and the between-class distance, and the calculation method is expressed as:

[0137]

[0138]

[0139]

[0140] In the formula, is the initial smoothing parameter, is the smoothing adjustment factor, is the average within-class variance of all classes, is the average between-class distance, is the variance of the th class, is the number of classes, is the th class mean vector, is the th class mean vector.

[0141] Step S133: Determine the boundary smoothing function according to the current smoothing parameter and the classification result, and obtain a new classification result from the boundary smoothing function, the weight parameter and the bias of the extreme learning machine.

[0142] Adopt a dynamic boundary smoothing mechanism to adjust the decision boundary through an adaptive function to handle the distribution characteristics of different classes. According to the distribution density and overlap degree of each class of data, dynamically adjust the classification boundary to enhance the performance of the model in dealing with multi-class and imbalanced data. The boundary smoothing function is expressed as:

[0143]

[0144] In the formula, is the boundary smoothing function; is the smoothing parameter after the th iteration, and the specific update method will be introduced in detail later; is the input data of the extreme learning machine.

[0145] Use the optimized weights and biases of the extreme learning machine, and the adjusted classification boundary to train the extreme learning machine. The way the extreme learning machine realizes classification is expressed as:

[0146]

[0147] In the formula, is the output of the extreme learning machine, is the input data matrix, is the activation function, specifically the Sigmoid activation function; is the normal distribution; is the noise variance.

[0148] Furthermore, the mean vector represents the class center of the sample distribution and is calculated based on the weighted average of the samples within each class, expressed as:

[0149]

[0150] In the formula, represents the sample set of the th class, is the sample weight, obtained based on the contribution rate of the sample features in the previous classification.

[0151] Furthermore, the calculation method of the within-class variance is expressed as:

[0152]

[0153] In the formula, is the variance of the th class.

[0154] Step S134: Determine the optimization objective function.

[0155] Align the distribution of the support vectors to ensure that the decision boundary between different classes is more reasonable and balanced. By adopting multi-objective optimization terms to optimize the balance and rationality of the decision boundary, reduce the within-class variance and maximize the between-class distance, thereby improving the accuracy and stability of classification. The calculation method is expressed as:

[0156]

[0157] In the formula, is the multi-objective optimization term; and are the weight factors of the within-class variance and the between-class distance respectively, is the variance of the th class.

[0158] Use the feedback mechanism to fine-tune and optimize the model to improve the generalization ability and reduce overfitting. The goal is to enable the model to maintain a high recognition accuracy when facing new and unseen data. The optimization objective function is:

[0159]

[0160] In the formula, is the total error of the extreme learning machine, and the training objective is to make as small as possible;

[0161] is the target output of the extreme learning machine, is the regularization parameter of the extreme learning machine, which aims to balance the fitting error and complexity of the model, thereby improving the generalization performance of the model. Preferably, is set to 0.3.

[0162] Step S135: Determine whether the optimization objective function satisfies the iteration termination condition.

[0163] If the judgment in step S135 is no, execute step S136; if the judgment in step S135 is yes, end the training of the extreme learning machine to obtain a classification model for determining the pore type of the input coal and rock scanning electron microscope image.

[0164] Step S136: Update the weight parameters and biases through backpropagation, and update the smoothing parameter based on the current iteration number.

[0165] The update method of the smoothing parameter is as follows:

[0166]

[0167] is the smoothing decay rate, preferably set to 0.3.

[0168] After step S136, return to execute step S133.

[0169] In one embodiment, an adaptive noise injection mechanism is adopted to dynamically inject noise into the input data during the training process to simulate various interferences that may be encountered in actual operations, thereby improving the adaptability of the model to noise. The calculation method of the noise is expressed as:

[0170]

[0171] where is the noise standard deviation applied to each sample, is the basic noise level, which is related to the overall noise characteristics of the dataset; is the amount of noise dynamically adjusted based on the characteristics of the input sample.

