Shale scanning electron microscope microstructure super-resolution imaging method, device and equipment

Abstract

The shale scanning electron microscope microstructure super-resolution imaging method, device and equipment are disclosed, including: obtaining a plurality of first high-resolution images of shale, the first high-resolution image is obtained by scanning electron microscope scanning on the shale; input each first high-resolution image into the degradation model, simulate the physical effect of fast imaging through the degradation model, generate the first low-resolution image corresponding to each first high-resolution image; according to a plurality of first high-resolution images and the first low-resolution image corresponding to each first high-resolution image, construct a resolution training set; through the resolution training set, take the first low-resolution image corresponding to each first high-resolution image as the input, take each first high-resolution image as the output, train the initial generative machine learning model, obtain the target generative machine learning model supporting the reconstruction of the lost microstructure of shale in the fast imaging process. To improve the generation efficiency of high-resolution images of shale.

Description

Technical Field

[0001] This application relates to the field of geological sample analysis, and in particular to a method, apparatus and equipment for super-resolution imaging of the microstructure of shale using scanning electron microscopy. Background Technology

[0002] The microstructure of shale, such as the morphology of nanopores, micropores, and microfractures in the matrix, directly determines the energy storage and permeability of shale. Precise characterization of these microstructures is a crucial foundation for fields such as oil and gas exploration, geological energy storage, and groundwater research.

[0003] Currently, high-resolution imaging techniques such as scanning electron microscopy and focused ion beam scanning electron microscopy are mainly used to directly obtain high-resolution microstructure images of shale, but these techniques suffer from high equipment costs and low imaging efficiency. Summary of the Invention

[0004] The main objective of this application is to propose a method, apparatus, and device for super-resolution imaging of the microstructure of shale using scanning electron microscopy, aiming to solve the problem of low efficiency in super-resolution imaging of the microstructure of shale using scanning electron microscopy.

[0005] To achieve the above objectives, a first aspect of this application proposes a method for super-resolution imaging of the microstructure of shale using scanning electron microscopy, the method comprising:

[0006] Multiple first high-resolution images of shale are acquired; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope; Each of the first high-resolution images is input into a degradation model, which simulates the physical effects of rapid imaging to generate a first low-resolution image corresponding to each of the first high-resolution images; the low-resolution image is an image with a resolution lower than the preset resolution. A resolution training set is constructed based on the plurality of first high-resolution images and the first low-resolution images corresponding to each first high-resolution image; Using the resolution training set, with the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, the initial generative machine learning model is trained to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

[0007] In some implementations, generating a first low-resolution image corresponding to each of the first high-resolution images by simulating the physical effects of rapid imaging using the degradation model includes: The physical effects of rapid imaging are simulated by the degradation model, and target processing is performed on each of the first high-resolution images to obtain the first low-resolution image corresponding to each of the first high-resolution images. The target processing includes at least one of the following: blurring, downsampling, and noise addition.

[0008] In some implementations, performing target processing on each of the first high-resolution images includes at least one of the following: The first high-resolution image is blurred by simulating Gaussian blur of the point spread function at low resolution. Perform downsampling processing on each of the first high-resolution images; Noise-adding processing is performed on each of the first high-resolution images by simulating mixed noise with a low signal-to-noise ratio during rapid imaging.

[0009] In some implementations, the initial generative machine learning model includes a generator network and a discriminator network; The step of training an initial generative machine learning model using the resolution training set, taking the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process, includes: The generator network converts the first low-resolution image in the resolution training set into a corresponding second high-resolution image. The discriminator network is used to distinguish between the second high-resolution image and the first high-resolution image corresponding to the resolution training set, respectively, to obtain the discrimination scores of the second high-resolution image and the first high-resolution image; the discrimination scores are used to indicate the authenticity of the image. Based on the discrimination scores of the second high-resolution image and the first high-resolution image, and the second high-resolution image, the loss value of the initial generative machine learning model is calculated using the target loss function; If the loss value is greater than a preset threshold, the initial generative machine learning model is adjusted using backpropagation based on the loss value until the loss value is less than or equal to the preset threshold, thereby obtaining the target generative machine learning model.

[0010] In some implementations, the target loss function includes perceptual loss and relativistic adversarial loss; The perceptual loss is used to indicate the similarity in morphology and texture between the microstructures in the second high-resolution image output by the initial generative machine learning model and the first high-resolution image. The relativistic adversarial loss is used to indicate the difference in realism between the second high-resolution image output by the initial generative machine learning model and the first high-resolution image.

[0011] In some embodiments, after training the initial generative machine learning model using the resolution training set to obtain the target generative machine learning model, the method further includes: By using acquisition parameters that match the physical effects of rapid imaging simulated by the degradation model, images of the shale to be analyzed are acquired to obtain a target low-resolution image of the shale to be analyzed. The low-resolution image of the shale to be analyzed is input into the target generative machine learning model, and the target generative machine learning model outputs a target high-resolution image. Based on the target high-resolution image, quantitative analysis is performed on the shale to be analyzed to obtain the rock physical parameters of the shale to be analyzed; the rock physical parameters are used to characterize the microstructure of the shale to be analyzed.

[0012] In some embodiments, the step of performing quantitative analysis on the shale to be analyzed based on the target high-resolution image to obtain the rock physical parameters of the shale to be analyzed includes: The target high-resolution image is segmented to obtain a segmentation mask; the segmentation mask is used to identify the microstructure in the target high-resolution image. Based on the segmentation mask, feature parameters of the microstructure in the target high-resolution image are calculated to obtain the rock physical parameters of the shale to be analyzed.

