Wellbore electrical imaging super-resolution reconstruction method, device and image reconstruction system

Abstract

The application discloses a well wall electric imaging super-resolution reconstruction method and device and an image reconstruction system. The method comprises the following steps: acquiring a plurality of well wall electric imaging data segments, wherein the well wall electric imaging data segments are obtained by cutting a data set of well electric imaging; performing downsampling on the plurality of well wall electric imaging data segments to obtain a plurality of training data groups, wherein the training data groups correspond to the well wall electric imaging data segments in a one-to-one manner; inputting the plurality of training data groups into an adversarial network for iterative training until the change value of a Wasserstein distance, the change value of a perception loss function value and the change value of a discriminator loss function value are less than corresponding threshold values, and a generator model is obtained, wherein the adversarial network comprises a generator network and a discriminator network, and the generator model is the trained generator network; and inputting a well wall electric imaging data segment to be reconstructed into the generator model to obtain a reconstructed well wall image, thereby solving the technical problem of low resolution of well wall electric imaging.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more specifically, to a method, apparatus, computer-readable storage medium, processor, and image reconstruction system for super-resolution reconstruction of wellbore electrical imaging. Background Technology

[0002] In the field of oil extraction, with the continuous growth of societal demand for oil, corresponding oil well extraction technologies are also constantly developing and innovating in methods and techniques to help discover oil resources. Electrical imaging logging is a logging technology used in oil and gas field exploration. It uses the electrodes of a logging instrument to reflect changes in resistivity around the wellbore through changes in current. The electrical imaging images generated by electrical imaging logging technology allow geological researchers to obtain formation information from the wellbore more intuitively. However, currently, due to the influence of the geological environment, the unsatisfactory accuracy and performance of logging instruments, and data transmission losses, the wellbore images obtained by electrical imaging logging are not clear enough, have many noise points, and poor image quality, hindering further downhole analysis tasks. To better display the details in electrical logging images, we urgently need a method to improve the resolution and quality of electrical logging images. Low-resolution electrical logging images are blurry and lack detailed information, while high-resolution electrical logging images, restored through reconstruction, have natural and clear textures, good visual effects, and more detailed information.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a method, apparatus, computer-readable storage medium, processor, and image reconstruction system for super-resolution reconstruction of wellbore electrical imaging, in order to at least solve the technical problem of low resolution in wellbore electrical imaging.

[0005] According to one aspect of the present invention, a super-resolution reconstruction method for wellbore electrical imaging is provided, comprising: acquiring multiple wellbore electrical imaging data segments, wherein the wellbore electrical imaging data segments are obtained by segmenting a well logging electrical imaging dataset; downsampling the multiple wellbore electrical imaging data segments to obtain multiple training data sets, wherein the training data sets correspond one-to-one with the wellbore electrical imaging data segments; inputting the multiple training data sets into an adversarial network for iterative training until the changes in Wasserstein distance, the changes in perceptual loss function, and the changes in discriminator loss function are less than corresponding thresholds, thereby obtaining a generator model, wherein the adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network; and inputting the wellbore electrical imaging data segments to be reconstructed into the generator model to obtain a reconstructed wellbore image.

[0006] Optionally, acquiring multiple wellbore electrical imaging data segments includes: acquiring wellbore electrical imaging data, wherein the wellbore electrical imaging data is obtained by electrical imaging of the wellbore; and cutting the wellbore electrical imaging data to obtain multiple wellbore electrical imaging data segments of the same size, wherein both the wellbore electrical imaging data and the wellbore electrical imaging data segments are matrices, and the number of columns in the wellbore electrical imaging data segments is equal to the number of columns in the wellbore electrical imaging data.

[0007] Optionally, the number of rows in the wellbore electrical imaging data segment is a predetermined multiple of the number of columns in the wellbore electrical imaging data segment, wherein the predetermined multiple is any one of 4 times, 3 times, and 2 times.

[0008] Optionally, the adversarial network is iteratively trained using multiple sets of training data until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than corresponding thresholds, thus obtaining a generator model. This includes: a training step, where multiple sets of training data are input into the adversarial network for one iteration of training; and a first calculation step, where the Wasserstein distance is calculated according to a first formula, where the first formula is... π(P train P G ) is P train and P G The set of all possible joint distributions γ, where x is the input training data set and y is the image data output by the generator network; the second calculation step is to calculate the perceptual loss function value according to the second formula, which is: in, The third calculation step involves calculating the discriminator loss function value according to the third formula, which is: The fourth calculation step involves calculating the absolute value of the difference between the Wasserstein distance and the Wasserstein distance of the previous iteration to obtain the change value of the Wasserstein distance; calculating the absolute value of the difference between the perceptual loss function value and the perceptual loss function value of the previous iteration to obtain the change value of the perceptual loss function value; and calculating the absolute value of the difference between the discriminator loss function value and the discriminator loss function value of the previous iteration to obtain the change value of the discriminator loss function value. The adjustment step involves adjusting the parameters of the generator network if any one of the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value is greater than or equal to the corresponding threshold. The training step, the first calculation step, the second calculation step, the third calculation step, the fourth calculation step, and the adjustment step are repeated at least once in sequence until the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value are less than the corresponding thresholds, thus obtaining the generator model.