[0172] Furthermore, the amount of noise is calculated according to the local density or variability of the sample, expressed as:

[0173]

[0174] In the formula, is the noise adjustment factor, is the data The local variance within its neighborhood; is the local density, calculated as the number of samples in the data neighborhood. Preferably, is set to 2.

[0175] After adding noise to the input data, the classification result of the extreme learning machine is expressed as:

[0176]

[0177] The pore type classification method for coal rock scanning electron microscope images provided in the first embodiment of the present invention, through an automated classification method, greatly reduces the time and error required for manual analysis, improves the work efficiency of deep coal rock reservoir exploration, and provides more scientific and efficient technical support for coal rock reservoir exploration and the analysis of reservoir gas migration characteristics.

[0178] In one embodiment, the extreme learning machine does not include a feature extraction layer, but instead uses the third sample set obtained after image augmentation to train an autoencoder based on boundary smoothing to obtain a feature extraction model; the output data of the feature extraction model is used to obtain a fourth sample set, which is input into the extreme learning machine and trained using a dynamic boundary smoothing mechanism to obtain a classification model.

[0179] The autoencoder includes an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs this low-dimensional representation into a feature vector that satisfies a similarity condition with the input data, that is, a feature vector that is as similar as possible to the input data.

[0180] Specifically, referring to Figure 4 as shown, the training process of the autoencoder is as follows:

[0181] Step S41: Input the third sample set into the autoencoder, and initialize the weight parameters and biases of the encoder and decoder of the autoencoder respectively.

[0182] In one embodiment, the initialization method is expressed as:

[0183]

[0184]

[0185]

[0186]

[0187] In the formula, represents the weight of the encoder, represents the bias of the encoder; represents the weight of the decoder, represents the bias of the decoder; denotes a random function; is the dimension of the encoder input layer, is the dimension of the encoder middle layer, is the dimension of the encoder output layer; is a function to generate a vector of all zeros.

[0188] Step S42: Based on the current weight parameters and biases of the encoder, obtain the output data of the encoder through an activation function. Based on the current weight parameters and biases of the decoder, obtain the output data of the decoder through an activation function. After the encoder performs boundary smoothing processing on the output data of the decoder, obtain the final smoothed output.

[0189] To handle outliers and boundary effects during the dimensionality reduction process, an automatic boundary smoothing strategy is adopted. By dynamically adjusting the processing of boundary values, the robustness of the model is enhanced. Specifically, it is achieved by smoothing the output of the encoder through a smoothing function, expressed as:

[0190]

[0191]

[0192] In the formula, is the final smoothed output of the encoder, is the smoothing function, is the smoothing intensity parameter; is an automatically adjusted smoothing degree threshold parameter, used to control the degree of smoothing processing and improve the adaptability to the boundaries of input data.

[0193] In one embodiment, the smoothing intensity parameter is determined by an adaptive method based on the local density of data, and the calculation method is expressed as:

[0194]

[0195] In the formula, Calculate the kernel density estimate in its neighborhood can accurately process data in high-density regions and sparse regions, improving the robustness and adaptability of the overall model; denotes the neighborhood; is the kernel density estimation function; is the function to take the average value; is a preset small constant. Preferably, is set to 0.001.

[0196] Step S43: Determine the autoencoder loss function based on the final smoothed output and the weight coefficients of the autoencoder.

[0197] During the training process of the autoencoder, data is processed through forward propagation and then the parameters of the autoencoder are updated using gradient descent through backpropagation of errors. During this process, the calculation method of the loss function of the autoencoder is expressed as:

[0198]

[0199]

[0200] In the formula, is the loss function of the autoencoder; is the input data of the autoencoder, is the output reconstructed by the decoder, is the Sigmoid activation function; is the L2 norm; is the output of the smoothed encoder; is the weight coefficient of the -th dimension of the autoencoder; is the number of samples input to the autoencoder in the current batch.

[0201] In one embodiment, the weight coefficients of the autoencoder are used to adjust the contribution of each dimension feature to the error, and these weights are dynamically allocated according to the importance of the features. The calculation method is expressed as:

[0202]

[0203] In the formula, is the feature variance of the input data of the -th dimension of the autoencoder; is the feature of the input data of the -th dimension of the autoencoder; is a small constant to avoid division by zero; is the feature of the input data of the -th dimension of the autoencoder; is a small constant to avoid division by zero; is the variance function.

