Rock multi-mineral microcosmic implicit three-dimensional reconstruction method and system based on deep learning

Abstract

The invention provides a rock multi-mineral microcosmic implicit three-dimensional reconstruction method and system based on deep learning, and the method comprises the steps: obtaining a fractal dimension function of a rock test piece based on a first microcosmic image sequence of the rock test piece, the first microcosmic image sequence representing a small number of images obtained based on the rock test piece; constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock test piece, and generating a plurality of rock two-dimensional images through the trained generative adversarial network; and under the condition that the two-dimensional images of the plurality of rocks meet a preset precision condition, reconstructing based on the two-dimensional images of the plurality of rock samples to obtain a rock three-dimensional model. According to the invention, through the constructed generative adversarial network based on the fractal dimension feature pyramid algorithm, rock three-dimensional reconstruction based on a small number of rock microscopic images is realized.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method and system for rock multi-mineral microscopic implicit three-dimensional reconstruction based on deep learning. Background Art

[0002] The microstructural characteristics of rock materials determine the flow evolution of various fluids inside the rock (such as oil, natural gas and CO2 hydrates, etc.), the evolution of mechanical properties and fracture mechanisms of deep rock reservoirs. Accurately characterizing the 3D (3D) microstructure of rock materials and studying their fracture process and mechanical property evolution have great application prospects and economic value for deep reservoir resource exploitation and space development.

[0003] The 3D reconstruction algorithms of rock materials mainly include direct 3D reconstruction, theoretical 3D reconstruction, random statistical 3D reconstruction technology and 3D reconstruction based on machine learning. These technologies can establish 3D structural models of rocks and promote the study of rock physical and mechanical properties and fracture laws. However, these existing technologies are limited by factors such as rock material size, imaging resolution, image data volume and economic use cost, and cannot establish accurate 3D models of rock microstructures. Summary of the invention

[0004] The present invention provides a method and system for implicit three-dimensional reconstruction of rock multi-mineral microstructures based on deep learning, which is used to solve the defects of the prior art that the rock three-dimensional reconstruction data is redundant and the data volume is small, resulting in insufficient accuracy, and to realize a method for three-dimensional reconstruction of rock microstructures with a small amount of data and higher accuracy.

[0005] The present invention provides a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning, comprising: Acquire a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; Based on the fractal dimension function of the rock specimen, a generative adversarial network based on a fractal dimension feature pyramid algorithm is constructed, and a plurality of two-dimensional images of the rock are generated by the trained generative adversarial network; In the case that the multiple two-dimensional images of the rocks meet the preset accuracy conditions, a three-dimensional model of the rock is reconstructed based on the multiple two-dimensional images of the rock samples.

[0006] According to a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention, before the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, the method further includes: Collecting microscopic multiphase images of the rock specimen to obtain the first microscopic image sequence; Segmenting the images in the first microscopic image sequence to obtain segmented images of each type of sub-phase structure of the rock specimen; The segmented images of each type of sub-phase structure of the rock specimen are encoded, and a data set for training the generative adversarial network is constructed based on the encoded segmented images.

[0007] According to a rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning provided by the present invention, the step of segmenting the images in the first microscopic image sequence to obtain the segmented images of each type of sub-phase structure of the rock specimen specifically includes: Inputting the images in the first microscopic image sequence into a pre-constructed deep learning segmentation model to obtain a segmentation image of each type of sub-phase structure of the rock specimen output by the deep learning segmentation model; Among them, the deep learning segmentation model constructs a loss function based on the rock image microstructure target loss, the real target and predicted target boundary compensation loss and the rock microstructure fractal dimension loss, and is trained using the mineral boundary difference as a low-rank fine-tuning parameter.

[0008] According to a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention, the step of encoding the segmented image of each type of sub-phase structure of the rock specimen and constructing a data set for training the generative adversarial network based on the encoded segmented image specifically includes: The segmented image corresponding to each image in the first microscopic image sequence is coded with numbers and mineral names according to a preset order; The segmented images with numbers and mineral name codes corresponding to each image in the first microscopic image sequence are merged, and the merged images are organized into a data set for training the generative adversarial network.