[0009] Optionally, the parameters of the generator network include the number of network layers and the convolution stride.

[0010] Optionally, after inputting the wellbore electrical imaging data fragment to be reconstructed into the generator model to obtain a reconstructed wellbore image, the method includes: calculating the peak signal-to-noise ratio and a similarity metric based on the reconstructed wellbore image; and evaluating the reconstructed wellbore image based on the peak signal-to-noise ratio and the similarity metric.

[0011] According to another aspect of the present invention, a super-resolution reconstruction apparatus for wellbore electrical imaging is also provided, comprising: an acquisition unit for acquiring multiple wellbore electrical imaging data segments, wherein the wellbore electrical imaging data segments are obtained by segmenting a dataset of well logging electrical imaging; a processing unit for downsampling the multiple wellbore electrical imaging data segments to obtain multiple training data sets, wherein the training data sets correspond one-to-one with the wellbore electrical imaging data segments; a training unit for iteratively training an adversarial network using the multiple training data sets as input until the change values ​​of the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than corresponding thresholds, thereby obtaining a generator model, wherein the adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network; and a reconstruction unit for inputting the wellbore electrical imaging data segments to be reconstructed into the generator model to obtain a reconstructed wellbore image.

[0012] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform any of the methods described.

[0013] According to another aspect of the present invention, a processor is also provided, the processor being configured to run a program, wherein the program, when running, executes any of the methods described herein.

[0014] According to another aspect of the present invention, an image reconstruction system is also provided, comprising: a wellbore electrical imaging device, one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for performing any one of the methods described.

[0015] In this embodiment of the invention, the above-mentioned super-resolution reconstruction method for wellbore electrical imaging firstly acquires multiple wellbore electrical imaging data segments, which are obtained by segmenting a well logging electrical imaging dataset; then, downsampling the multiple wellbore electrical imaging data segments yields multiple training data sets, which correspond one-to-one with the wellbore electrical imaging data segments; subsequently, the multiple training data sets are input into an adversarial network for iterative training until the changes in Wasserstein distance, perceptual loss function, and discriminator loss function are less than the corresponding thresholds, resulting in a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network; finally, the wellbore electrical imaging data segments to be reconstructed are input into the generator model to obtain a reconstructed wellbore image. This method reconstructs wellbore images using generative adversarial networks, transforming low-resolution electro-optical imaging images into high-resolution images, thus obtaining high-definition electro-optical imaging images. This helps well logging analysts better analyze the detailed conditions of oil wells and has the advantages of being more targeted, faster in computation, more flexible, and more effective, solving the technical problem of low resolution in existing wellbore electro-optical imaging technologies. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0017] Figure 1 This is a flowchart of a super-resolution reconstruction method for wellbore electrical imaging according to an embodiment of this application;

[0018] Figure 2This is a schematic diagram of the Gaussian pyramid sampling principle according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of the structure of a generative adversarial network according to an embodiment of this application;

[0020] Figure 4 This is a loss curve diagram of the network training process according to an embodiment of this application;

[0021] Figure 5 These are comparison images of wellbore electrical imaging reconstruction results according to embodiments of this application;

[0022] Figure 6 This is a schematic diagram of a super-resolution reconstruction apparatus for wellbore electrical imaging according to an embodiment of this application. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention 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 the invention 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 a 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.

[0025] As mentioned in the background section, the resolution of existing wellbore electrical imaging is low. To address the above problem, in a typical embodiment of this application, a method, apparatus, computer-readable storage medium, processor, and image reconstruction system for super-resolution reconstruction of wellbore electrical imaging are provided.

[0026] According to an embodiment of this application, a super-resolution reconstruction method for wellbore electrical imaging is provided.

[0027] Figure 1 This is a flowchart of a super-resolution reconstruction method for wellbore electrical imaging according to an embodiment of this application. For example... Figure 1 As shown, the method includes the following steps:

[0028] Step S101: Acquire multiple wellbore electrical imaging data segments, which are obtained by cutting the well logging electrical imaging dataset.

[0029] Step S102: Downsample multiple of the above-mentioned wellbore electrical imaging data segments to obtain multiple training data sets, and the training data sets correspond one-to-one with the above-mentioned wellbore electrical imaging data segments;

[0030] Step S103: Input multiple sets of the above training data into the adversarial network for iterative training until the changes in Wasserstein distance, perceptual loss function, and discriminator loss function are less than the corresponding thresholds, and obtain the generator model. The adversarial network includes a generator network and a discriminator network. The generator model is the trained generator network.

