Pore recognition method and device in rock image based on deep learning

Abstract

The application discloses a pore recognition method and device in a rock image based on deep learning, and the method comprises the following steps: acquiring a three-dimensional CT slice image collected from a rock; forming input image data of a pre-trained rock pore recognition model based on the three-dimensional CT slice image, wherein the rock pore recognition model comprises a multi-scale pore feature extraction network and a pore recognition network; extracting pore feature data of multiple scales from the input image data by using the multi-scale pore feature extraction network; taking the pore feature data of each scale as the input of the pore recognition network respectively, and recognizing each pore in the three-dimensional CT slice image and three-dimensional position area data of each pore.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a pore recognition method in a rock image based on deep learning and a computing device. BACKGROUND

[0002] With the resources constantly approaching to the deep part, the development of rock layer (such as coal bed gas) is paid more and more attention. The key to study how the gas migrates in the rock is to obtain the distribution of pores, but due to the research scale of coal rock, the fine pore distribution cannot be obtained. The existing research shows that the fractal characteristics of pores have scale invariance. The pore recognition method in the prior art has low pore recognition accuracy and is difficult to distinguish noise and pores. When the pore recognition is inaccurate, the accurate pore distribution cannot be obtained. SUMMARY

[0003] In order to solve the existing technical problems, the present application provides a pore recognition method in a rock image based on deep learning and a computing device, which can improve the accuracy of pore recognition.

[0004] In a first aspect, a pore recognition method in a rock image based on deep learning is provided, comprising: obtaining a three-dimensional CT slice image collected from a rock; forming input image data of a pre-trained rock pore recognition model based on the three-dimensional CT slice image, wherein the rock pore recognition model comprises a multi-scale pore feature extraction network and a pore recognition network; extracting pore feature data of multiple scales from the input image data by using the multi-scale pore feature extraction network; taking the pore feature data of each scale as the input of the pore recognition network respectively to recognize each pore in the three-dimensional CT slice image and three-dimensional position region data of each pore.

[0005] In a second aspect, a computing device is provided, comprising a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the pore recognition method in the rock image based on deep learning provided in the first aspect of the present application.

[0006] The present application can directly recognize the pores in the three-dimensional CT slice image through the deep learning network, and can improve the accuracy of pore recognition based on the pore feature data of different scales. Moreover, after recognizing the three-dimensional position region data of the pores, the shape factor of the pores is directly output through the pore shape factor calculation network, which is convenient for obtaining more accurate pore distribution data in the rock image. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 It is an application environment diagram of the pore recognition method in the rock image based on deep learning in an embodiment.

[0008] Figure 2 A flow chart of a method for identifying pores in a rock image based on deep learning in an embodiment;

[0009] Figure 3 A network structure diagram of an example rock pore identification model in an embodiment;

[0010] Figure 4 A flow chart of training a rock pore identification model in an embodiment;

[0011] Figure 5 A network structure diagram of training an example rock pore identification model in an embodiment;

[0012] Figure 6 A flow chart of training a rock pore identification model in another embodiment;

[0013] Figure 7 A schematic diagram of a device for identifying pores in a rock image based on deep learning in an embodiment;

[0014] Figure 8 A schematic diagram of a computing device in an embodiment. DETAILED DESCRIPTION

[0015] The technical solutions of the present application will be further described in detail below in combination with the accompanying drawings and specific embodiments.

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terminology used in the description of the present application herein only for the purpose of describing specific embodiments and is not intended to limit the scope of protection of the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0017] In the following description, the expression "some embodiments" describes a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0018] Reference is made to Figure 1Fig. 1 is an application environment diagram of a method for identifying pores in rock images based on deep learning according to an embodiment. The method for identifying pores in rock images based on deep learning is applied in a computing device 10, which can obtain three-dimensional computed tomography (CT) slice images. The three-dimensional CT slice images are three-dimensional slice images in a three-dimensional image reconstructed from three-dimensional measurement projection images, where the three-dimensional slice images are three-dimensional slice images in the three-dimensional image. The three-dimensional measurement projection images can be obtained by performing three-dimensional CT scanning on a rock by a detection device. A radiation source in the detection device rotates around the rock to obtain three-dimensional measurement projection images of the rock from different angles. These projection data are then used to reconstruct a three-dimensional image of the scanned region. The process of obtaining measurement projection images involves attenuation of the radiation source, and different densities of tissues absorb the radiation source to different extents, which is used to distinguish different tissue structures. In the scanning, the radiation ball tube emits a narrow beam of radiation, and the radiation attenuation is measured by the detector opposite the ball tube. The data collected by the detection device is the radiation projection and contour data. The computing device 10 can reconstruct the three-dimensional measurement projection images to obtain a three-dimensional image. In some embodiments, the process of reconstructing the three-dimensional measurement projection images to obtain a three-dimensional image can also be calculated in other devices, and the computing device 10 directly obtains the reconstructed three-dimensional image and then slices the three-dimensional image to obtain three-dimensional CT slice images.