[0204] Step S44: Update the weight parameters and biases of the encoder and decoder through backpropagation of the loss function.

[0205] The calculation method of the parameter update amount for updating the parameters in each iteration is expressed as:

[0206]

[0207]

[0208]

[0209]

[0210] In the formula, is the update amount of the weight parameter of the encoder; is the update amount of the bias parameter of the encoder; is the update amount of the weight parameter of the decoder; is the update amount of the bias parameter of the decoder; is the learning rate of the autoencoder in the

[0211] Further, update the parameters of the autoencoder, and the update method is expressed as:

[0212]

[0213]

[0214]

[0215]

[0216] In the formula, represents the parameter update operation.

[0217] Optimize the training process of the autoencoder using the convex hull convergence strategy. By adjusting the learning step size and weight update strategy, make the network converge quickly. The learning rate is dynamically adjusted according to the data distribution characteristics in the current iteration, and the expression is:

[0218]

[0219] In the formula, is the learning rate of the initial autoencoder; is the learning rate of the autoencoder in the is the learning rate decay factor of the autoencoder, is the average diameter of the convex hull in the iteration,

[0220] is the diameter of the initial convex hull. The learning rate decreases as the diameter of the convex hull decreases to prevent instability caused by too large a step size in the later stage of learning.

[0221] The preset iteration termination condition can be reaching the preset maximum number of iterations; preferably, the preset maximum number of iterations is set to 1000 times.

[0222] If the judgment in step S45 is no, return to step S42; if the judgment in step S45 is yes, end the training of the autoencoder and obtain the feature extraction model for image feature extraction.

[0223] The pore type classification method for coal and rock scanning electron microscope images provided by the embodiments of the present invention uses an autoencoder based on boundary smoothing to extract pore features. The autoencoder compresses high-dimensional data into a low-dimensional space through the structures of the encoder and the decoder, and avoids outliers and boundary effects during the dimensionality reduction process through a boundary smoothing strategy. Through the adaptive adjustment of the smoothing intensity, the adaptability of the model to the local structure of the data is enhanced, and the stability and accuracy of feature extraction are improved.

[0224] The above has introduced that both the first sample set and the second sample set contain scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in coal and rock reservoirs. Further, the scanning electron microscope images of primary pores include the scanning electron microscope images of intercellular pores and intergranular pores; the scanning electron microscope images of metamorphic pores include the scanning electron microscope images of gas pores, shrinkage pores, and interchain pores; the scanning electron microscope images of mineral pores include the scanning electron microscope images of dissolution pores, intercrystalline pores, and mold pores.

[0225] The following introduces the data acquisition and label annotation process:

[0226] The embodiments of the present invention relate to collecting scanning electron microscope image data of coal and rock reservoirs, which are derived from cores in multiple coalbed methane coring wells in the same well area. The collection work is obtained through a field emission scanning electron microscope, and high-precision images can be obtained. The collected image data is first preprocessed, including denoising and contrast enhancement, to ensure data quality. All image data is stored in the standard TIFF format; the collected scanning electron microscope image data is all stored at a resolution of 512 × 512 pixels, and each pixel uses a 16-bit depth to maintain the original spectral characteristics of the image, ensuring that every detail can be accurately captured; the number of sampled images is at least 1000 or more to ensure clear images and typical pore characteristics; the collected data is labeled. During the data labeling process, pores are divided into four main genetic categories: primary pores, metamorphic pores, mineral pores, and strain pores.