[0031] Step S104: Input the above-mentioned wellbore electrical imaging data fragment to be reconstructed into the above-mentioned generator model to obtain the reconstructed wellbore image.

[0032] In the above-described super-resolution reconstruction method for wellbore electrical imaging, firstly, multiple wellbore electrical imaging data segments are acquired, which are obtained by segmenting the well logging electrical imaging dataset; then, the multiple wellbore electrical imaging data segments are downsampled to obtain multiple training data sets, which correspond one-to-one with the wellbore electrical imaging data segments; subsequently, the multiple training data sets are input into an adversarial network for iterative training until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thus obtaining a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network; finally, the wellbore electrical imaging data segments to be reconstructed are input into the generator model to obtain the reconstructed wellbore image. This method reconstructs wellbore images using generative adversarial networks, transforming low-resolution electro-optical imaging images into high-resolution images, thus obtaining high-definition electro-optical imaging images. This helps well logging analysts better analyze the detailed conditions of oil wells and has the advantages of being more targeted, faster in computation, more flexible, and more effective, solving the technical problem of low resolution in existing wellbore electro-optical imaging technologies.

[0033] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0034] It should also be noted that the Gaussian pyramid method is used to downsample the well logging electrical imaging to generate low-resolution images. By stacking images of different sizes together, with the highest resolution image at the bottom, and following the rule that each layer's size is half that of the layer above, the images are stacked in a pyramid shape according to their resolution. Figure 2 As shown, downsampling is performed to generate a training set, which includes high-resolution images and low-resolution images.

[0035] In one embodiment of this application, acquiring multiple wellbore electrical imaging data segments includes: acquiring wellbore electrical imaging data, wherein the wellbore electrical imaging data is obtained by electrical imaging of the wellbore; and segmenting the wellbore electrical imaging data to obtain multiple wellbore electrical imaging data segments of the same size. Both the wellbore electrical imaging data and the wellbore electrical imaging data segments are matrices, and the number of columns in the wellbore electrical imaging data segments is equal to the number of columns in the wellbore electrical imaging data. Specifically, in the segmented wellbore electrical imaging data segments, the number of rows in the matrix is ​​an integer multiple of the number of columns. For example, if the dataset contains wellbore electrical imaging data from 13 wells, and the width of the wellbore electrical imaging dataset is 192 and the width is 144, the dataset is segmented into small sample segments of 144×576 and 192×768. Subsequently, an equidistant sampling method is used to allocate a proportional number of test and training sets according to the proportion of small samples in the dataset.

[0036] In one embodiment of this application, the number of rows in the aforementioned wellbore electrical imaging data segment is a predetermined multiple of the number of columns in the aforementioned wellbore electrical imaging data segment, wherein the predetermined multiple is any one of 4 times, 3 times, and 2 times. Specifically, the predetermined multiple is any one of 4 times, 3 times, and 2 times to cut electrical imaging segments of normal image size.

[0037] It should be noted that some data values ​​were abnormal due to the detector not activating in time during data acquisition, and these needed to be deleted or discarded. Also, because some datasets had an odd number of rows, the output data size differed from the original data size during image downsampling, necessitating adjustment of the image row count.

[0038] In one embodiment of this application, multiple sets of training data are input into the adversarial network for iterative training until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than corresponding thresholds, thus obtaining a generator model. This includes: a training step, inputting multiple sets of training data into the adversarial network for one iterative training; and a first calculation step, calculating the Wasserstein distance according to a first formula, wherein the first formula is... π(P train P G ) is Ptrain and P G The set of all possible joint distributions γ, where x is the input training data set and y is the image data output by the generator network; the second calculation step is to calculate the perceptual loss function value according to the second formula, which is: in, The third calculation step involves calculating the discriminator loss function value according to the third formula, which is: The fourth calculation step involves calculating the absolute value of the difference between the Wasserstein distance and the Wasserstein distance from the previous iteration to obtain the change in the Wasserstein distance; calculating the absolute value of the difference between the perceptual loss function value and the perceptual loss function value from the previous iteration to obtain the change in the perceptual loss function value; and calculating the absolute value of the difference between the discriminator loss function value and the discriminator loss function value from the previous iteration to obtain the change in the discriminator loss function value. In the adjustment step, if any one of the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value is greater than or equal to the corresponding threshold, the parameters of the generator network are adjusted. This process is repeated at least once in sequence through the training step, the first calculation step, the second calculation step, the third calculation step, the fourth calculation step, and the adjustment step until the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value are less than the corresponding thresholds, thus obtaining the generator model. Specifically, as shown... Figure 3 As shown, low-resolution well logging electrical images from the training set are input into the initial generator model. Then, the reconstructed well logging electrical images from the output and the original well logging electrical images from the training set are used as inputs to the discriminator network. The discriminator network is iteratively trained until it can distinguish between low-resolution and original well logging electrical images. The generator network is then trained again until any one of the following—the change in Wasserstein distance, the change in the perceptual loss function, and the change in the discriminator loss function—is less than or equal to the corresponding threshold. Figure 4 As shown, the discriminator remains essentially unchanged, and its discrimination accuracy is continuously improved. Based on the judgment results of the discriminator network, the generator network is trained by backpropagation to improve the reconstruction performance of the generator network, enabling it to reconstruct higher quality logging electrical imaging.