[0019] Referring to Fig. 2 Figure 2 Fig. 3 is a flowchart of a method for identifying pores in rock images based on deep learning according to an embodiment. The method for identifying pores in rock images based on deep learning is applied in a computing device, and the method for identifying pores in rock images based on deep learning includes the following steps:

[0020] S11, obtaining three-dimensional CT slice images collected from a rock.

[0021] S12, based on the three-dimensional CT slice images, forming input image data of a pre-trained rock pore identification model, wherein the rock pore identification model includes a multi-scale pore feature extraction network and a pore identification network.

[0022] In this embodiment, the rock pore identification model is trained based on a training data set. The rock pore identification model includes a multi-scale pore feature extraction network and a pore identification network. The multi-scale pore feature extraction network is used to extract pore features of different sizes. Since the sizes of the pores in the rock layer are diverse, extracting pore features of different scales can facilitate the subsequent pore identification network to identify pores of different sizes.

[0023] S13, extracting a plurality of scales of pore feature data from the input image data by using a multi-scale pore feature extraction network.

[0024] In the embodiment, the multi-scale pore feature extraction network can be a multi-scale convolutional neural network (CNN), the multi-scale pore feature extraction network includes a plurality of convolutional layers of different sizes, and the plurality of convolutional layers are connected. The convolutional layers of different sizes respectively adopt convolutional kernels of different sizes (such as 1x1, 3x3, 5x5) to capture local features of different scales. By outputting respective pore feature data at different convolutional layers, a plurality of scales of pore feature data are obtained.

[0025] For example Figure 3 As shown in the figure, Figure 3 is a network structure diagram of a rock pore recognition model in an embodiment; three different scales of pore feature data are extracted from a three-dimensional CT slice image, which are first scale pore feature data, second scale pore feature data, and third scale pore feature data. The three scales of pore feature data have different pore sizes.

[0026] S14, taking each scale of pore feature data as an input of a pore recognition network respectively, and recognizing each pore in the three-dimensional CT slice image and three-dimensional position region data of each pore.

[0027] In the embodiment, the pore recognition network is mainly used to detect pores and three-dimensional position region data of the pores according to different scales of input pore feature data. The pore recognition network includes a detection head, which is used to recognize boundaries of the pores and centers of the pores. The three-dimensional position region data includes boundary data of the pores and three-dimensional coordinates of the centers of the pores. The detection head includes, but is not limited to, a convolutional layer, a fully connected layer, an activation layer, a pooling layer, and the like. For example, the detection head can be a detection head network structure in a Faster R-CNN model or a network structure in a YOLO detection head.

[0028] For example Figure 3 As shown in the figure, the first scale pore feature data, the second scale pore feature data, and the third scale pore feature data are taken as inputs of the pore recognition network respectively, and first pore recognition results, second pore recognition results, and third pore recognition results are obtained respectively; then the first pore recognition results, the second pore recognition results, and the third pore recognition results are fused to obtain a final pore recognition result, so that pores of different sizes can be included in the final pore recognition result.

[0029] In the above embodiment, the three-dimensional CT slice image collected from the rock is input into a pre-trained rock pore recognition model, different scale pore feature data can be directly extracted through a multi-scale pore feature extraction network, the different scale pore feature data is taken as the input of a pore recognition network respectively, a plurality of different scale pore recognition results are obtained, the different scale pore recognition results are fused, and each pore in the three-dimensional CT slice image and three-dimensional position region data of each pore are obtained. The deep learning network can directly recognize the pores in the three-dimensional CT slice image, and the recognition is based on different scale pore feature data, so that the accuracy of pore recognition is improved, and more accurate pore distribution data in the rock image can be obtained.