[0013] To achieve the above objectives, a second aspect of this application provides a super-resolution imaging device for the microstructure of shale using scanning electron microscopy, the device comprising: The acquisition module is used to acquire multiple first high-resolution images of the shale; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope; The generation module is used to input each of the first high-resolution images into a degradation model, and to simulate the physical effects of rapid imaging through the degradation model to generate a first low-resolution image corresponding to each of the first high-resolution images; the low-resolution image is an image with a resolution lower than the preset resolution; A construction module is used to construct a resolution training set based on the plurality of first high-resolution images and the first low-resolution images corresponding to each of the first high-resolution images; The training module is used to train an initial generative machine learning model using the resolution training set, taking the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

[0014] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the shale scanning electron microscope microstructure super-resolution imaging method described in the first aspect.

[0015] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the shale scanning electron microscope microstructure super-resolution imaging method described in the first aspect.

[0016] To achieve the above objectives, embodiments of this application may provide a computer program product, wherein the instructions in the computer program product, when executed by the processor of an electronic device, cause the electronic device to implement the shale scanning electron microscope microstructure super-resolution imaging method described in the first aspect.

[0017] The method, apparatus, and equipment for super-resolution imaging of shale microstructure using scanning electron microscopy proposed in this application acquire multiple first high-resolution images of shale. These high-resolution images are those with a resolution greater than or equal to a preset resolution. The first high-resolution images are obtained by scanning the shale using a scanning electron microscope. Each first high-resolution image is input into a degradation model, which simulates the physical effects of rapid imaging to generate a first low-resolution image corresponding to each first high-resolution image. The low-resolution image is an image with a resolution less than the preset resolution. A resolution training set is constructed based on the multiple first high-resolution images and their corresponding first low-resolution images. Using this resolution training set, and with the first low-resolution images corresponding to each first high-resolution image as input and the first high-resolution images as output, an initial generative machine learning model is trained to obtain a target generative machine learning model that supports the reconstruction of the shale microstructure lost during rapid imaging. The resolution of the image output by the target generative machine learning model is the same as the resolution of the high-resolution image obtained by scanning electron microscopy. Therefore, by aligning the degradation model with the actual physical mechanisms of rapid imaging, the constructed resolution training set can accurately match real-world application scenarios. Furthermore, the trained target-generative machine learning model can specifically learn the microstructural features of shale and the degradation patterns of rapid imaging, effectively restoring key details such as pore morphology and connectivity. The reconstruction effect is superior to traditional image processing algorithms and general generative models. Thus, in practical applications, rapid imaging technology can be used to acquire low-resolution images of shale, and the target-generative machine learning model can then be used to reconstruct the microstructure of shale lost during rapid imaging, resulting in a super-resolution image of the shale. Moreover, the super-resolution image has the same resolution as the high-resolution image output by the scanning electron microscope, thereby significantly improving the generation efficiency of high-resolution shale images. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the super-resolution imaging method for the microstructure of shale using scanning electron microscopy provided in the embodiments of this application; Figure 2 This is a schematic diagram of the implementation process of the super-resolution imaging method for the microstructure of shale using scanning electron microscopy provided in the embodiments of this application; Figure 3 This is a schematic diagram of an application workflow for a new sample provided in an embodiment of this application; Figure 4 This is a comparison diagram of a low-resolution image and a virtual high-resolution image provided in an embodiment of this application; wherein, (a) is a low-resolution image and (b) is a virtual high-resolution image; Figure 5 This is a schematic diagram of a process for quantitative analysis of shale provided in an embodiment of this application; Figure 6This is a schematic diagram of the structure of the super-resolution imaging device for the microstructure of shale scanning electron microscope provided in the embodiments of this application; Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0021] 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 this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] The study of the microstructure of porous geological media such as sandstone, carbonate rocks, and shale is of great significance for oil and gas extraction, groundwater pollution control, and carbon dioxide geological sequestration. In order to accurately obtain key parameters such as pore distribution, porosity, and permeability of porous media, high-resolution imaging techniques such as nano-scanning (CT) and scanning electron microscopy (SEM) are usually required.

[0023] However, high-resolution imaging techniques typically suffer from limitations such as long imaging times, high equipment costs, and small fields of view. In contrast, rapid imaging techniques such as low-magnification microscopy and low-dose CT, while significantly improving imaging speed and expanding the observation range, are limited by physical imaging conditions and radiation dose, resulting in lower image resolution and the loss of a large amount of detailed information characterizing the microstructure of rocks. Traditional image super-resolution reconstruction methods, such as bicubic interpolation, rely solely on mathematical operations on pixel values, making it difficult to recover physically meaningful true geological microstructures, often leading to blurred image edges and loss of texture details.

[0024] Based on this, embodiments of this application provide a method, apparatus, and device for super-resolution imaging of shale scanning electron microscopy microstructures, aiming to solve the problem of low efficiency in super-resolution imaging of shale scanning electron microscopy microstructures.

[0025] The shale scanning electron microscope microstructure super-resolution imaging method, apparatus, and device provided in this application are specifically described through the following embodiments. First, the shale scanning electron microscope microstructure super-resolution imaging method in this application embodiment is described.

[0026] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0027] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0028] The shale scanning electron microscope (SEM) microstructure super-resolution imaging method, apparatus, and equipment provided in this application relate to the field of shale geological sample analysis. The shale SEM microstructure super-resolution imaging method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the shale SEM microstructure super-resolution imaging method, but is not limited to the above forms.