[0039] It should be noted that the generator network described above is a U-shaped generator network, combined with a discriminator network to form a generative adversarial network (GAN), which generates high-resolution imaging images through adversarial processing. The generator consists of three parts: a backbone feature extraction part, a reinforcement feature extraction part, and a generation part, ultimately resulting in an effective feature layer that predicts and generates high-resolution well logging electrical imaging. Subsequently, the discriminator network uses a convolutional module to extract features, obtaining feature maps. Finally, a linear layer transforms the feature dimensions, outputting a one-dimensional tensor as the classification result.

[0040] In one embodiment of this application, the parameters of the generator network include the number of network layers and the convolution stride. Specifically, the network structure can be adjusted and retrained when the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function exceed the corresponding thresholds, thereby improving the reconstruction performance of the generator network.

[0041] In one embodiment of this application, after inputting the aforementioned wellbore electrical imaging data fragment to be reconstructed into the aforementioned generator model to obtain a reconstructed wellbore image, the method includes: calculating the peak signal-to-noise ratio (PSNR) and a similarity metric based on the reconstructed wellbore image; and evaluating the reconstructed wellbore image based on the PSNR and the similarity metric. Specifically, the formula for calculating the PSNR is as follows: Where MAX represents the maximum possible pixel value of the image, X(i,j) represents the pixel value of the pixel at coordinate (i,j) in image X, Y(i,j) represents the pixel value of the pixel at coordinate (i,j) in image Y, H is the height of the image, W is the width of the image, and the similarity metric SSIM is calculated as SSIM(X,Y)=L(X,Y)·C(X,Y)·S(X,Y), where, Used to characterize the brightness difference between image X and image Y. Used to characterize the contrast difference between image X and image Y. μ is used to characterize the contrast difference between image X and image Y. X and μ Y σ is the mean of the pixel values ​​at each point in images X and Y. X and σ Y σ represents the variance of the pixel values ​​at each point in images X and Y. XY Let C1, C2, and C3 be the covariance of pixel values ​​at each point in images X and Y, and C3 be three constants. Peak signal-to-noise ratio (PSNR) and result similarity metric are used as two quality evaluation indicators to measure the super-resolution reconstruction effect of well logging electrical imaging. This evaluation assesses the quality of the well logging electrical imaging reconstruction. Figure 5 As shown, the reconstructed image is compared with the original image, and the reconstructed image has a higher resolution.

[0042] This application also provides a super-resolution reconstruction apparatus for wellbore electrical imaging. It should be noted that the super-resolution reconstruction apparatus for wellbore electrical imaging provided in this application can be used to execute the super-resolution reconstruction method for wellbore electrical imaging provided in this application. The following describes the super-resolution reconstruction apparatus for wellbore electrical imaging provided in this application.

[0043] Figure 6 A schematic diagram of a super-resolution reconstruction apparatus for wellbore electrical imaging according to an embodiment of this application. (See diagram below.) Figure 6 The device includes:

[0044] The acquisition unit 10 is used to acquire multiple wellbore electrical imaging data segments, which are obtained by cutting the well logging electrical imaging dataset.

[0045] Processing unit 20 is used to downsample multiple of the above-mentioned wellbore electrical imaging data segments to obtain multiple training data sets, wherein each training data set corresponds one-to-one with the above-mentioned wellbore electrical imaging data segment.

[0046] Training unit 30 is used to input multiple sets of training data into the adversarial network for iterative training until the changes in Wasserstein distance, perceptual loss function, and discriminator loss function are less than the corresponding thresholds, thereby obtaining a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network.

[0047] The reconstruction unit 40 is used to input the above-mentioned wellbore electrical imaging data fragments to be reconstructed into the above-mentioned generator model to obtain the reconstructed wellbore image.