[0030] In some embodiments, the rock pore recognition model comprises a pore shape factor calculation network, and the method further comprises:

[0031] The three-dimensional position region data of each pore is taken as the input of the pore shape factor calculation network, and the shape factor of each pore is calculated.

[0032] In this embodiment, the shape factor of the pore is a dimensionless parameter for describing the geometric shape of the pore. For example, in petroleum engineering, the pore shape factor of the rock affects the permeation of oil and gas. When the pore shape is regular (the shape factor is large), the flow channel of oil and gas and other fluids in the rock pore is relatively smooth, and the permeability is high. Therefore, when detecting the pore of the rock, the shape factor of the pore also needs to be accurately detected. The pore shape factor calculation network is used to construct the three-dimensional structure data of the pore according to the three-dimensional position region data of the pore, and calculate the shape factor of the pore by using the formula of the principal axis length method. The pore shape factor calculation network includes but is not limited to a convolution layer, a pooling layer, a full connection layer and an output layer. The convolution layer is used to capture the local boundary information and subtle shape changes of the pore, the main function of the pooling layer is to down-sample the features extracted by the convolution layer, and the full connection layer is located at the end of the network, and integrates and classifies the features extracted by the previous convolution layer, the pooling layer and the like. It maps the input feature vector to the output space, and is used for predicting the final value of the pore shape factor, and the output layer outputs the shape factor of the pore.

[0033] In the above embodiment, the deep learning network can directly recognize the pores in the three-dimensional CT slice image, and the recognition is based on different scale pore feature data, so that the accuracy of pore recognition is improved; and after the three-dimensional position region data of the pore is recognized, the shape factor of the pore is directly output through the pore shape factor calculation network, so that more accurate pore distribution data in the rock image can be obtained subsequently.

[0034] In some embodiments, the method further comprises:

[0035] training the rock pore recognition model.

[0036] As shown in the figure, the flow steps of training the rock pore recognition model include the following: Figure 4

[0037] S41, obtaining a training data set.

[0038] In the embodiment, each training sample in the training data set includes a three-dimensional CT slice sample image collected from a rock sample and a sample label of the three-dimensional CT slice sample image, wherein the sample label includes a three-dimensional position label region of a sample pore and a shape factor label of the sample pore. The three-dimensional position label region is used to indicate the three-dimensional position of the pore in the slice sample image. The shape factor label is used to indicate the parameter value of the pore geometry.

[0039] In the embodiment, for each training sample, the porosity, the maximum pore radius, the minimum pore radius, and the shape factor of the training sample can be calculated after the three-dimensional position label region of the sample pore is known. The shape factor ω = 4πA / L 2 , A is the sample pore surface area, and L is the pore diameter length.

[0040] S42, constructing an initial rock pore recognition model.

[0041] In the embodiment, the initial rock pore recognition model includes an initial multi-scale pore feature extraction network, an initial pore recognition network, and an initial pore shape factor calculation network.

[0042] S43, obtaining a training sample from the training data set as an input training sample, and iteratively training the initial rock pore recognition model based on the input training sample until a training termination condition is met, to obtain a pre-trained rock pore recognition model.

[0043] Optionally, in the process of each iteration training, S43 specifically includes the following steps:

[0044] extracting pore sample features of multiple scales from the input training sample by using the multi-scale pore feature extraction network in the current iteration;

[0045] inputting the pore sample features of each scale into the pore recognition network in the current iteration as an input, to identify the sample pore and the three-dimensional position region data of the sample pore in the input training sample, and to calculate a loss value between the three-dimensional position region data of the sample pore and the three-dimensional position label region of the sample pore based on a first loss function, to obtain a first loss value;

[0046] ​The three-dimensional position region data of the sample pore is taken as an input of a pore shape factor calculation network in the current iteration, a shape factor corresponding to the sample pore is calculated, and a loss value between the shape factor corresponding to the sample pore and a shape factor label of the sample pore is calculated based on a second loss function, to obtain a second loss value;

[0047] A total loss value in the current iteration is obtained based on the first loss value and the second loss value.