[0029] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0030] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0031] Figure 1 This is a flowchart illustrating the super-resolution imaging method for the microstructure of shale using scanning electron microscopy provided in this application. Please refer to [link / reference]. Figure 1 The super-resolution imaging method for the microstructure of shale using scanning electron microscopy provided in this application may include, but is not limited to, steps S101 to S104.

[0032] Step S101: Obtain multiple first high-resolution images of the shale; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope.

[0033] In this step, multiple first high-resolution images of the shale can be obtained by scanning the shale with a scanning electron microscope (SEM). The SEM can be a scanning electron microscope, a focused ion beam scanning electron microscope (FIE), etc., and no specific limitation is made here.

[0034] High-resolution images are images with a resolution greater than or equal to the preset resolution. The preset resolution can be set according to the needs of the actual application scenario, and no specific limitation is made here.

[0035] In some implementations, taking into account the performance of existing scanning electron microscopes, the preset resolution is preferably set to 1024×1024 pixels or higher, such as 2048×2048 pixels or 4096×4096 pixels, to ensure that the first high-resolution image can clearly present the microstructural details of shale, such as pore morphology, grain boundaries, and pore connectivity paths.

[0036] In some implementations, when imaging shale with a scanning electron microscope to obtain a high-resolution image, the high-resolution image can be segmented to obtain multiple first high-resolution images, thereby rapidly increasing the number of samples.

[0037] In other implementations, similarly, first high-resolution images of other porous media can be obtained for subsequent dataset construction and model training steps, resulting in a target generative machine learning model adapted to other porous geological media, thereby improving the generation efficiency of high-resolution images of other porous geological media. These other porous geological media may include sandstone, carbonate rocks, porous basalt, etc., without specific limitations here.

[0038] Step S102: Input each of the first high-resolution images into the degradation model, and use the degradation model to simulate the physical effects of fast imaging to generate a first low-resolution image corresponding to each of the first high-resolution images; the low-resolution image is an image with a resolution lower than the preset resolution.

[0039] In this step, each first high-resolution image is input into a degradation model. The degradation model simulates the physical effects of rapid imaging to degrade the first high-resolution images, generating a first low-resolution image that corresponds one-to-one with each first high-resolution image. The low-resolution image is an image with a resolution lower than a preset resolution, and its resolution can match the output resolution of actual rapid imaging devices (such as 256×256 pixels, 512×512 pixels), ensuring that the subsequent model training scenario is consistent with the actual application scenario.

[0040] The degradation model is used to simulate the physical effects of fast imaging devices during the imaging process. These physical effects are the core reasons for the reduced resolution and loss of microstructure in fast imaging images. Specifically, they can include blurring caused by the point spread function (PSF) of the optical system, downsampling caused by the pixel size limitation of the detector, and increased noise due to the shortened imaging time.

[0041] Furthermore, a degradation model can be constructed based on the physical characteristics of actual fast imaging equipment. That is, by analyzing the physical mechanism of the fast imaging process or fitting the difference between the real fast imaging low-resolution image and the corresponding high-resolution image, degradation parameters (such as blur kernel size, noise intensity, and sampling ratio) can be determined to ensure that the low-resolution image generated by the degradation model can truly reproduce the degradation effect of fast imaging.

[0042] The input to the degradation model is multiple first high-resolution images, and the output of the degradation model is the low-resolution image corresponding to each first high-resolution image.

[0043] Step S103: Construct a resolution training set based on the plurality of first high-resolution images and the first low-resolution images corresponding to each of the first high-resolution images.

[0044] In this step, a resolution training set is constructed based on multiple first high-resolution images and the first low-resolution images corresponding to each first high-resolution image.

[0045] The resolution training set consists of several image pairs, each containing a first low-resolution image and a corresponding first high-resolution image. The first low-resolution image serves as the input sample, and the first high-resolution image serves as the label sample, guiding the initial generative machine learning model to learn the mapping relationship from low-resolution images to high-resolution images.

[0046] Step S104: Using the resolution training set, with the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, train the initial generative machine learning model to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

[0047] In this step, the initial generative machine learning model can employ architectures such as generative adversarial networks (GANs), diffusion models, and generative models based on self-attention mechanisms (Transformers). Preferably, a GAN is used, and through adversarial training between the generator and discriminator in the GAN, images with rich detail and high fidelity can be generated.

[0048] During training, the initial generative machine learning model learns the feature distribution of shale, which not only restores the high-frequency details of the image, but also reconstructs the pore connectivity and particle morphology that conform to the geological and physical laws, thereby accurately restoring the microstructural information such as pore morphology, pore size distribution, and particle arrangement lost during rapid imaging.

[0049] In this implementation, the degradation model is aligned with the actual physical mechanism of rapid imaging, enabling the constructed resolution training set to accurately match the actual application scenario. Furthermore, the trained target-generative machine learning model can specifically learn the microstructural features of shale and the degradation law of rapid imaging, effectively restoring key details such as pore morphology and connectivity. The reconstruction effect is superior to traditional image processing algorithms and general generative models. Thus, in practical applications, low-resolution images of shale can be obtained using rapid imaging technology, and the target-generative machine learning model can then be used to reconstruct the microstructure of shale lost during rapid imaging, resulting in a super-resolution image of shale. Moreover, the super-resolution image has the same resolution as the high-resolution image output by the scanning electron microscope, thereby significantly improving the generation efficiency of high-resolution shale images.

[0050] In some implementations, step S102 may include the following step S201: Step S201: Simulate the physical effects of rapid imaging using the degradation model, perform target processing on each of the first high-resolution images, and obtain the first low-resolution image corresponding to each of the first high-resolution images; The target processing includes at least one of the following: blurring, downsampling, and noise addition.