[0227] Primary pores include intercellular pores and intergranular pores. Intercellular pores are formed by the remains of the cell structure of coal-forming plants and are usually the result of incomplete decomposition of plant tissues during the coalification process. These pores retain the original anatomical structure characteristics of coal-forming plants; the pores are generally larger than 10 μm, mostly circular or elliptical in shape, unevenly distributed, and are mostly found in the vitrinite and fusinite of low to medium metamorphic coals. The pores are often filled with minerals. In high-metamorphic coals, these pores become scarce and irregular in shape due to compaction and mineral filling during the coalification process. Intergranular pores are formed by the voids between clastic particles in coal. The particles may come from the debris of coal-forming plants, minerals, or the deposition of other heterogeneous components; the pore diameters of these pores are usually larger than 1 μm, and the shapes are irregular. In low to medium metamorphic coals, the intergranular pores are more developed, with larger pore diameters and higher concentration; while in high-metamorphic coals, the pores between particles are affected by the compaction effect, the pore diameters decrease, and the number decreases.

[0228] Metamorphic pores mainly include gas pores, shrinkage pores, and inter-chain pores. Gas pores are mainly formed during the coal metamorphism process by the generation and release of thermogenic gases. As the coal rank increases, thermogenic gases are continuously generated, and the process of their expulsion leaves pores within the organic matter. Developmental characteristics: The gas pores in low-rank coals are regular in shape and small in pore size; while the gas pores in high-rank coals can be divided into two types: (1) residual gas pores, formed by the deformation of primary pores under pressure, mostly in short linear shapes; (2) secondary gas pores, formed by later gas emissions or fracture effects, mostly in cluster distributions, with circular or elliptical shapes. Shrinkage pores are formed during coalification due to the volume shrinkage of organic matter caused by dehydration, degassing, and polycondensation, as well as the separation of minerals and organic matter. Developmental characteristics: Such pores are mainly concentrated in high-rank coals and are distributed in the interfacial region between organic matter and minerals. They have larger pore sizes and belong to mesopores (2 - 50 nm) and macropores (>50 nm). Their shapes are complex and often interconnected in clusters. They are the most important connected pores in high-rank coals and can significantly affect the migration of coalbed methane. Inter-chain pores are formed by the polycondensation between organic matter molecular chains during coalification. These pores are mainly distributed in the organic matter of coal and reflect the structural changes at the molecular level. The pore sizes are mainly less than 10 nm and the shapes are irregular. As the coal rank increases, the pore size gradually decreases, while the concentration shows a trend of first decreasing and then increasing. These pores make important contributions to the gas adsorption properties of coal.

[0229] Strain pores are formed due to the relative displacement between particles, shear sliding between minerals and organic matter, or local compressive deformation in the coal seam under tectonic deformation or stress. They usually form in the shear zones, compression-shear zones, or tensile fracture zones of the coal seam and are the direct result of the internal structural adjustment of the coal seam caused by external forces. The morphology of strain pores is diverse and mainly shows irregular shapes such as linear, fissure-like, or arc-shaped. The pore size ranges from micrometers to millimeters and is controlled by the deformation intensity. The strain pores in weakly deformed areas are mainly tiny fissures, while large-scale slip fissures or even through-going pores will appear in strongly deformed areas. Strain pores usually have high connectivity, but strong compaction or slip may cause some pores to close. In addition, the distribution of strain pores is significantly directional, usually consistent with the direction of the tectonic stress field, and has an important impact on fluid migration.

[0230] The identification and annotation of each type of pore are reviewed by domain experts to ensure the accuracy of the annotation. The basic characteristics of various pores are shown in Table 1 and Figures 5a - 5i as follows. Among them, Figure 5a is intercellular pore, Figure 5b is intergranular pore, Figure 5c is gas pore, Figure 5d is shrinkage pore and inter-chain pore, Figure 5e and Figure 5f is dissolution pore, Figure 5g is intercrystalline pore,Figure 5h is the mold hole, Figure 5i is the strain hole.

[0231] Table 1. Pore types, origins and development characteristics of different origins in coal and rock

[0232]

[0233]

[0234] Example Two

[0235] Example Two of the present invention provides a specific application of a pore type classification method for coal and rock scanning electron microscope images.