[0048] When the total loss value in the current iteration does not satisfy the training termination condition, a training sample is continuously obtained from the training data set as an input training sample, and the iterative training is continuously performed until the training termination condition is satisfied, and the rock pore recognition model after satisfying the training termination condition is taken as a pre-trained rock pore recognition model.

[0049] In the embodiment, the network structure of the rock pore recognition model in training is the same as that of the trained rock pore recognition model, and only the model parameters of the rock pore recognition model in training have not been fixed, and need to be iteratively trained to find the optimal model parameters. The first loss function and the second loss function can be the same or different, for example, can be a mean square error loss function, a cross-entropy loss function, etc. When there are multiple sample pores in an input training sample, for each sample pore, a loss value between the three-dimensional position region data corresponding to each sample pore and the three-dimensional position label region is calculated, and the loss values between the three-dimensional position region data corresponding to each sample pore and the three-dimensional position label region are accumulated, to obtain the first loss value. Similarly, for each sample pore, a loss value between the shape factor corresponding to each sample pore and the shape factor label is calculated, and the loss values between the shape factor corresponding to each sample pore and the shape factor label are accumulated, to obtain the second loss value. The first loss value and the second loss value can be weighted to obtain the total loss value.

[0050] For example, Figure 5As shown in a network structure diagram of an example training rock pore recognition model in an embodiment, input of a multi-scale pore feature extraction network is formed based on an input training sample, first, second, and third scale pore sample features are respectively output by the multi-scale pore feature extraction network, and are respectively taken as input of a pore recognition network, to obtain first, second, and third sample pore recognition results, the three recognition results are integrated to obtain sample pore and three-dimensional position region data of the sample pore of the input training sample, then based on the three-dimensional position region data of the sample pore obtained in this iteration, first and second loss values are respectively calculated, and finally a total loss value is obtained, when it is determined based on the total loss value that iteration training needs to be continued, input training samples are continuously obtained for training. In the training process, the rock pore recognition model continuously learns position features and shape features of sample pores of different scales from the input training samples, learns position features and boundary feature data of sample pores with a three-dimensional position label region of the sample pore as a training target, and learns shape features of sample pores with a shape factor label of the sample pore as a training target.

[0051] In the above embodiment, in the training process, two different losses are integrated, which enables the rock pore recognition model to learn more features and accelerate the convergence speed of the rock pore recognition model, thereby improving the pore recognition accuracy of the rock pore recognition model.

[0052] In some embodiments, as shown in FIG. 43, S43 can further include the following steps: Figure 6

[0053] S431, obtaining three-dimensional position region data of a current sample pore in each input training sample output by the rock pore recognition model meeting an iteration training condition.

[0054] In this embodiment, the iteration training condition includes that the number of iterations reaches a number of one round of iteration, for example, 1000 times is set as the number of one round of iteration, after one round of iteration is completed, the rock pore recognition model after the iteration can be obtained, part or all of the training samples in the training data set can be taken as input training samples, and the three-dimensional position region data of the current sample pore in each input training sample can be output by the rock pore recognition model after the iteration.

[0055] S432, obtaining target input training samples meeting a screening condition based on the three-dimensional position region data of the current sample pore in each input training sample.

[0056] Optionally, the obtaining of the target input training samples meeting the screening condition based on the three-dimensional position region data of the current sample pore in each input training sample includes:

[0057] ​Based on the three-dimensional position region data of the current sample pore in each input training sample, the porosity corresponding to each input training sample is calculated.

[0058] The input training sample with a porosity less than a preset porosity is selected as a target input training sample.

[0059] In this embodiment, the porosity represents the ratio of the pore volume to the total volume. For any input training sample, after obtaining the three-dimensional position region data of the current sample pore in the input training sample, the volume of all pores can be calculated, and thus the porosity of each input training sample can be calculated.