[0051] Furthermore, the target processing performed on each of the first high-resolution images in step S201 may include at least one of the following: The first high-resolution image is blurred by simulating Gaussian blur of the point spread function at low resolution. Perform downsampling processing on each of the first high-resolution images; Noise-adding processing is performed on each of the first high-resolution images by simulating mixed noise with a low signal-to-noise ratio during rapid imaging.

[0052] In this implementation, the degradation model is used to reproduce the image quality degradation process caused by faster imaging speed, reduced radiation dose, or limited equipment precision.

[0053] Because image details can be smoothed due to the diffraction limit of the optical system or the limitations of the detector response speed during rapid imaging, target manipulation can include blurring to simulate this effect.

[0054] In some implementations, blurring can be performed on each first high-resolution image by simulating Gaussian blurring of the point spread function at low resolution. Specifically, a two-dimensional Gaussian kernel can be used to convolve the first high-resolution image. The standard deviation of the Gaussian kernel is proportional to the degree of blurring in the imaging system, and the standard deviation of the Gaussian kernel can be set according to a preset ratio between low and high resolution. For example, if simulating a 4x degradation, the corresponding Gaussian kernel size is set to simulate the imaging effect under large light spots, thereby eliminating high-frequency texture details in the image.

[0055] Because the detector pixel size of low-resolution imaging devices (such as low-magnification microscopes or low-resolution CT scanners) is typically larger than that of high-resolution devices, the number of sampling points per unit area is reduced. Therefore, target manipulation may include downsampling processing.

[0056] In some implementations, downsampling can be performed by downsampling the blurred image (or directly downsampling the first high-resolution image). Specifically, this can be achieved through methods such as bilinear interpolation, bicubic interpolation, or nearest-neighbor interpolation. For example, setting the downsampling factor to... k ( k If the integer is greater than 1 (such as 2, 4, or 8), then the image size will be reduced to 1 / 2 of its original size in both the length and width directions. k The total number of pixels is reduced to 1 / 3 of the original. k 2 This directly reduces the spatial resolution of the image.

[0057] Since rapid imaging often means shortening exposure time or reducing radiation dose, it results in insufficient signal strength, introducing a large amount of noise, i.e., a low signal-to-noise ratio. Therefore, target manipulation may include noise addition processing.

[0058] In some implementations, noise addition processing can be performed on each first high-resolution image by simulating mixed noise with a low signal-to-noise ratio during rapid imaging. Specifically, a mixed noise model is superimposed on each first high-resolution image. The mixed noise can consist of Gaussian noise and Poisson noise. Gaussian noise is used to simulate the thermal noise of electronic circuits, and its amplitude follows a Gaussian distribution. Poisson noise is used to simulate the random fluctuations (shot noise) of photon or electron impact detectors, and its intensity is related to the intensity of the signal itself. The intensity parameter of the noise can be adjusted according to a preset signal-to-noise ratio (SNR) value. For example, setting a low SNR value (such as 20 dB or lower) makes the generated first low-resolution image exhibit obvious graininess and texture coarseness, realistically reflecting the image quality under rapid imaging conditions.

[0059] In a preferred embodiment, the target processing sequentially includes: first, performing Gaussian blurring on the first high-resolution image to simulate the physical limitations of the optical system; then, downsampling the blurred image to reduce pixel density; and finally, adding mixed noise to the downsampled image to simulate a low signal-to-noise ratio environment. After these three steps, the generated first low-resolution image can maximally reproduce the image features output by a fast imaging device in a real-world scene, providing a reliable data foundation for subsequent training of a high-fidelity generative machine learning model.

[0060] It should be noted that the order of the three target processes is not limited in this application embodiment. They can be processed according to the actual situation. The combination of the three target processes is also not limited. The purpose of the degradation model in this application embodiment is to handle specific types of degradation found in low-cost data in the real world.

[0061] In some implementations, the initial generative machine learning model includes a generator network and a discriminator network.

[0062] Step S104 may include the following steps S301 to S304: Step S301: Convert the first low-resolution image in the resolution training set into the corresponding second high-resolution image through the generator network; Step S302: The discriminator network is used to distinguish between the second high-resolution image and the first high-resolution image corresponding to the resolution training set, respectively, to obtain the discrimination scores of the second high-resolution image and the first high-resolution image; the discrimination scores are used to indicate the authenticity of the image. Step S303: Based on the discrimination scores of the second high-resolution image and the first high-resolution image, and the second high-resolution image, calculate the loss value of the initial generative machine learning model using the target loss function; Step S304: If the loss value is greater than a preset threshold, the initial generative machine learning model is adjusted according to the loss value using backpropagation until the loss value is less than or equal to the preset threshold, thereby obtaining the target generative machine learning model.

[0063] In some implementations, the target loss function includes perceptual loss and relativistic adversarial loss; The perceptual loss is used to indicate the similarity in morphology and texture between the microstructures in the second high-resolution image output by the initial generative machine learning model and the first high-resolution image. The relativistic adversarial loss is used to indicate the difference in realism between the second high-resolution image output by the initial generative machine learning model and the first high-resolution image.

[0064] In this implementation, the initial generative machine learning model can adopt a generative adversarial network (GAN) architecture, which includes a generator network and a discriminator network. The two networks are trained through adversarial learning to improve the target generative machine learning model's ability to reconstruct the microstructure of shale.