[0236] Core sample X was obtained from the coalbed methane reservoir of Well A1 in the xx Basin, with a total porosity of 6.91%. High-resolution scanning electron microscope images were collected using a field emission scanning electron microscope (FE-SEM). The collected images had a resolution of 512×512 pixels, a 16-bit gray depth was used to retain the detailed features of the pores, and the number of images was 3,212 FE-SEM images. All samples were preprocessed, including denoising and contrast enhancement, to ensure that the image quality met the analysis requirements. The collected images were labeled by domain experts, and the pores were classified into four major genetic types: primary pores (intercellular pores and intergranular pores), metamorphic pores (gas pores, shrinkage pores, interchain pores), mineral pores (dissolution pores, intercrystalline pores, mold holes), and strain holes. During labeling, pore morphology, size, and development characteristics were referred to to ensure the accuracy of classification. Due to the limited number of collected samples, a generative adversarial network (GAN) based on dual manifold optimization was used to augment the samples. The augmented dataset covered different pore morphologies and sizes, and the generated samples were visually consistent with the original data while significantly increasing the data diversity. An autoencoder based on boundary smoothing was used to reduce the dimensionality of the augmented dataset. The encoder compressed the high-dimensional features into low-dimensional representations, and the decoder reconstructed the features to ensure information integrity. At the same time, a boundary smoothing strategy was introduced to reduce the impact of outliers on the model performance. The reduced-dimensional features were used for subsequent classification model training. An extreme learning machine (ELM) based on a dynamic boundary smoothing mechanism was used to classify the pore categories. The model dynamically adjusted the classification boundary during training, optimized the within-class variance and between-class distance, and significantly improved the classification accuracy and generalization ability. The final classification results are shown in the following table, with an error within ±0.3% compared with the manual recognition results (Table 2).

[0237] Table 2. Comparison of automatic and manual recognition results of pores in Sample X of Well A1 in the xx Basin

[0238]

[0239] According to the above results, it is found that in Sample X1, pores of different genetic types exhibit significant distribution characteristics and proportion features. Primary pores (intercellular pores and intergranular pores) have a relatively low areal porosity (0.06% - 0.02%) in Sample X1 due to compaction and mineral filling during coalification. The porosity is relatively small, and the automatically and manually identified porosities are close, with small errors (absolute errors are 0.01% and 0.06% respectively), indicating that they have tended to be underdeveloped in Sample X1.

[0240] Metamorphic pores, including gas pores, shrinkage pores, and inter-chain pores, are the main pore types in Sample X1. Among them, gas pores have a relatively high areal porosity (0.42%), and a relatively large porosity (automatically identified as 2.90% and manually identified as 3.20%), showing good connectivity, but with a relatively high error (absolute error is 0.30%), reflecting their complexity. Shrinkage pores and inter-chain pores respectively show medium areal porosities (0.10%), and the automatically and manually identified porosity values are close (absolute error of shrinkage pores is 0.01% and that of inter-chain pores is 0.29%). Inter-chain pores especially show micropore characteristics inside the coal rock, which is crucial for adsorbing coalbed methane.

[0241] Mineral pores (dissolution pores, intercrystalline pores, and mold pores) account for a relatively low proportion in Sample X1, with an areal porosity of 0.08% - 0.12%. Among them, intercrystalline pores and dissolution pores respectively show relatively high porosities (0.83% and 0.55%), indicating that these pores may be well developed in local areas of the reservoir and have certain storage and fluid migration capabilities. The areal porosity and porosity of mold pores are both 0, indicating that they are basically underdeveloped in Sample X1.

[0242] Strain pores (such as structural deformation pores) are particularly significant in the tectonic action area, with an areal porosity of 0.10% and a relatively high porosity (0.69% - 0.50%). The absolute error between automatic and manual identification is 0.19%, reflecting the transformation effect of tectonic stress on the pores of the coal reservoir in Sample X1. These pores usually have good connectivity and are important channels for gas migration.

[0243] Generally speaking, pores in Sample X1 are mainly metamorphic pores, mineral pores and strain pores are developed in local areas, and primary pores are basically underdeveloped. This pore characteristic is of great significance for the storage and migration of coalbed methane.