[0048] In the aforementioned super-resolution reconstruction device for wellbore electrical imaging, the acquisition unit acquires multiple wellbore electrical imaging data segments, which are obtained by segmenting the dataset of well logging electrical imaging; the processing unit downsamples the multiple wellbore electrical imaging data segments to obtain multiple training data sets, which correspond one-to-one with the wellbore electrical imaging data segments; the training unit inputs the multiple training data sets into the adversarial network for iterative training until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thus obtaining a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network; the reconstruction unit inputs the wellbore electrical imaging data segments to be reconstructed into the generator model to obtain the reconstructed wellbore image. This device reconstructs wellbore images using generative adversarial networks, transforming low-resolution electro-optical imaging images into high-resolution images to obtain high-definition electro-optical imaging images. This helps well logging analysts better analyze the detailed conditions of oil wells and has the advantages of being more targeted, faster in processing, more flexible, and more effective. It solves the technical problem of low resolution in existing wellbore electro-optical imaging technologies.

[0049] It should be noted that the Gaussian pyramid method is used to downsample the well logging electrical imaging to generate low-resolution images. By stacking images of different sizes together, with the highest resolution image at the bottom, and following the rule that each layer's size is half that of the layer above, the images are stacked in a pyramid shape according to their resolution. Figure 2 As shown, downsampling is performed to generate a training set, which includes high-resolution images and low-resolution images.

[0050] In one embodiment of this application, the aforementioned unit includes an acquisition module and a processing module. The acquisition module acquires wellbore electrical imaging data, which is obtained by performing electrical imaging on the wellbore. The processing module segments the wellbore electrical imaging data to obtain multiple segments of the same size. Both the wellbore electrical imaging data and the wellbore electrical imaging data segments are matrices, and the number of columns in each wellbore electrical imaging data segment is equal to the number of columns in the wellbore electrical imaging data. Specifically, in the segmented wellbore electrical imaging data segments, the number of rows in the matrix is ​​an integer multiple of the number of columns. For example, if the dataset contains wellbore electrical imaging data from 13 wells, and the width of the wellbore electrical imaging dataset is 192 and the width is 144, the dataset is segmented into small sample segments of 144×576 and 192×768. Subsequently, an equidistant sampling method is used to allocate a proportional number of test and training sets according to the proportion of small samples in the dataset.

[0051] In one embodiment of this application, the number of rows in the aforementioned wellbore electrical imaging data segment is a predetermined multiple of the number of columns in the aforementioned wellbore electrical imaging data segment, wherein the predetermined multiple is any one of 4 times, 3 times, and 2 times. Specifically, the predetermined multiple is any one of 4 times, 3 times, and 2 times to cut electrical imaging segments of normal image size.

[0052] It should be noted that some data values ​​were abnormal due to the detector not activating in time during data acquisition, and these needed to be deleted or discarded. Also, because some datasets had an odd number of rows, the output data size differed from the original data size during image downsampling, necessitating adjustment of the image row count.

[0053] In one embodiment of this application, the training unit includes a training module, a first calculation module, a second calculation module, a third calculation module, a fourth calculation module, an adjustment module, and a repetition module. The training module performs a training step, inputting multiple sets of training data into the adversarial network for one iteration of training. The first calculation module performs a first calculation step, calculating the Wasserstein distance according to a first formula, where the first formula is... π(P train P G ) is P train and P G The set of all possible joint distributions γ, where x is the input training data set and y is the image data output by the generator network; the second calculation module is used to perform the second calculation step, calculating the perceptual loss function value according to the second formula, which is: in, The aforementioned third calculation module is used to perform the third calculation step, calculating the discriminator loss function value according to the third formula, wherein the third formula is: The fourth calculation module is used to execute the fourth calculation step, calculating the absolute value of the difference between the Wasserstein distance and the Wasserstein distance of the previous iteration to obtain the change value of the Wasserstein distance; calculating the absolute value of the difference between the perceptual loss function value and the perceptual loss function value of the previous iteration to obtain the change value of the perceptual loss function; calculating the absolute value of the difference between the discriminator loss function value and the discriminator loss function value of the previous iteration to obtain the change value of the discriminator loss function; the adjustment module is used to execute the adjustment step, adjusting the parameters of the generator network when any one of the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value is greater than or equal to the corresponding threshold; the repetition module is used to execute the training step, the first calculation step, the second calculation step, the third calculation step, the fourth calculation step, and the adjustment step at least once in sequence until the changes in the Wasserstein distance, the perceptual loss function value, and the discriminator loss function value are less than the corresponding threshold, thus obtaining the generator model. Specifically, as shown... Figure 3 As shown, low-resolution well logging electrical images from the training set are input into the initial generator model. Then, the reconstructed well logging electrical images from the output and the original well logging electrical images from the training set are used as inputs to the discriminator network. The discriminator network is iteratively trained until it can distinguish between low-resolution and original well logging electrical images. The generator network is then trained again until any one of the following—the change in Wasserstein distance, the change in the perceptual loss function, and the change in the discriminator loss function—is less than or equal to the corresponding threshold. Figure 4 As shown, the discriminator remains essentially unchanged, and its discrimination accuracy is continuously improved. Based on the judgment results of the discriminator network, the generator network is trained by backpropagation to improve the reconstruction performance of the generator network, enabling it to reconstruct higher quality logging electrical imaging.