[0065] In some implementations, a generator network is used to process the data in the resolution training set. The first low-resolution image in the resolution training set (i.e., a blurry, noisy image obtained by simulating fast imaging) is input to the generator network. The generator network extracts features from the first low-resolution image and attempts to recover lost high-frequency details through a series of convolutions, residual connections, and upsampling operations, outputting a transformed second high-resolution image. At this point, the second high-resolution image is a high-resolution version "guessed" by the generator, and its quality is usually poor in the early stages of training.

[0066] Subsequently, a discriminator network is used to evaluate the realism of the images. The discriminator network receives two types of images as input: a second-high resolution image generated by the generator network, and a first-high resolution image (Ground Truth) from the resolution training set. The discriminator network extracts features and classifies these two types of images, outputting a discrimination score for each. This discrimination score indicates the degree of realism of the image or the probability that it belongs to a real image. For example, a score closer to 1 indicates that the discriminator considers the image more like a real high-resolution image, while a score closer to 0 indicates that the discriminator considers the image to be fake or of substandard quality.

[0067] Based on the data obtained from the above steps, the total loss value of the initial generative machine learning model is calculated using the objective loss function. The objective loss function can include perceptual loss and relativistic adversarial loss.

[0068] Specifically, the perceptual loss is used to indicate the morphological and textural similarity of the microstructures in the second high-resolution image output by the initial generative machine learning model to the first high-resolution image. Instead of directly comparing pixel values, the perceptual loss compares the image representations in a high-level feature space. Furthermore, a deep convolutional neural network, such as the VGG-19 network, pre-trained on existing large image datasets like ImageNet can be used as a feature extractor. The generated second high-resolution image and the real first high-resolution image are respectively input into this pre-trained network to extract feature maps from specific hidden layers, such as ReLU activation layers. The distance between these two sets of feature maps (which can be Euclidean distance) is calculated as the perceptual loss value. By minimizing the perceptual loss, the model is forced to recover texture information relevant to human visual perception, as well as the edge morphology of rock grains and pores, thereby reconstructing a geological image with clear microstructures.

[0069] Relativistic adversarial loss is used to indicate the difference in realism between the second high-resolution image output by the initial generative machine learning model and the first high-resolution image. Traditional adversarial losses typically only consider the probability of a sample being "real" or "fake," while relativistic adversarial loss attempts to estimate the probability that "the real image is more real than the fake image." Specifically, the discriminator network outputs not only a decision score but also a relative comparison between the generated pair of images (a real first high-resolution image and a generated second high-resolution image). For the discriminator, the goal is to maximize the probability that the real image is real relative to the generated image. For the generator, the goal is to minimize the probability that the generated image is fake relative to the real image, i.e., to make the discriminator consider the generated image to be more real than the real image, or at least indistinguishable from it. By providing the generator with richer gradient information through relativistic adversarial loss, the vanishing gradient or mode collapse problems commonly encountered in traditional adversarial network (GAN) training are effectively mitigated. This results in the generated second high-resolution image having sharper texture details and a more natural overall image, making it difficult for humans or algorithms to distinguish.

[0070] In some implementations, the final loss function value can be a weighted sum of the perceptual loss and the relativistic adversarial loss. The weights of the perceptual loss and the relativistic adversarial loss can be set according to the actual situation, and no specific restrictions are made here.

[0071] Furthermore, after calculating the loss value, it is determined whether the loss value is greater than a preset threshold to ascertain whether the initial generative machine learning model has converged. Specifically, if the loss value is greater than the preset threshold, it indicates that the initial generative machine learning model has not been sufficiently trained. At this point, based on the loss value, the gradient is calculated using backpropagation (such as gradient descent), and the network parameters (weights and biases) of the generator and discriminator networks are updated. Subsequently, the above steps are repeated using the next batch of training data for the next round of training.

[0072] Training continues until the loss value is less than or equal to a preset threshold, indicating that the model has converged and the generator network has the ability to reconstruct high-quality microstructures from low-resolution images. At this point, training is stopped, and the trained generator network and related parameters are fixed, thus obtaining the target generative machine learning model.

[0073] The preset threshold can be set according to the actual situation, and no specific limit is set here, such as 0.001.

[0074] In this implementation, by minimizing the target loss function, the initial generative machine learning model can significantly improve the realism of the image while maintaining the accuracy of the microstructure and texture of the geological medium, thereby obtaining a target generative machine learning model that can accurately reconstruct details lost during rapid imaging.

[0075] In some embodiments, after step S104, the shale scanning electron microscope microstructure super-resolution imaging method provided in this application embodiment may further include the following steps S401 to S403: Step S401: By using acquisition parameters that match the physical effects of rapid imaging simulated by the degradation model, image acquisition is performed on the shale to be analyzed to obtain a target low-resolution image of the shale to be analyzed. Step S402: Input the low-resolution image of the shale to be analyzed into the target generative machine learning model, and output the target high-resolution image through the target generative machine learning model; Step S403: Based on the target high-resolution image, perform quantitative analysis on the shale to be analyzed to obtain the rock physical parameters of the shale to be analyzed; the rock physical parameters are used to characterize the microstructure of the shale to be analyzed.

[0076] Step S403 may also include the following: The target high-resolution image is segmented to obtain a segmentation mask; the segmentation mask is used to identify the microstructure in the target high-resolution image. Based on the segmentation mask, feature parameters of the microstructure in the target high-resolution image are calculated to obtain the rock physical parameters of the shale to be analyzed.

[0077] In this implementation, after training the initial generative machine learning model with a resolution training set to obtain a high-precision target generative machine learning model, the target generative machine learning model can be applied to actual geological exploration or core analysis.

[0078] Specifically, when imaging the shale to be analyzed (such as newly drilled core samples from deep underground), a rapid imaging strategy is adopted to improve detection efficiency.