[0060] Due to the pore volume effect, a part of the pores is difficult to effectively identify due to noise interference, and the characteristics of this part of the pores are that they have a lower attenuation coefficient than the surrounding image neighborhood (i.e., this part of the pore image generally displays a darker region). In addition, the pore fractal feature has scale invariance, and thus the porosity obtained based on the three-dimensional position region data of the current sample pore is generally lower than the true porosity of the input training sample. Therefore, the pores need to be reconstructed, that is, the pore region is increased in the target region meeting the requirements to enhance the features of this part, that is, to enhance the data set of this part, so that the rock pore recognition model in the training can learn more features of this darker region, and thus the porosity obtained based on the three-dimensional position region data of the recognized sample pore is closer to the true porosity.

[0061] S433, based on the three-dimensional position region data of the current sample pore in the target input training sample, a polyhedron corresponding to each current sample pore in the target input training sample is generated, a plurality of polyhedrons are obtained, and the average voxel value of each polyhedron is calculated.

[0062] In this embodiment, when there are a plurality of current sample pores for the target input training sample, a polyhedron is generated for each current sample pore, and the process of generating the polyhedron can be based on the three-dimensional boundary data of the current sample pore.

[0063] S434, according to the average voxel value of each polyhedron, a target polyhedron meeting the condition is selected.

[0064] In this embodiment, since it is necessary to enhance the pore data set of the darker region, the polyhedron with an average voxel value lower than a preset voxel value is selected as the target polyhedron.

[0065] S435, in the target polyhedron, a target sample pore meeting the pore condition is generated, and a sample label corresponding to the target sample pore is generated, the target input training sample is updated based on the target sample pore and the sample label corresponding to the target sample pore, and the updated target input training sample is updated to the training data set to continue training the rock pore recognition model.

[0066] Optionally, S435 specifically further comprises:

[0067] obtaining a first parameter and a second parameter of the target input training sample, wherein the first parameter represents a characteristic size ratio of the fractal structure at adjacent two scales, and the second parameter represents a base element number ratio covering the fractal structure at adjacent scales;

[0068] obtaining a maximum pore radius in the target input training sample according to three-dimensional position region data of a current sample pore in the target input training sample;

[0069] updating the maximum pore radius according to the first parameter to obtain an updated maximum pore radius;

[0070] determining a number of the target sample pores according to the second parameter and taking the updated maximum pore radius as a radius of the target sample pore;

[0071] generating the target sample pores in the target polyhedron according to the number of the target sample pores and the radius of the target sample pore.

[0072] In the embodiment, for the target input training sample, the first parameter and the second parameter can be obtained according to the three-dimensional position label region of the sample pore in the target input training sample and by using multi-scale coverage and fractal scaling law analysis. The calculation of this part is prior art, and will not be described here. The maximum pore radius is the maximum distance from the center of the pore to the boundary. Therefore, the number of the target sample pores and the radius of the target sample pore can be used to randomly generate a pore region meeting the conditions in the target polyhedron. The geometric distribution of the pore meets the scale invariance, which is specifically reflected in the first parameter and the second parameter. The first parameter is the scaling rate of the fractal pore size, which represents the proportion of each pore size to the previous pore size in the fractal iteration. In the fractal iteration, each iteration generates smaller pores. According to the first parameter to update the maximum pore radius, the self-similarity of the pore structure at different iteration levels can be ensured. At the same time, by dynamically adjusting the maximum pore radius through the first parameter, the model can better adapt to different types of coal matrix and different geological conditions.

[0073] The second parameter is the scaling coverage of the fractal pore number, which represents the proportion of the total number of pores in each iteration to the total number of pores in the previous iteration in the fractal iteration process. In the fractal iteration process, each iteration generates more pores. According to the second parameter to determine the number of new pores after each iteration to ensure the consistent self-similarity of the pore distribution.

[0074] The maximum pore radius and the pore number are updated by the first parameter and the second parameter in the fractal terrain theory, so that the pore structure of the rock can be more accurately characterized, and the self-similarity and universality of the model are improved.

[0075] Optionally, the updated maximum pore radius is equal to a ratio of the maximum pore radius and the first parameter.

[0076] Optionally, the number of the target sample pores is equal to a product of the second parameter and the number of the current sample pores in the target input training sample.