[0054] It should be noted that the generator network described above is a U-shaped generator network, combined with a discriminator network to form a generative adversarial network (GAN), which generates high-resolution imaging images through adversarial processing. The generator consists of three parts: a backbone feature extraction part, a reinforcement feature extraction part, and a generation part, ultimately resulting in an effective feature layer that predicts and generates high-resolution well logging electrical imaging. Subsequently, the discriminator network uses a convolutional module to extract features, obtaining feature maps. Finally, a linear layer transforms the feature dimensions, outputting a one-dimensional tensor as the classification result.

[0055] In one embodiment of this application, the parameters of the generator network include the number of network layers and the convolution stride. Specifically, the network structure can be adjusted and retrained when the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function exceed the corresponding thresholds, thereby improving the reconstruction performance of the generator network.

[0056] In one embodiment of this application, after inputting the wellbore electrical imaging data fragment to be reconstructed into the generator model to obtain a reconstructed wellbore image, the apparatus includes an evaluation unit. The evaluation unit includes a fifth calculation module and an evaluation module. The fifth calculation module is used to calculate the peak signal-to-noise ratio (PSNR) and a similarity metric based on the reconstructed wellbore image. The evaluation module is used to evaluate the reconstructed wellbore image based on the PSNR and the similarity metric. Specifically, the formula for calculating the PSNR is as follows: Where MAX represents the maximum possible pixel value of the image, X(i,j) represents the pixel value of the pixel at coordinate (i,j) in image X, Y(i,j) represents the pixel value of the pixel at coordinate (i,j) in image Y, H is the height of the image, W is the width of the image, and the similarity metric SSIM is calculated as SSIM(X,Y)=L(X,Y)·C(X,Y)·S(X,Y), where, Used to characterize the brightness difference between image X and image Y. Used to characterize the contrast difference between image X and image Y. μ is used to characterize the contrast difference between image X and image Y. X and μ Y σ is the mean of the pixel values ​​at each point in images X and Y. X and σ Y σ represents the variance of the pixel values ​​at each point in images X and Y. XY Let C1, C2, and C3 be the covariance of pixel values ​​at each point in images X and Y, and C3 be three constants. Peak signal-to-noise ratio (PSNR) and result similarity metric are used as two quality evaluation indicators to measure the super-resolution reconstruction effect of well logging electrical imaging. This evaluation assesses the quality of the well logging electrical imaging reconstruction. Figure 5 As shown, the reconstructed image is compared with the original image, and the reconstructed image has a higher resolution.

[0057] Example 1

[0058] Step 1: Due to the detector not starting in time during some data acquisition in this embodiment, the acquired data value was abnormally -99999.0, which needs to be deleted and discarded. Since some datasets have an odd number of rows, the output data size is inconsistent with the original data size during image downsampling; therefore, the last row of these odd-numbered rows must be deleted.

[0059] Step 2: Fragment the dataset. The depth of each well is divided into segments four times its width, generating electrical imaging fragments of normal image size. In this embodiment, the dataset contains electrical imaging data from 13 wellbore walls. The wellbore electrical imaging dataset has widths of 192 and 144, which are then fragmented into small sample segments of 144×576 and 192×768. Subsequently, an equidistant sampling method is used to allocate equal numbers of test and training sets according to the proportion of small samples in the dataset.

[0060] Step 3: Generate training data through downsampling. The Gaussian pyramid method is used to downsample the well logging electrical imaging to generate low-resolution images. This method combines sampling and Gaussian smoothing filtering for multi-scale representation. Specifically, images of different sizes are stacked together, with the highest resolution image at the bottom. Each layer is stacked half the size of the layer above it, forming a pyramid shape based on image resolution. (See diagram below.) Figure 2 As shown.

[0061] Step 4: Design a U-shaped generator network and combine it with a discriminator network to form a generative adversarial network (GAN) to generate high-resolution imaging images through adversarial processing. The generator consists of three parts: a backbone feature extraction part, a enhancement feature extraction part, and a generation part. Finally, an effective feature layer is obtained to predict and generate high-resolution well logging electrical imaging. The backbone feature extraction process includes three convolutional blocks and three downsampling layers. Each convolutional block has two convolutional layers with a kernel size of 3*3 and a stride of 1. In the downsampling layers, a max-pooling layer with a kernel size of 3*3 and a stride of 2 is used. The enhancement feature extraction part includes three convolutional blocks, three upsampling layers, and three skip connections. The final generation stage includes a convolutional operation with a kernel size of 3*3 and a stride of 1. Detailed network structures are shown in Table 1.