[0079] The equipment parameters used for image acquisition (such as voltage, current, and exposure time for CT scans, or magnification for microscopes) must match the physical effects simulated by the previous degradation model. For example, if the degradation model simulates a fast imaging effect with a resolution of 1 / 4 of the original image, then in actual acquisition, the resolution of the imaging equipment should be set to the corresponding low-resolution mode, or the exposure time should be shortened to produce a similar signal-to-noise ratio.

[0080] By scanning the sample using the parameters described above, a low-resolution image of the target, containing the overall structure of the sample but with blurred details, can be quickly obtained. Thanks to the use of rapid imaging parameters, the time cost of this process is significantly reduced, enabling rapid screening of a large number of samples.

[0081] The low-resolution images of the shale to be analyzed are input into the previously trained target generative machine learning model. Internally, the target generative machine learning model, through feature extraction and mapping, can identify blurry patterns in the low-resolution images and reconstruct clear details based on prior knowledge of the shale microstructure learned during training. The target generative machine learning model outputs the corresponding high-resolution target image. The high-resolution target image is close to the effect obtained by using high-precision equipment for long-term slow scanning in terms of resolution and texture detail, and contains key microstructural information lost during rapid imaging, such as tiny pore throats and particle surface roughness.

[0082] After obtaining a high-resolution image of the target, image processing and quantitative analysis are performed on the high-resolution image of the target to extract parameters characterizing the physical properties of the rock.

[0083] Furthermore, the target high-resolution image can be segmented to distinguish different components in the image, for example, separating regions representing pore space (typically darker) from regions representing rock skeleton or granular matrix (typically brighter).

[0084] When segmenting a high-resolution target image, image processing algorithms can be used, such as adaptive thresholding, watershed algorithms, or edge detection-based segmentation algorithms. Alternatively, deep learning-based semantic segmentation networks (such as U-Net) or multi-task deep learning models (such as Mask R-CNN) can be employed. Preferably, deep learning-based semantic segmentation networks (such as U-Net) or multi-task deep learning models (such as Mask R-CNN) are used.

[0085] After segmentation, a segmentation mask corresponding to the original image size is generated. In the segmentation mask, different pixel values ​​are assigned specific meanings; for example, a pixel value of "1" represents a pore region, and a pixel value of "0" represents a rock skeleton region (or vice versa). This segmentation mask accurately identifies the microstructure in the target high-resolution image, clearly outlining the shape, size, and spatial distribution of the pores.

[0086] After obtaining the segmentation mask, it is used as the basic data to calculate the feature parameters of the microstructure in the high-resolution image of the target, thereby obtaining quantitative rock physical parameters. Rock physical parameters may include the porosity of inorganic and organic pores, the area ratio of organic matter, and the pore size distribution.

[0087] It should be noted that the physical parameters of the rock can be calculated according to the actual situation, and no specific limitations are made here.

[0088] In this implementation, compared with traditional high-resolution imaging methods, this embodiment significantly shortens the image acquisition time and reduces the dependence on expensive high-precision equipment. At the same time, since the target generative machine learning model can accurately reconstruct the microstructure, the calculated rock physical parameters have high accuracy and reliability, which can effectively guide geological exploration and oil and gas development work.

[0089] The following describes the implementation process of the super-resolution imaging method for the microstructure of shale using scanning electron microscopy provided in this application: Step 210: Acquire a set of reference high-resolution (HR) images. This can be accomplished through detailed, costly analysis of a limited number of representative geological samples from the target strata. For example, using scanning electron microscopy (SEM), several large-area, high-magnification image mosaics (MAPS) with high resolution (e.g., 10 nm / pixel) are acquired on an ion-beam polished shale surface. These MAPS images are then divided into a large number of smaller HR image patches, such as 50,000 patches (i.e., the first high-resolution images).

[0090] Step 220 involves synthesizing a set of corresponding low-resolution (LR) images (i.e., the first low-resolution images) from the HR images. This step is crucial, going beyond simple image blurring. A physically realistic degradation model is applied to each HR image patch to simulate artifacts and features of fast, low-cost SEM imaging. This degradation model comprises a series of operations: (i) applying Gaussian blur to simulate the point spread function of SEM at lower resolutions; (ii) downsampling the image to the target low resolution (e.g., 100 nm / pixel); and (iii) adding a physically representative noise model, such as a combination of Poisson and Gaussian noise, to simulate the low signal-to-noise ratio during rapid scanning. This ensures that the model is trained to handle the specific types of degradation found in real-world, low-cost data.

[0091] In step 230, a generative machine learning model (i.e., the initial generative machine learning model) is trained using paired LR-HR image patches. This model is preferably a Generative Adversarial Network (GAN), more preferably an Enhanced Super-Resolution GAN (ESRGAN) architecture, which is well-suited for generating realistic textures. The ESRGAN consists of a generator network and a discriminator network. The generator, preferably employing a deep architecture with Residual Dense Blocks (RRDB) in the residuals and omitting batch normalization layers to improve stability and generalization, learns to convert LR input images into virtual high-resolution (vHR) images. The discriminator learns to distinguish the generator's output from real HR images. These two networks are trained adversarially, with the generator aiming to produce vHR images that the discriminator cannot distinguish from real HR images. The training process utilizes a composite loss function, including perceptual loss and relativistic adversarial loss, to guide the generator to produce geologically realistic and detailed microstructures. The trained model, representing the mapping from LR to HR, is stored in machine learning module 116.

[0092] Reference Figure 3 This demonstrates the application workflow for analyzing new, unknown samples. The process is designed to be fast and low-cost.