[0244] The embodiments of the present invention comprehensively utilize various technical means such as data collection and annotation, data augmentation, feature extraction, and deep learning classifiers to achieve the automatic identification function of pores of multiple genetic categories in coal-rock gas reservoirs. By introducing a generative adversarial network to augment the training data, the problems of insufficient data and poor model generalization ability are solved; a neural network parameter optimization method based on ecosystem optimization is adopted to improve the training efficiency and accuracy of the model; at the same time, combining a non-linear potential mapping autoencoder and an improved quantum-coded high-order neural network classifier further enhances the model's processing ability and classification accuracy for complex geological data.

[0245] Based on the inventive concept of the present invention, the embodiments of the present invention further provide a method for establishing a pore type classification model for coal-rock scanning electron microscope images, including:

[0246] Using the first sample set, training the generator and discriminator of the generative adversarial network by adopting a dual manifold optimization mechanism to obtain an image generation model;

[0247] Obtaining a second sample set, using the image generation model for image augmentation to establish a third sample set, where both the first sample set and the second sample set include scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in coal-rock reservoirs;

[0248] Based on the third sample set, training an extreme learning machine by adopting a dynamic boundary smoothing mechanism to obtain a classification model for determining the pore type of the input coal-rock scanning electron microscope image.

[0249] Based on the inventive concept of the present invention, the embodiments of the present invention further provide a pore type classification model for coal-rock scanning electron microscope images, where the classification model is used to determine the pore type of the input coal-rock scanning electron microscope image, and the classification model is established through the following steps:

[0250] Using the first sample set, training the generator and discriminator of the generative adversarial network by adopting a dual manifold optimization mechanism to obtain an image generation model;

[0251] Obtaining a second sample set, using the image generation model for image augmentation to establish a third sample set, where both the first sample set and the second sample set include scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in coal-rock reservoirs;

[0252] Based on the third sample set, training an extreme learning machine by adopting a dynamic boundary smoothing mechanism to obtain a classification model.

[0253] Regarding the establishment method of the above classification model, it has been described in detail in Embodiment 1, and will not be elaborated here.

[0254] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a computer storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the above method is implemented.

[0255] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a server, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the above method is implemented.

[0256] Unless otherwise specifically stated, terms such as processing, computing, calculating, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate on and transform data represented as physical (such as electronic) quantities in the registers or memories of the processing system into other data similarly represented as physical quantities in the memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals can be represented using any of a variety of different technologies and methods. For example, the data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0257] It should be understood that the specific order or hierarchy of steps in the disclosed processes is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The appended method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy.

[0258] In the above detailed description, various features are combined in a single embodiment to simplify the present disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are stated in each claim. On the contrary, as reflected in the appended claims, the present invention lies in a state with fewer features than all the features of the disclosed single embodiment. Therefore, the appended claims are hereby expressly incorporated into the detailed description, where each claim stands alone as a separate preferred embodiment of the present invention.

[0259] Those skilled in the art should also understand that all the illustrative logical blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments herein can be implemented as electronic hardware, computer software, or a combination thereof. To clearly illustrate the interchangeability between hardware and software, the above various illustrative components, blocks, modules, circuits, and steps have been generally described in terms of their functions. Whether such a function is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Skilled technicians can implement the described functions in a flexible manner for each specific application. However, such implementation decisions should not be construed as departing from the protection scope of the present disclosure.

[0260] The steps of the methods or algorithms described in conjunction with the embodiments herein can be directly embodied as hardware, software modules executed by a processor, or a combination thereof. The software modules can be located in a RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the art. An exemplary storage medium is connected to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC. The ASIC can be located in a user terminal. Of course, the processor and the storage medium can also exist as discrete components in the user terminal.

[0261] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that execute the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented inside the processor or outside the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well-known in the art.