[0009] According to a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention, the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen specifically includes: Constructing a generator of the generative adversarial network based on the fractal dimension function of the rock specimen; Constructing a discriminator of the generative adversarial network based on the difference between the generated distribution and the true distribution; A gradient penalty function of the discriminator is constructed based on the boundary similarity parameters.

[0010] According to a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention, the preset accuracy conditions include: generalization evaluation index, fractal dimension standard deviation and accuracy rate; When the generalization evaluation index is less than a first preset threshold, the fractal dimension standard deviation is not greater than a second preset threshold, and the accuracy is not less than a third preset threshold, it is considered that the two-dimensional image of the rock sample meets the preset accuracy conditions.

[0011] The present invention also provides a rock multi-mineral microscopic implicit three-dimensional reconstruction system based on deep learning, comprising: An acquisition module, configured to acquire a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; A generation module is used to construct a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generate a plurality of two-dimensional images of the rock specimen through the trained generative adversarial network; The reconstruction module is used to reconstruct a three-dimensional rock model based on the two-dimensional images of the multiple rock samples when the two-dimensional images of the multiple rock samples meet a preset accuracy condition.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for implicit three-dimensional microscopic reconstruction of multi-minerals of rocks based on deep learning as described in any one of the above-mentioned methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the deep learning-based rock multi-mineral microscopic implicit three-dimensional reconstruction methods described above.

[0014] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the deep learning-based rock multi-mineral microscopic implicit three-dimensional reconstruction methods described above.

[0015] The deep learning-based implicit three-dimensional reconstruction method and system of rock multi-mineral microstructure provided by the present invention generates a two-dimensional image of rock with a fractal dimension function similar to the microstructure image of the rock specimen by constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm, and determines the optimal network with the best generalization ability in combination with preset accuracy conditions. Finally, multiple two-dimensional images of rocks are generated based on the determined generative adversarial network, and the rock three-dimensional model is rendered, so that the final rock three-dimensional model can accurately reflect the microstructure of the rock sample, facilitate the precise study of the microscopic mechanical properties and fracture mechanism of the rock based on the generated rock three-dimensional model, and realize the three-dimensional reconstruction of the rock based on a small number of rock microscopic images. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is one of the flow charts of the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 2 (a) is a schematic diagram of a microstructure image of a rock specimen in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 2 (b) is a deep learning segmentation model in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 2 (c) is a 0-pore segmentation image in the rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning provided by the present invention; Figure 2 (d) is a 1-feldspar segmentation image in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 2 (e) is a 2-quartz segmentation image in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 2 (f) is a 3-muscovite segmentation image in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 3 (a) is a schematic diagram of a microstructure image with digital and mineral name coding input into the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 3 (b) is a plurality of two-dimensional images of rocks generated by MGAN in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 3 (c) is a 3D data set obtained based on multiple two-dimensional images of rocks in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 4 (a) is a three-dimensional rock model reconstructed in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 4(b) is a 0-pore three-dimensional model reconstructed in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 4 (c) is a 1-feldspar 3D model reconstructed in the rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning provided by the present invention; Figure 4 (d) is a 2-quartz three-dimensional model reconstructed in the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the present invention; Figure 4 (e) is a 3-muscovite 3D model reconstructed in the rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning provided by the present invention; Figure 5 It is a structural schematic diagram of a rock multi-mineral microscopic implicit three-dimensional reconstruction system based on deep learning provided by the present invention; Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Combine the following Figures 1 to 4 The present invention introduces a rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning, such as Figure 1 As shown, including: Step 101, acquiring a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; Based on the processed rock specimens, high-precision modern imaging technology is used to perform high-resolution digital imaging of the microstructure of the rock material to obtain a first microscopic image sequence, wherein the first microscopic image sequence contains a small amount of rock microscopic multiphase images. In this embodiment, one image is collected from each of the upper, middle and lower layers of the rock sample, and the obtained three images are used as the first microscopic image sequence.

[0020] Optionally, each image in the first microscopic image sequence is segmented to obtain a microscopic image of each sub-type of material, and then based on the microscopic image of each type of sub-phase structure, the fractal dimension function of the rock specimen is obtained.