[0093] Step 310: Using SEM, prepare and image one or more new geological samples (e.g., drill cuttings) under rapid and low-cost conditions to obtain one or more new LR images. Acquisition parameters (e.g., resolution, dwell time) are normalized to match the parameters simulated by the deterioration model used in training.

[0094] In step 320, a new LR image is fed as input into the trained generative model (i.e., the objective generative machine learning model) stored in the machine learning module. The model processes the LR input and, in step 330, outputs a corresponding vHR image.

[0095] like Figure 4 As shown in (a), this is an LR input image. Due to the low resolution, fine features become blurred or completely missing. Figure 4 (b) shows the corresponding vHR output image generated by the model. The vHR image exhibits significantly enhanced detail, including the reconstruction of nanoscale features (such as organic pores and microcracks) that were not resolved in the LR image.

[0096] Step 340 involves quantitative analysis of the vHR images to extract one or more rock physical parameters. For example... Figure 5As shown in the detailed flowchart, this analysis process does not rely on a simple thresholding method, as thresholding is sensitive to noise and artifacts. Instead, the vHR images are processed by a robust segmentation model, such as a U-Net or Mask R-CNN architecture, which has been trained separately on manually labeled HR images to perform semantic or instance segmentation. This segmentation model identifies and delineates different phases and features in the image. Based on the segmentation map, this module calculates key parameters such as total porosity, organic matter area fraction, pore size distribution, and fracture aperture. The calculated parameters are output for reservoir modeling and evaluation.

[0097] The following experiments illustrate the effectiveness of the super-resolution imaging method for shale scanning electron microscopy microstructure provided in the embodiments of this application: 1. Control group setup: Method A (Example of this application): Training data is generated using a physically realistic degradation model that includes Gaussian blur, downsampling, and Poisson-Gaussian mixed noise.

[0098] Comparison Method B (Conventional Method 1): The most conventional degradation method in the field is adopted, that is, only the standard "bicubic interpolation downsampling" is used to generate training data without adding additional blurring and noise.

[0099] Comparison Method C (Conventional Method 2): A slightly optimized conventional method is adopted, which uses Gaussian blur and downsampling, but does not add a Poisson-Gaussian noise model that conforms to physical properties.

[0100] 2. Experimental procedure: Using identical high-resolution (HR) source images, three different training datasets were generated using the three methods described above. These three datasets were then used to train three structurally identical ESRGAN models (models A, B, and C). Subsequently, using the same batch of real, never-used low-resolution (LR) test images, vHR images were generated from the three trained models, and quantitative rock physical parameters were calculated on the vHR images, comparing them to "ground truth" values ​​(i.e., parameters calculated directly from the original HR images).

[0101] 3. Experimental Results and Analysis: Verification of Quantitative Recovery Capacity.

[0102] In this experiment, a representative shale sample was selected, whose "true" total porosity, obtained through precise measurement, was 8.0%. After low-cost imaging simulation (i.e., the degradation process), a large number of nanopores disappeared in the corresponding low-resolution (LR) image, and the calculated total porosity was only 3.5%. This means that 4.5% of the absolute porosity information was lost in the low-resolution image. Table 1 below shows a comparison of the quantitative recovery capabilities of different models for the lost porosity.

[0103] Table 1

[0104] The quantitative results in Table 1 show that all super-resolution methods increased porosity to varying degrees, but their recovery capabilities differed significantly. Comparing methods B and C, their overly simplistic degradation models resulted in AI models that merely learned to "make the image clearer" without learning to "reconstruct the true, lost nanopores." Consequently, their porosity recovery rates were only 13.3% and 44.4%, respectively, making their quantitative results completely unreliable and unsuitable for engineering decision-making.

[0105] Method A of this application, through its specific, physics-based training data generation strategy, enables the AI ​​model not only to learn how to improve visual quality but also how to quantitatively reconstruct microscopic pores invisible in low-resolution images. Its recovery rate of lost porosity reaches as high as 95.6%, and the calculated total porosity value (7.8%) is close to the true ground value (8.0%). This qualitative leap from "visual restoration" to "quantitative reconstruction" is something that those skilled in the art could not directly foresee from a simple combination of known techniques, powerfully demonstrating the extraordinary effects and core inventiveness of the present invention.

[0106] Please see Figure 6 This application also provides a shale scanning electron microscope (SEM) microstructure super-resolution imaging device 600, which can realize the above-mentioned shale SEM microstructure super-resolution imaging method. The shale SEM microstructure super-resolution imaging device 600 includes: The acquisition module 601 is used to acquire multiple first high-resolution images of the shale; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope; The generation module 602 is used to input each of the first high-resolution images into a degradation model, and generate a first low-resolution image corresponding to each of the first high-resolution images by simulating the physical effects of fast imaging through the degradation model; the low-resolution image is an image with a resolution lower than the preset resolution; The construction module 603 is used to construct a resolution training set based on the plurality of first high-resolution images and the first low-resolution images corresponding to each first high-resolution image; The training module 604 is used to train an initial generative machine learning model using the resolution training set, taking the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

[0107] The specific implementation of the shale scanning electron microscope microstructure super-resolution imaging device 600 is basically the same as the specific implementation of the shale scanning electron microscope microstructure super-resolution imaging method described above, and will not be repeated here.

[0108] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described super-resolution imaging method for the microstructure of shale using scanning electron microscopy. This electronic device can be any smart terminal, including desktop computers, tablets, mobile phones, and in-vehicle computers.