[0062] Table 1

[0063]

[0064]

[0065] The discriminator network is a binary classifier. The first convolution transforms the input 3-channel well logging electrical imaging data into 64 3×3 filters. Five convolutional modules in the middle extract features, yielding 512 feature maps. Finally, a linear layer transforms the feature dimension, outputting a one-dimensional tensor as the classification result. Its structural parameters are shown in Table 2.

[0066] Table 2

[0067] name Kernel function Step length Normalization Activation function enter 128×128 / 3 Conv0 3×3 / 64 1 - Leaky RELU Conv1 3×3 / 64 2 BN Leaky RELU Conv2 3×3 / 128 1 BN Leaky RELU Conv3 3×3 / 128 2 BN Leaky RELU Conv4 3×3 / 512 1 BN Leaky RELU Conv5 3×3 / 512 2 BN Leaky RELU

[0068] Step 5: The specific implementation process of training the network designed in the above steps involves using the pre-trained network model as the initial generator model. Low-resolution well logging electrical images from the training set are input into the initial generator model. Then, the reconstructed well logging electrical images and the original well logging electrical images from the training set are input into the discriminator network. The discriminator network is trained iteratively five times until it can distinguish between low-resolution and original well logging electrical images. Then, the generator network is trained. The discrimination accuracy of the discriminator is continuously improved. Based on the judgment results of the discriminator network, the generator network is trained through backpropagation to improve the network's reconstruction performance, enabling it to reconstruct higher-quality well logging electrical images. In the testing and reconstruction phases, only the low-resolution well logging electrical images from the test set need to be input into the trained generator network model to obtain the reconstructed high-resolution well logging electrical images. The parameters are: initial learning rate of 0.0001, decay rate of 0.1, and batch size of 128.

[0069] Step 6, Image Reconstruction and Evaluation. This patent uses two quality evaluation indicators, peak signal-to-noise ratio and result similarity metric, as the basis for measuring the super-resolution reconstruction effect of well logging electrical imaging to evaluate the quality of well logging electrical imaging reconstruction.

[0070] To objectively evaluate the reconstruction effect and efficiency of the reconstruction algorithm in this chapter and other algorithms, this patent calculated the PSNR and SSIM values ​​of the original and reconstructed well logging electrical images on the test set, and recorded the reconstruction time. Super-resolution reconstruction of well logging electrical images was performed using SRResNet, SRUNet, SRResNet_GAN, and SRUNet_GAN methods, respectively. The obtained PSNR, SSIM, and training time results are shown in Table 3.

[0071] Table 3

[0072] SRResNet SRUNet SRResNet_GAN SRUNet_GAN PSNR 31.479dB 31.279dB 31.556dB 31.452dB SSIM 0.865 0.857 0.867 0.862 Training time 585s 96s 2626s 1305s

[0073] As can be seen from the table above, compared with traditional interpolation algorithms, the metrics of this invention are improved, the reconstruction effect is better, and it has superior performance and faster speed. Figure 5As shown, the super-resolution reconstruction model is clearer and has better results than traditional interpolation algorithms.

[0074] This application also provides an image reconstruction system, including: a wellbore electrical imaging device, one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for performing any of the above methods.

[0075] The aforementioned image reconstruction system includes a wellbore electrical imaging device, one or more processors, a memory, and one or more programs. The programs are stored in the memory and configured to be executed by the processors. Each program includes methods for performing one of the aforementioned techniques. This system reconstructs wellbore images using a generative adversarial network (GAN) approach, converting low-resolution electrical imaging images into high-resolution images, thus obtaining high-definition electrical imaging. This helps well logging analysts better analyze the detailed conditions of oil wells, offering advantages such as greater targeting, faster processing speed, higher flexibility, and better results. It solves the technical problem of low resolution in existing wellbore electrical imaging technologies.

[0076] The aforementioned super-resolution reconstruction device for wellbore electrical imaging includes a processor and a memory. The acquisition unit, processing unit, training unit, and reconstruction unit are all stored as program units in the memory, and the processor executes the aforementioned program units stored in the memory to achieve the corresponding functions.

[0077] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the low resolution problem in existing wellbore electrical imaging technologies.

[0078] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0079] This invention provides a computer-readable storage medium including a stored program, wherein the program, when running, controls the device containing the computer-readable storage medium to execute the method.

[0080] This invention provides a processor for running a program, wherein the program executes the method described above during runtime.

[0081] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0082] Step S101: Acquire multiple wellbore electrical imaging data segments, which are obtained by cutting the well logging electrical imaging dataset.