[0077] In the above embodiment, the target input training sample with a generally lower porosity than the real porosity of the input training sample is selected, and then the polyhedron corresponding to each current sample pore is generated based on the three-dimensional position region data of the current sample pore of the target input training sample, the target polyhedron in the darker region is selected, the data set of the pore region in the target polyhedron is enhanced, and the updated target input training sample is updated to the training data set to continue training the rock pore recognition model. The rock pore recognition model in the training can learn more features of the darker region, so that the porosity obtained based on the three-dimensional position region data of the recognized sample pore is closer to the real porosity, so that the rock pore recognition model after training can more accurately recognize the pores in the darker region, and the accuracy of pore recognition is improved.

[0078] In another aspect of the present application, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the pore recognition method in a rock image based on deep learning as described in any embodiment of the present application.

[0079] In the computer program product, the optional implementation form of the program module architecture of the computer program implementing each step of the target recognition method can be a pore recognition device in a rock image based on deep learning.

[0080] Please refer to Figure 7 In an embodiment of the present application, a pore recognition device in a rock image based on deep learning is provided, which includes: an acquisition module 71 configured to acquire a three-dimensional CT slice image collected from a rock; an identification module 72 configured to form input image data of a pre-trained rock pore recognition model based on the three-dimensional CT slice image, wherein the rock pore recognition model includes a multi-scale pore feature extraction network and a pore recognition network; the identification module 72 is further configured to extract pore feature data of multiple scales from the input image data by using the multi-scale pore feature extraction network; and the identification module 72 is further configured to identify each pore in the three-dimensional CT slice image and three-dimensional position region data of each pore by taking the pore feature data of each scale as an input of the pore recognition network.

[0081] Optionally, the recognition module 72 is further configured to:

[0082] input the three-dimensional position region data of each of the pores into the pore shape factor calculation network to calculate the shape factor of each of the pores.

[0083] Optionally, the pore recognition apparatus based on deep learning in the rock image further comprises a training module 73 configured to:

[0084] train the rock pore recognition model.

[0085] The training of the rock pore recognition model comprises:

[0086] obtaining a training data set, each training sample in the training data set comprising a three-dimensional CT slice sample image collected from a rock sample and a sample label of the three-dimensional CT slice sample image, wherein the sample label comprises a three-dimensional position label region of a sample pore and a shape factor label of the sample pore;

[0087] constructing an initial rock pore recognition model;

[0088] obtaining a training sample from the training data set as an input training sample, and iteratively training the initial rock pore recognition model based on the input training sample until a training termination condition is met to obtain a pre-trained rock pore recognition model.

[0089] Optionally, the training module 73 is further configured to:

[0090] extracting pore sample features of multiple scales from the input training sample by using a multi-scale pore feature extraction network in a current iteration;

[0091] inputting the pore sample features of each scale into a pore recognition network in the current iteration to recognize a sample pore in the input training sample and three-dimensional position region data of the sample pore, and calculating a loss value between the three-dimensional position region data of the sample pore and a three-dimensional position label region of the sample pore based on a first loss function to obtain a first loss value;

[0092] inputting the three-dimensional position region data of the sample pore into a pore shape factor calculation network in the current iteration to calculate a shape factor corresponding to the sample pore, and calculating a loss value between the shape factor corresponding to the sample pore and a shape factor label of the sample pore based on a second loss function to obtain a second loss value;

[0093] obtaining a total loss value in the current iteration based on the first loss value and the second loss value;

[0094] When the total loss value in the current iteration does not satisfy the training termination condition, continue to obtain a training sample from the training data set as an input training sample, continue the iterative training until the training termination condition is satisfied, and take the rock pore recognition model satisfying the training termination condition as a pre-trained rock pore recognition model.

[0095] Optionally, the training module 73 is further configured to:

[0096] obtain three-dimensional position region data of a current sample pore in each input training sample from the rock pore recognition model satisfying the iterative training condition;

[0097] obtain a target input training sample meeting a screening condition based on the three-dimensional position region data of the current sample pore in each input training sample;

[0098] generate a polyhedron corresponding to each current sample pore in the target input training sample based on the three-dimensional position region data of the current sample pore in the target input training sample, obtain a plurality of polyhedrons, and calculate an average voxel value of each polyhedron;

[0099] screen a target polyhedron meeting a condition according to the average voxel value of each polyhedron;

[0100] generate a target sample pore meeting a pore condition in the target polyhedron and generate a sample label corresponding to the target sample pore, update the target input training sample based on the target sample pore and the sample label corresponding to the target sample pore, and update the updated target input training sample to the training data set to continue training the rock pore recognition model.