[0021] Step 102, constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generating a plurality of two-dimensional images of the rock through the trained generative adversarial network; It can be understood that the fractal dimension function of the rock specimen characterizes the shape and distribution of the sub-phase structure of the rock specimen. Therefore, a generative adversarial network (MGAN) based on the fractal dimension feature pyramid algorithm is constructed and trained based on the fractal dimension function of the rock specimen.

[0022] That is to integrate the pyramid algorithm into the traditional generative adversarial network (GAN), and capture the structural complexity of image data at various resolutions by calculating the fractal dimension functions at different scales. This makes the images generated by the MGAN network trained based on a small number of images similar to the fractal dimension of the microstructure images of rock specimens, and enables the trained MGAN to generate two-dimensional images of rocks with higher accuracy.

[0023] Optionally, during the training process, a GAN network with the best generalization ability is selected as the trained GAN model.

[0024] By training a generative adversarial network to generate multiple two-dimensional images of rocks, the generated images have a fractal dimension function that is closer to the rock specimens, and can more realistically reflect the structural characteristics of the rock specimens, so that they can be used to study the microscopic mechanical properties and fracture mechanisms of rocks.

[0025] Step 103: When the plurality of two-dimensional images of the rock meet a preset accuracy condition, a three-dimensional model of the rock is reconstructed based on the plurality of two-dimensional images of the rock.

[0026] In order to make the two-dimensional images of rocks generated by GAN reflect the microstructure of rock specimens more realistically, after the GAN training is completed, its accuracy needs to be evaluated through pre-set preset accuracy conditions.

[0027] Optionally, the preset accuracy condition includes a generalization ability evaluation condition and / or a model accuracy evaluation condition. The threshold of each evaluation condition can be determined based on experience, and is usually set to be greater than 90%, such as setting the model accuracy to be greater than 90%.

[0028] When the two-dimensional image of the rock generated by GAN meets the preset accuracy conditions, the generative adversarial network at this time is considered to be the optimal network with the best generalization ability. Therefore, based on the generative adversarial network at this time, it is possible to sequentially generate multiple layers of two-dimensional images of the rock, and then render them into a three-dimensional model of the rock. Among them, the two-dimensional image of the rock is a characteristic image of the rock microstructure.

[0029] For example, 300 layers of two-dimensional images of rocks are generated with an image size of 300×300. The generated 300 two-dimensional images of rocks are constructed into a three-dimensional data set, and a 300×300×300 rock three-dimensional model is rendered.

[0030] The present invention generates a two-dimensional image of rock that is similar to the fractal dimension function of the microstructure image of the rock specimen by constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm, and determines the optimal network with the best generalization ability in combination with preset accuracy conditions. Finally, multiple two-dimensional images of rocks are generated based on the determined generative adversarial network, and a three-dimensional model of the rock is rendered, so that the final three-dimensional rock model can accurately reflect the microstructure of the rock sample, facilitate the precise study of the microscopic mechanical properties and fracture mechanism of the rock based on the generated three-dimensional rock model, and realize the three-dimensional reconstruction of the rock based on a small number of rock microscopic images.

[0031] In the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning of the present invention, before the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, the method further includes: Collecting microscopic multiphase images of the rock specimen to obtain the first microscopic image sequence; Segmenting the images in the first microscopic image sequence to obtain segmented images of each type of sub-phase structure of the rock specimen; The segmented images of each type of sub-phase structure of the rock specimen are encoded, and a data set for training the generative adversarial network is constructed based on the encoded segmented images.

[0032] Rock specimens with certain standard sizes are processed and manufactured. In this embodiment, standard sandstone thin section specimens with a length, width and height of 20 mm, 20 mm and 2 mm respectively are manufactured. High-precision modern imaging technology such as EBSD (Electron Backscatter Diffraction, scanning electron microscope) imaging technology is used to perform high-resolution digital imaging of the microstructure of the sandstone material. In this embodiment, the resolution of the obtained microstructure image is 5 μm, and the corresponding microscopic multiphase image is 500 pixels and the size is 500×500.