[0109] Please see Figure 7 , Figure 7 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and called and executed by the processor 901 to execute the shale scanning electron microscope microstructure super-resolution imaging method of the embodiments of this application. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0110] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described shale scanning electron microscope microstructure super-resolution imaging method.

[0111] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0112] Alternatively, this application embodiment can provide a computer program product for implementation, wherein when the instructions in the computer program product are executed by the processor of an electronic device, the electronic device implements the shale scanning electron microscope microstructure super-resolution imaging method in the above embodiment.

[0113] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0114] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0116] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0117] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0118] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0120] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0122] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A super-resolution imaging method for the microstructure of shale using scanning electron microscopy, characterized in that, The method includes: Multiple first high-resolution images of shale are acquired; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope; Each of the first high-resolution images is input into a degradation model, which simulates the physical effects of rapid imaging to generate a first low-resolution image corresponding to each of the first high-resolution images; the low-resolution image is an image with a resolution lower than the preset resolution. A resolution training set is constructed based on the plurality of first high-resolution images and the first low-resolution images corresponding to each first high-resolution image; Using the resolution training set, with the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, the initial generative machine learning model is trained to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

2. The method according to claim 1, characterized in that, The step of simulating the physical effects of rapid imaging using the degradation model to generate a first low-resolution image corresponding to each of the first high-resolution images includes: The physical effects of rapid imaging are simulated by the degradation model, and target processing is performed on each of the first high-resolution images to obtain the first low-resolution image corresponding to each of the first high-resolution images. The target processing includes at least one of the following: blurring, downsampling, and noise addition.

3. The method according to claim 2, characterized in that, The target processing performed on each of the first high-resolution images includes at least one of the following: The first high-resolution image is blurred by simulating Gaussian blur of the point spread function at low resolution. Perform downsampling processing on each of the first high-resolution images; Noise-adding processing is performed on each of the first high-resolution images by simulating mixed noise with a low signal-to-noise ratio during rapid imaging.

4. The method according to claim 1, characterized in that, The initial generative machine learning model includes a generator network and a discriminator network; The step of training an initial generative machine learning model using the resolution training set, taking the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process, includes: The generator network converts the first low-resolution image in the resolution training set into a corresponding second high-resolution image. The discriminator network is used to distinguish between the second high-resolution image and the first high-resolution image corresponding to the resolution training set, respectively, to obtain the discrimination scores of the second high-resolution image and the first high-resolution image; the discrimination scores are used to indicate the authenticity of the image. Based on the discrimination scores of the second high-resolution image and the first high-resolution image, and the second high-resolution image, the loss value of the initial generative machine learning model is calculated using the target loss function; If the loss value is greater than a preset threshold, the initial generative machine learning model is adjusted using backpropagation based on the loss value until the loss value is less than or equal to the preset threshold, thereby obtaining the target generative machine learning model.

5. The method according to claim 4, characterized in that, The target loss function includes perceptual loss and relativistic adversarial loss; The perceptual loss is used to indicate the similarity in morphology and texture between the microstructures in the second high-resolution image output by the initial generative machine learning model and the first high-resolution image. The relativistic adversarial loss is used to indicate the difference in realism between the second high-resolution image output by the initial generative machine learning model and the first high-resolution image.

6. The method according to claim 1, characterized in that, After training the initial generative machine learning model using the resolution training set to obtain the target generative machine learning model, the method further includes: By using acquisition parameters that match the physical effects of rapid imaging simulated by the degradation model, images of the shale to be analyzed are acquired to obtain a target low-resolution image of the shale to be analyzed. The low-resolution image of the shale to be analyzed is input into the target generative machine learning model, and the target generative machine learning model outputs a target high-resolution image. Based on the target high-resolution image, quantitative analysis is performed on the shale to be analyzed to obtain the rock physical parameters of the shale to be analyzed; the rock physical parameters are used to characterize the microstructure of the shale to be analyzed.

7. The method according to claim 6, characterized in that, The step of quantitatively analyzing the shale to be analyzed based on the target high-resolution image to obtain the rock physical parameters of the shale to be analyzed includes: The target high-resolution image is segmented to obtain a segmentation mask; the segmentation mask is used to identify the microstructure in the target high-resolution image. Based on the segmentation mask, feature parameters of the microstructure in the target high-resolution image are calculated to obtain the rock physical parameters of the shale to be analyzed.

8. A super-resolution imaging device for the microstructure of shale using scanning electron microscopy, characterized in that, The device includes: The acquisition module is used to acquire multiple first high-resolution images of the shale; the high-resolution images are images with a resolution greater than or equal to a preset resolution, and the first high-resolution images are obtained by scanning the shale with a scanning electron microscope; The generation module is used to input each of the first high-resolution images into a degradation model, and to simulate the physical effects of rapid imaging through the degradation model to generate a first low-resolution image corresponding to each of the first high-resolution images; the low-resolution image is an image with a resolution lower than the preset resolution; A construction module is used to construct a resolution training set based on the plurality of first high-resolution images and the first low-resolution images corresponding to each of the first high-resolution images; The training module is used to train an initial generative machine learning model using the resolution training set, taking the first low-resolution image corresponding to each of the first high-resolution images as input and each of the first high-resolution images as output, to obtain a target generative machine learning model that supports the reconstruction of the microstructure of the shale lost during the rapid imaging process; the resolution of the image output by the target generative machine model is the same as the resolution of the high-resolution image obtained by the scanning electron microscope.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the shale scanning electron microscope microstructure super-resolution imaging method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the super-resolution imaging method for the microstructure of shale by scanning electron microscopy as described in any one of claims 1 to 7.

Images