[0101] Optionally, the training module 73 is further configured to:

[0102] calculate a porosity corresponding to each input training sample based on the three-dimensional position region data of the current sample pore in each input training sample;

[0103] select an input training sample with a porosity less than a preset porosity as a target input training sample.

[0104] Optionally, the training module 73 is further configured to:

[0105] obtain a first parameter and a second parameter of the target input training sample, wherein the first parameter represents a characteristic size ratio of a fractal structure at adjacent two scales, and the second parameter represents a primitive number ratio of a covering fractal structure at adjacent scales;

[0106] obtain a maximum pore radius in the target input training sample according to the three-dimensional position region data of the current sample pore in the target input training sample.

[0107] updating the maximum pore radius according to the first parameter to obtain an updated maximum pore radius;

[0108] determining the number of the target sample pores according to the updated maximum pore radius as the radius of the target sample pores and the second parameter;

[0109] generating the target sample pores in the target polyhedron according to the number of the target sample pores and the radius of the target sample pores.

[0110] Optionally, the updated maximum pore radius is equal to a ratio of the maximum pore radius and the first parameter.

[0111] Optionally, the number of the target sample pores is equal to a product of the second parameter and the number of the current sample pores in the target input training sample.

[0112] Referring to Figure 8 In another aspect, the present application provides a computing device 10, comprising a memory 3011 and a processor 3012, the memory 3011 stores a computer program, and the computer program is executed by the processor to make the processor 3012 execute the steps of the pore recognition method in the rock image based on deep learning provided by any of the above embodiments. The computing device 10 is, for example, a desktop computer, a laptop computer, a tablet computer, a handheld computer, a smart speaker, a server, etc., a mobile phone (for example, a smart phone, a wireless phone, etc.), a wearable device (for example, a pair of smart glasses or a smart watch) or the like.

[0113] The processor 3012 is a control center, which connects various parts of the computer device through various interfaces and lines, executes various functions of the computer device and processes data by running or executing software programs and / or modules stored in the memory 3011 and calling data stored in the memory 3011. Optionally, the processor 3012 can include one or more processing cores; preferably, the processor 3012 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user pages and application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 3012.

[0114] The memory 3011 can be used to store software programs and modules, and the processor 3012 executes various function applications and data processing by running the software programs and modules stored in the memory 3011. The memory 3011 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 3011 can include a high-speed random access memory, and can also include a nonvolatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device. Accordingly, the memory 3011 can also include a memory controller to provide the processor 3012 with access to the memory 3011.

[0115] In another aspect, the embodiment of the present application also provides a storage medium storing a computer program, and the computer program is executed by a processor to make the processor execute the steps of the pore identification method in a rock image based on deep learning provided by any one of the above-mentioned embodiments of the present application.