[0033] It should be noted that in the present embodiment, when obtaining microscopic multiphase images, there is no need to divide the rock specimen into layers as much as possible. To obtain the microscopic multiphase image of each layer of the rock specimen, only a small number of images are required to complete the three-dimensional reconstruction of the rock.

[0034] Specifically, in this embodiment, a microscopic structure image is obtained from the top, middle and bottom of the rock sample respectively, and the three microscopic images are used as the first microscopic image sequence.

[0035] In other feasible implementations, the rock sample may be divided into two parts, an upper part and an lower part, and a microstructure image is obtained from each of the upper part and the lower part, and the obtained two images are used as the first microscopic image sequence.

[0036] On this basis, each microscopic multiphase image is segmented to obtain the segmented image of each type of subphase structure of the rock sample in each image, such as Figure 2 Since sandstone is used as the rock sample in this embodiment, the sub-phase structure of the rock sample includes Figure 2 The pores shown in (c) Figure 2 The feldspar shown in (d) Figure 2 The quartz shown in (e) and Figure 2 Muscovite shown in (f).

[0037] That is to say, for each microstructure image, it is segmented to obtain segmented images of these four types of sub-phase structures.

[0038] Furthermore, if Figure 2 As shown in the figure, the segmentation structure of each sub-phase corresponding to each microstructure is encoded as the label of the image, and a dataset for training MGAN is constructed based on the encoded image, so that the image output by MGAN also contains the encoding of each sub-phase structure, such as Figure 3 As shown in (c) in .

[0039] In the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning of the present invention, before the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, the method further includes: The segmented image corresponding to each image in the first microscopic image sequence is coded with numbers and mineral names according to a preset order; The segmented images with numbers and mineral name codes corresponding to each image in the first microscopic image sequence are merged, and the merged images are organized into a data set for training the generative adversarial network.

[0040] Optionally, the segmented images are coded with numbers and mineral names in a clockwise order from top to bottom as a preset order.

[0041] Code, for example, 0-pore structure, 1-feldspar, 2-quartz, 3-muscovite.

[0042] Furthermore, for each microstructure image, all segmented images with numbers and mineral code names are merged to obtain microstructure images encoded with numbers and mineral names, and used as the dataset for training MGAN, such as Figure 3 As shown in (a) in .

[0043] In the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning of the present invention, the step of segmenting the images in the first microscopic image sequence to obtain the segmented images of each type of sub-phase structure of the rock specimen specifically includes: Inputting the images in the first microscopic image sequence into a pre-constructed deep learning segmentation model to obtain a segmentation image of each type of sub-phase structure of the rock specimen output by the deep learning segmentation model; Among them, the deep learning segmentation model constructs a loss function based on the rock image microstructure target loss, the real target and predicted target boundary compensation loss and the rock microstructure fractal dimension loss, and is trained using the mineral boundary difference as a low-rank fine-tuning parameter.

[0044] In order to achieve accurate identification and segmentation of each type of sub-phase structure, a deep learning segmentation model (MFD-UNet) based on multi-phase feature loss function and low-rank fine-tuning parameters is pre-constructed in this embodiment.

[0045] Specifically, the multiphase feature loss function is composed of the weighted average of the rock image microstructure target loss, the boundary compensation loss between the real target and the predicted target, and the rock microstructure fractal dimension loss: ; In the formula, is the total loss function, is the target loss function of the rock image microstructure, Compensate the loss function for the boundary between the real target and the predicted target, is the fractal dimension loss function of rock microstructure; , and They are the weight values ​​of the three types of loss functions.

[0046] To predict the microstructure prototype coefficient, is the true microstructure prototype coefficient, To predict the microstructure boundary prototype coefficient, is the prototype coefficient of the real microstructure boundary.

[0047] In order to further improve the segmentation accuracy, the parameters of MFD-UNet are fine-tuned using the mineral boundary difference as a low-rank fine-tuning parameter: ; In the formula, represents the low-rank fine-tuning parameter, N represents the image dimension, represents the prediction boundary, Indicates the true boundary.