[0262] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments. However, those of ordinary skill in the art should recognize that the various embodiments can be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the protection scope of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, this term is covered in a manner similar to the term "including," as interpreted when "including" is used as a transitional word in the claims. In addition, any use of the term "or" in the claims or the specification is intended to mean "non-exclusive or." The terms "first," "second," etc. are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

Claims

1. A method for classifying pore types in coal-rock scanning electron microscope images, characterized in that, Including: Determining the pore type of the coal-rock scanning electron microscope image based on a classification model, and the classification model is established through the following steps: Using a first sample set, training a generator and a discriminator of a generative adversarial network by adopting a dual manifold optimization mechanism to obtain an image generation model; Obtaining a second sample set, using the image generation model to perform image augmentation, and establishing a third sample set. Both the first sample set and the second sample set include scanning electron microscope images of primary pores, metamorphic pores, mineral pores, and strain pores in a coal-rock reservoir; Based on the third sample set, training an extreme learning machine by adopting a dynamic boundary smoothing mechanism to obtain a classification model.

2. The method according to claim 1, wherein The training of the generator and the discriminator of the generative adversarial network by adopting the dual manifold optimization mechanism using the first sample set includes: Initializing the weight parameters of the generator and the discriminator of the generative adversarial network respectively; Inputting a random noise vector into the generator, inputting the image generated by the generator and the first sample set into the discriminator, and obtaining the authenticity discrimination result of the input image; Adopting a dual manifold optimization mechanism to determine the objective function of the generator, determining the generator loss function by maximizing the misclassification probability of the discriminator and minimizing the objective function, and determining the discriminator loss function by maximizing the sum of the correct classification of real images and the misclassification of generated images; Updating the weight parameters of the generator based on the gradient of the generator loss function, and updating the weight parameters of the discriminator based on the gradient of the discriminator loss function; If the iteration termination condition is not satisfied, return to execute the step of inputting the random noise vector into the generator.

3. The method according to claim 2, wherein The determining of the objective function of the generator by adopting the dual manifold optimization mechanism includes: Determining the difference between the random noise vector and the generated image, and taking the square of the L2 norm of the difference as the first objective term; Taking the result of multiplying the square of the F norm of the product of the difference and the gradient of the random noise vector by the coefficient controlling the importance of the high-order derivative term as the second objective term; Taking the result of multiplying the sum of the first objective term and the second objective term by the weight adjustment coefficient of the dual manifold as the objective function of the generator.

4. The method according to claim 1, characterized in that, The training of the extreme learning machine by adopting the dynamic boundary smoothing mechanism based on the third sample set includes: Inputting the third sample set into the extreme learning machine, and initializing the weight parameters and biases of the extreme learning machine; Determining the average within-class variance and the average between-class distance according to the initialized classification result to obtain an initial smoothing parameter; Determining a boundary smoothing function according to the current smoothing parameter and the classification result, and obtaining a new classification result from the boundary smoothing function, the weight parameters, and the biases of the extreme learning machine; Determining an optimization objective function. If the optimization objective function does not satisfy the iteration termination condition, updating the weight parameters and biases through backpropagation, updating the smoothing parameter based on the current iteration number, and returning to execute the step of determining the boundary smoothing function according to the current smoothing parameter and the classification result.

5. The method according to claim 4, characterized in that, The obtaining of the initial smoothing parameter includes: Determining the ratio of the average within-class variance to the average between-class distance, and taking the product of the ratio and the smoothing adjustment factor as the initial smoothing parameter.

6. The method according to claim 4, wherein The determining of the boundary smoothing function according to the current smoothing parameter and the classification result includes: Determining the boundary smoothing function through the following formula according to the current smoothing parameter and the classification result: ； In the formula, is the boundary smoothing function, is the input data of the extreme learning machine, For the The smoothing parameter after iterations, For the The mean vector of the class, For the The mean vector of the class, is the number of categories, is the L2 norm.

7. The method according to claim 4, wherein The determining of the optimization objective function includes: ； ； In the formula, is the optimization objective function, is the output data of the extreme learning machine, is the target output of the extreme learning machine, is the regularization parameter of the extreme learning machine, is the weight parameter of the extreme learning machine, is the multi-objective optimization term, and are the weight factors of the within-class variance and the between-class distance respectively, is the class variance, is the class mean vector, is the class mean vector, is the number of classes, is the L2 norm.