[0116] Those skilled in the art can understand that all or part of the processes in the method provided by the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium, and the program can include the processes of the above-mentioned embodiments when executed. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0117] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. The protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for identifying pores in a rock image based on deep learning, characterized by, The method comprises: obtaining a three-dimensional CT slice image of a rock sample; forming input image data of a pre-trained rock pore recognition model based on the three-dimensional CT slice image, wherein the rock pore recognition model comprises a multi-scale pore feature extraction network and a pore recognition network; extracting pore feature data of multiple scales from the input image data by using the multi-scale pore feature extraction network; inputting the pore feature data of each scale into the pore recognition network respectively to identify each pore in the three-dimensional CT slice image and three-dimensional position region data of each pore. The method further comprises: training the rock pore recognition model, wherein the training of the rock pore recognition model comprises: obtaining a training data set, each training sample in the training data set comprising a three-dimensional CT slice sample image of a rock sample and a sample label of the three-dimensional CT slice sample image, wherein the sample label comprises three-dimensional position region label of a sample pore and shape factor label of the sample pore; constructing an initial rock pore recognition model; obtaining a training sample from the training data set as an input training sample, iteratively training the initial rock pore recognition model based on the input training sample until a training termination condition is met to obtain the pre-trained rock pore recognition model; wherein the obtaining of the training sample from the training data set as the input training sample, the iteratively training of the initial rock pore recognition model based on the input training sample until the training termination condition is met to obtain the pre-trained rock pore recognition model comprises: obtaining three-dimensional position region data of a current sample pore in each input training sample output by the rock pore recognition model meeting an iteration training condition; obtaining a target input training sample meeting a screening condition based on the three-dimensional position region data of the current sample pore in each input training sample; generating polyhedrons corresponding to each current sample pore in the target input training sample based on the three-dimensional position region data of the current sample pore in the target input training sample to obtain a plurality of polyhedrons and calculating average voxel values of the polyhedrons; screening target polyhedrons meeting a condition according to the average voxel values of the polyhedrons; generating a target sample pore meeting a pore condition in the target polyhedrons and generating a sample label corresponding to the target sample pore, updating the target input training sample based on the target sample pore and the sample label corresponding to the target sample pore, and updating the updated target input training sample to the training data set for further training of the rock pore recognition model. 2.The rock image-based pore identification method using deep learning according to claim 1, wherein, The rock pore recognition model comprises a pore shape factor calculation network, and the method further comprises: inputting the three-dimensional position region data of each pore into the pore shape factor calculation network to calculate a shape factor of each pore. 3.The method of claim 1, wherein, the obtaining of the training sample from the training data set as the input training sample, the iteratively training of the initial rock pore recognition model based on the input training sample until the training termination condition is met to obtain the pre-trained rock pore recognition model comprises: extracting, from the input training sample in the current iteration, pore sample features at multiple scales by using a multi-scale pore feature extraction network; inputting the pore sample features at each scale into a pore recognition network in the current iteration, to identify sample pores in the input training sample and three-dimensional position region data of the sample pores, and calculating a loss value between the three-dimensional position region data of the sample pores and three-dimensional position region labels of the sample pores based on a first loss function, to obtain a first loss value; inputting the three-dimensional position region data of the sample pores into a pore shape factor calculation network in the current iteration, to calculate a shape factor corresponding to the sample pores, and calculating a loss value between the shape factor corresponding to the sample pores and a shape factor label of the sample pores based on a second loss function, to obtain a second loss value; obtaining a total loss value in the current iteration based on the first loss value and the second loss value; when the total loss value in the current iteration does not satisfy the training termination condition, continuing to obtain training samples from the training data set as input training samples, and continuing the iterative training until the training termination condition is satisfied, and taking the rock pore recognition model after the training termination condition is satisfied as the pre-trained rock pore recognition model. 4.The method of claim 1, wherein, The obtaining of the target input training sample that meets the screening condition based on the three-dimensional position region data of the current sample pore in each input training sample includes: calculating a porosity corresponding to each input training sample based on the three-dimensional position region data of the current sample pore in each input training sample; selecting an input training sample with a porosity less than a preset porosity as the target input training sample. 5.The rock image-based pore identification method using deep learning according to claim 1, wherein, The pore condition includes the number of target sample pores and the radius of the target sample pores, and the method further includes: obtaining a first parameter and a second parameter of the target input training sample, wherein the first parameter represents a characteristic size ratio of fractal structures at two adjacent scales, and the second parameter represents a primitive number ratio of covering fractal structures at adjacent scales; obtaining a maximum pore radius in the target input training sample according to the three-dimensional position region data of the current sample pore in the target input training sample; updating the maximum pore radius according to the first parameter to obtain an updated maximum pore radius; taking the updated maximum pore radius as the radius of the target sample pore and determining the number of the target sample pores according to the second parameter; generating the target sample pore in the target polyhedron according to the number of the target sample pores and the radius of the target sample pore. 6.The rock image-based pore identification method using deep learning according to claim 5, wherein, The updated maximum pore radius is equal to a ratio of the maximum pore radius and the first parameter. 7.The rock image-based pore identification method using deep learning according to claim 5, wherein, The number of the target sample pores is equal to a product of the second parameter and the number of the current sample pore in the target input training sample.

8. A computing device, comprising: The method comprises a memory and a processor, the memory stores a computer program, the computer program is executed by the processor, and the processor executes the pore identification method in the rock image based on deep learning as claimed in any one of claims 1 to 7.

Images