[0048] In the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning of the present invention, the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen specifically includes: Constructing a generator of the generative adversarial network based on the fractal dimension function of the rock specimen; Constructing a discriminator of the generative adversarial network based on the difference between the generated distribution and the true distribution; A gradient penalty function of the discriminator is constructed based on the boundary similarity parameters.

[0049] In order to generate a more accurate two-dimensional rock image based on a small amount of rock microstructure images, in this embodiment, a generative adversarial network based on a fractal dimension feature pyramid algorithm is constructed.

[0050] Specifically, the generator of the generative adversarial network is constructed based on the fractal dimension function of the rock specimen. , construct a discriminator based on the difference between the generated distribution and the true distribution , construct a gradient penalty function based on boundary similarity parameters : ; In the formula, PDF is the conditional probability density, MP fd represents the fractal dimension function, N s represents the number of similar objects, S represents the scaling factor, E For expectations, K is the gradient penalty function, P gdata Generates distributions for rock microstructures, P tdata is the true distribution, λ is the gradient penalty function, is the discriminator gradient, is the L2 norm.

[0051] Through the above method, a generalized adversarial deep learning network (MGAN) for multi-layer image generation based on the fractal dimension function feature pyramid algorithm (FPN) can be constructed. A dataset with digital and mineral name codes is used as input parameters, and hyperparameters such as the batch size, learning rate, and number of iterations of the MGAN model are adjusted to train the MGAN.

[0052] In the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning of the present invention, the preset accuracy conditions include: generalization evaluation index, fractal dimension standard deviation and accuracy rate; When the generalization evaluation index is less than a first preset threshold, the fractal dimension standard deviation is not greater than a second preset threshold, and the accuracy is not less than a third preset threshold, it is considered that the two-dimensional image of the rock sample meets the preset accuracy conditions.

[0053] In this implementation, in order to obtain the optimal MGAN model with the best generalization ability, the generalization evaluation index, fractal dimension standard deviation and accuracy are used as preset accuracy conditions to evaluate the generation accuracy of the MGAN model obtained by preliminary training: ; In the formula, F precision is the generalization evaluation index (the closer it is to 0, the better the generalization performance), STD fd is the standard deviation of the fractal dimension, is the accuracy, TP, TN, FP and FN are true positives, true negatives, false positives and false negatives, respectively.

[0054] The rock microstructure dataset with numbers and mineral name codes (such as 0-pore structure, 1-feldspar, 2-quartz, 3-muscovite) was used as the input parameters of the training set (30%), test set and validation set (70%) according to the division ratio. The MGAN model was trained by adjusting the hyperparameters such as the batch size (2), learning rate (0.98) and number of iterations (100). When it is close to 0, that is, less than the first preset threshold determined by experience, it is considered that MGAN has the best generalization ability. is not greater than a second preset threshold value, which is 0.1 in this embodiment, and When the value is not less than the third preset threshold value, which is 0.95 in this embodiment, the model accuracy is considered to be the highest.

[0055] On this basis, the final MGAN model is used to generate a two-dimensional microstructure feature image of sequential multi-layer rocks, and a three-dimensional model of the rock is rendered, such as Figure 4 shown.

[0056] It should be noted that the MGAN trained on the rock microstructure dataset with numbers and mineral name codes also outputs two-dimensional images with numbers and mineral name codes, such as Figure 3 As shown in (c), on this basis, the three-dimensional model is rendered based on the output two-dimensional image, as shown in Figure 4 In (a), the three-dimensional model of each subphase structure can also be obtained by extracting the code, such as Figure 4 As shown in (b), (c), (d) and (e).

[0057] The rock multi-mineral microscopic implicit three-dimensional reconstruction system based on deep learning provided by the present invention is described below. The rock multi-mineral microscopic implicit three-dimensional reconstruction system based on deep learning described below and the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning described above can be referenced to each other.

[0058] like Figure 5 As shown, the rock multi-mineral microscopic implicit 3D reconstruction system based on deep learning includes an acquisition module 501, a generation module 502 and a reconstruction module 503: An acquisition module 501 is used to acquire a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; Based on the processed rock specimens, high-precision modern imaging technology is used to perform high-resolution digital imaging of the microstructure of the rock material to obtain a first microscopic image sequence, wherein the first microscopic image sequence contains a small amount of rock microscopic multiphase images. In this embodiment, one image is collected from each of the upper, middle and lower layers of the rock sample, and the obtained three images are used as the first microscopic image sequence.