8. The method according to claim 4, wherein Updating the smoothing parameter based on the current iteration number includes: Updating the smoothing parameter based on the current iteration number using the following formula: ； In the formula, is the smoothing parameter after the -th iteration, is the initial smoothing parameter, is the smoothing decay rate, and t is the number of iterations.

9. The method according to claim 1, wherein After establishing the third sample set, it further includes: Training an autoencoder based on boundary smoothing using the third sample set to obtain a feature extraction model; correspondingly, Training an extreme learning machine based on the third sample set using a dynamic boundary smoothing mechanism includes: Obtaining a fourth sample set from the output data of the feature extraction model and training an extreme learning machine using a dynamic boundary smoothing mechanism.

10. The method according to claim 9, characterized in that, The autoencoder includes an encoder and a decoder. The encoder is used to perform boundary smoothing on the output data of the decoder through the following smoothing function: ； Among them, is a smoothing function, is the output data of the decoder, is the smoothing intensity parameter; is the smoothing degree threshold parameter.

11. The method according to claim 10, wherein The smoothing intensity parameter is determined using an adaptive method based on the local density of the data.

12. The method according to claim 10, wherein Training the autoencoder based on boundary smoothing using the third sample set includes: Inputting the third sample set into the autoencoder and initializing the weight parameters and biases of the encoder and decoder of the autoencoder respectively; Based on the current weight parameters and biases of the encoder, obtaining the output data of the encoder through an activation function. Based on the current weight parameters and biases of the decoder, obtaining the output data of the decoder through an activation function, and performing boundary smoothing on the output data of the decoder by the encoder to obtain the final smoothed output; Determining the loss function of the autoencoder based on the final smoothed output and the weight coefficients of the autoencoder; Updating the weight parameters and biases of the encoder and decoder through the backpropagation of the loss function; If the iteration termination condition is not satisfied, return to execute obtaining the output data of the encoder through the activation function based on the current weight parameters and biases of the encoder.

13. The method according to any one of claims 1 to 12, characterized in that, The primary pore scanning electron microscope images include the scanning electron microscope images of intercellular pores and interparticle pores; the metamorphic pore scanning electron microscope images include the scanning electron microscope images of gas pores, shrinkage pores, and interchain pores; the mineral pore scanning electron microscope images include the scanning electron microscope images of dissolution pores, intercrystalline pores, and mold pores.

14. A method for establishing a pore type classification model of coal rock scanning electron microscope images, characterized in that, It includes: Training the generator and discriminator of a generative adversarial network using a dual manifold optimization mechanism with the first sample set to obtain an image generation model; Obtaining a second sample set, using the image generation model for image augmentation to establish a third sample set. The first sample set and the second sample set both contain the scanning electron microscope images of the primary pores, metamorphic pores, mineral pores, and strain pores of the coal-rock reservoir; Training an extreme learning machine based on the third sample set using a dynamic boundary smoothing mechanism to obtain a classification model for determining the pore type of the input coal-rock scanning electron microscope image.

15. A pore type classification model for coal-rock scanning electron microscope images, characterized in that, The classification model is used to determine the pore type of the input coal-rock scanning electron microscope image. The classification model is established through the following steps: Training the generator and discriminator of a generative adversarial network using a dual manifold optimization mechanism with the first sample set to obtain an image generation model; Obtaining a second sample set, using the image generation model for image augmentation to establish a third sample set. The first sample set and the second sample set both contain the scanning electron microscope images of the primary pores, metamorphic pores, mineral pores, and strain pores of the coal-rock reservoir; Training an extreme learning machine based on the third sample set using a dynamic boundary smoothing mechanism to obtain a classification model.

16. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the method according to any one of claims 1 to 14 is implemented.

17. A server, characterized in that, Comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the method according to any one of claims 1 to 14 is implemented.

Images