[0059] Optionally, each image in the first microscopic image sequence is segmented to obtain a microscopic image of each sub-type of material, and then based on the microscopic image of each type of sub-phase structure, the fractal dimension function of the rock specimen is obtained.

[0060] A generation module 502 is used to construct a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generate a plurality of two-dimensional images of the rock specimen through the trained generative adversarial network; It can be understood that the fractal dimension function of the rock specimen characterizes the shape and distribution of the sub-phase structure of the rock specimen. Therefore, a generative adversarial network (MGAN) based on the fractal dimension feature pyramid algorithm is constructed and trained based on the fractal dimension function of the rock specimen.

[0061] That is to integrate the pyramid algorithm into the traditional generative adversarial network (GAN), and capture the structural complexity of image data at various resolutions by calculating the fractal dimension functions at different scales. This makes the images generated by the MGAN network trained based on a small number of images similar to the fractal dimension of the microstructure images of rock specimens, and enables the trained MGAN to generate two-dimensional images of rocks with higher accuracy.

[0062] Optionally, during the training process, a GAN network with the best generalization ability is selected as the trained GAN model.

[0063] By training a generative adversarial network to generate multiple two-dimensional images of rocks, the generated images have a fractal dimension function that is closer to the rock specimens, and can more realistically reflect the structural characteristics of the rock specimens, so that they can be used to study the microscopic mechanical properties and fracture mechanisms of rocks.

[0064] The reconstruction module 503 is used to reconstruct a three-dimensional rock model based on the two-dimensional images of the multiple rock samples when the two-dimensional images of the multiple rock samples meet a preset accuracy condition.

[0065] In order to make the two-dimensional images of rocks generated by GAN reflect the microstructure of rock specimens more realistically, after the GAN training is completed, its accuracy needs to be evaluated through pre-set preset accuracy conditions.

[0066] Optionally, the preset accuracy condition includes a generalization ability evaluation condition and / or a model accuracy evaluation condition. The threshold of each evaluation condition can be determined based on experience, and is usually set to be greater than 90%, such as setting the model accuracy to be greater than 90%.

[0067] When the two-dimensional image of the rock generated by GAN meets the preset accuracy conditions, the generative adversarial network at this time is considered to be the optimal network with the best generalization ability. Therefore, based on the generative adversarial network at this time, it is possible to sequentially generate multiple layers of two-dimensional images of the rock, and then render them into a three-dimensional model of the rock. Among them, the two-dimensional image of the rock is a characteristic image of the rock microstructure.

[0068] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute a rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning, the method comprising: obtaining a fractal dimension function of the rock specimen based on a first microscopic image sequence of the rock specimen, wherein the first microscopic image sequence represents a small number of images obtained based on the rock specimen; constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, generating multiple two-dimensional images of the rock through the trained generative adversarial network; and reconstructing a rock three-dimensional model based on the two-dimensional images of the multiple rock specimens when the two-dimensional images of the multiple rock specimens meet the preset accuracy conditions.

[0069] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0070] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the rock multi-mineral microscopic implicit three-dimensional reconstruction method based on deep learning provided by the above-mentioned methods, and the method includes: obtaining a fractal dimension function of the rock specimen based on a first microscopic image sequence of the rock specimen, wherein the first microscopic image sequence represents a small number of images obtained based on the rock specimen; constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generating multiple two-dimensional images of rocks through the trained generative adversarial network; when the two-dimensional images of the multiple rocks meet the preset accuracy conditions, reconstructing a three-dimensional rock model based on the two-dimensional images of the multiple rock samples.

[0071] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the deep learning-based rock multi-mineral microscopic implicit three-dimensional reconstruction method provided by the above-mentioned methods, the method comprising: obtaining a fractal dimension function of the rock specimen based on a first microscopic image sequence of the rock specimen, wherein the first microscopic image sequence represents a small number of images obtained based on the rock specimen; constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generating multiple two-dimensional images of rocks through the trained generative adversarial network; and reconstructing a three-dimensional rock model based on the two-dimensional images of the multiple rock specimens when the two-dimensional images of the multiple rock specimens meet preset accuracy conditions.

[0072] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0073] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning, characterized in that: include: Acquire a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; Based on the fractal dimension function of the rock specimen, a generative adversarial network based on a fractal dimension feature pyramid algorithm is constructed, and a plurality of two-dimensional images of the rock are generated by the trained generative adversarial network; In the case that the multiple two-dimensional images of the rocks meet the preset accuracy conditions, a three-dimensional model of the rock is reconstructed based on the multiple two-dimensional images of the rock samples.

2. The rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning according to claim 1 is characterized in that: Before the step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, the method further includes: Collecting microscopic multiphase images of the rock specimen to obtain the first microscopic image sequence; Segmenting the images in the first microscopic image sequence to obtain segmented images of each type of sub-phase structure of the rock specimen; The segmented images of each type of sub-phase structure of the rock specimen are encoded, and a data set for training the generative adversarial network is constructed based on the encoded segmented images.

3. The rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning according to claim 2 is characterized in that: The step of segmenting the images in the first microscopic image sequence to obtain the segmented images of each type of sub-phase structure of the rock specimen specifically includes: Inputting the images in the first microscopic image sequence into a pre-constructed deep learning segmentation model to obtain a segmentation image of each type of sub-phase structure of the rock specimen output by the deep learning segmentation model; Among them, the deep learning segmentation model constructs a loss function based on the rock image microstructure target loss, the real target and predicted target boundary compensation loss and the rock microstructure fractal dimension loss, and is trained using the mineral boundary difference as a low-rank fine-tuning parameter.

4. The rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning according to claim 2 is characterized in that: The step of encoding the segmented image of each type of sub-phase structure of the rock specimen and constructing a data set for training the generative adversarial network based on the encoded segmented image specifically includes: The segmented image corresponding to each image in the first microscopic image sequence is coded with numbers and mineral names according to a preset order; The segmented images with numbers and mineral name codes corresponding to each image in the first microscopic image sequence are merged, and the merged images are organized into a data set for training the generative adversarial network.

5. The rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning according to any one of claims 1 to 4, characterized in that: The step of constructing a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen specifically includes: Constructing a generator of the generative adversarial network based on the fractal dimension function of the rock specimen; Constructing a discriminator of the generative adversarial network based on the difference between the generated distribution and the true distribution; A gradient penalty function of the discriminator is constructed based on the boundary similarity parameters.

6. The rock multi-mineral microscopic implicit 3D reconstruction method based on deep learning according to any one of claims 1 to 4, characterized in that: The preset accuracy conditions include: generalization evaluation index, fractal dimension standard deviation and accuracy; When the generalization evaluation index is less than a first preset threshold, the fractal dimension standard deviation is not greater than a second preset threshold, and the accuracy is not less than a third preset threshold, it is considered that the two-dimensional image of the rock sample meets the preset accuracy conditions.

7. A rock multi-mineral microscopic implicit three-dimensional reconstruction system, characterized in that: include: An acquisition module, configured to acquire a fractal dimension function of the rock sample based on a first microscopic image sequence of the rock sample, wherein the first microscopic image sequence represents a small number of images acquired based on the rock sample; A generation module is used to construct a generative adversarial network based on a fractal dimension feature pyramid algorithm based on the fractal dimension function of the rock specimen, and generate a plurality of two-dimensional images of the rock specimen through the trained generative adversarial network; The reconstruction module is used to reconstruct a three-dimensional rock model based on the two-dimensional images of the multiple rock samples when the two-dimensional images of the multiple rock samples meet a preset accuracy condition.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the rock multi-mineral microscopic implicit three-dimensional reconstruction method as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the rock multi-mineral microscopic implicit three-dimensional reconstruction method as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the rock multi-mineral microscopic implicit three-dimensional reconstruction method as claimed in any one of claims 1 to 6 is implemented.

Images