[0083] Step S102: Downsample multiple of the above-mentioned wellbore electrical imaging data segments to obtain multiple training data sets, and the training data sets correspond one-to-one with the above-mentioned wellbore electrical imaging data segments;

[0084] Step S103: Input multiple sets of the above training data into the adversarial network for iterative training until the changes in Wasserstein distance, perceptual loss function, and discriminator loss function are less than the corresponding thresholds, and obtain the generator model. The adversarial network includes a generator network and a discriminator network. The generator model is the trained generator network.

[0085] Step S104: Input the above-mentioned wellbore electrical imaging data fragment to be reconstructed into the above-mentioned generator model to obtain the reconstructed wellbore image.

[0086] The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0087] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0088] Step S101: Acquire multiple wellbore electrical imaging data segments, which are obtained by cutting the well logging electrical imaging dataset.

[0089] Step S102: Downsample multiple of the above-mentioned wellbore electrical imaging data segments to obtain multiple training data sets, and the training data sets correspond one-to-one with the above-mentioned wellbore electrical imaging data segments;

[0090] Step S103: Input multiple sets of the above training data into the adversarial network for iterative training until the changes in Wasserstein distance, perceptual loss function, and discriminator loss function are less than the corresponding thresholds, and obtain the generator model. The adversarial network includes a generator network and a discriminator network. The generator model is the trained generator network.

[0091] Step S104: Input the above-mentioned wellbore electrical imaging data fragment to be reconstructed into the above-mentioned generator model to obtain the reconstructed wellbore image.

[0092] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units described above can be 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 units or modules may be electrical or other forms.

[0094] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Furthermore, the functional units in the various embodiments of the present invention 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.

[0096] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, 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 several 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 described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0097] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A super-resolution reconstruction method for wellbore electrical imaging, characterized in that, include: Multiple wellbore electrical imaging data segments are acquired, wherein the wellbore electrical imaging data segments are obtained by cutting the dataset of well logging electrical imaging; Multiple wellbore electrical imaging data segments are downsampled to obtain multiple training data sets, and the training data sets correspond one-to-one with the wellbore electrical imaging data segments. Multiple sets of training data are input into the adversarial network for iterative training until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thus obtaining a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network. The wellbore electrical imaging data fragment to be reconstructed is input into the generator model to obtain the reconstructed wellbore image; The adversarial network is iteratively trained using multiple sets of training data until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thus obtaining the generator model, which includes: The training step involves inputting multiple sets of training data into the adversarial network for one iteration of training. The first calculation step involves calculating the Wasserstein distance according to a first formula, where the first formula is: , for and All possible joint distributions of the combination The set of data, where x is the input training data set and y is the image data output by the generator network; The second calculation step involves calculating the perceptual loss function value according to the second formula, which is: ,in, , , ; The third calculation step involves calculating the discriminator loss function value according to the third formula, which is: ; The fourth calculation step involves calculating the absolute value of the difference between the Wasserstein distance and the Wasserstein distance of the previous iteration to obtain the change value of the Wasserstein distance; calculating the absolute value of the difference between the perceptual loss function value and the perceptual loss function value of the previous iteration to obtain the change value of the perceptual loss function value; and calculating the absolute value of the difference between the discriminator loss function value and the discriminator loss function value of the previous iteration to obtain the change value of the discriminator loss function value. The adjustment step involves adjusting the parameters of the generator network if any one of the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function is greater than or equal to the corresponding threshold. The training step, the first calculation step, the second calculation step, the third calculation step, the fourth calculation step, and the adjustment step are repeated at least once in sequence until the change values ​​of the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thus obtaining the generator model.

2. The method according to claim 1, characterized in that, Acquire multiple wellbore electrical imaging data fragments, including: Acquire electrical imaging data of the wellbore, wherein the electrical imaging data of the wellbore is obtained by electrical imaging of the wellbore. The wellbore electrical imaging data is segmented to obtain multiple wellbore electrical imaging data segments of the same size. Both the wellbore electrical imaging data and the wellbore electrical imaging data segments are matrices, and the number of columns in the wellbore electrical imaging data segments is equal to the number of columns in the wellbore electrical imaging data.

3. The method according to claim 2, characterized in that, The number of rows in the wellbore electrical imaging data segment is a predetermined multiple of the number of columns in the wellbore electrical imaging data segment, and the predetermined multiple is any one of 4 times, 3 times and 2 times.

4. The method according to claim 1, characterized in that, The parameters of the generator network include the number of network layers and the convolution stride.

5. The method according to claim 1, characterized in that, After inputting the wellbore electrical imaging data fragment to be reconstructed into the generator model to obtain the reconstructed wellbore image, the method includes: The peak signal-to-noise ratio and similarity metric are calculated based on the reconstructed wellbore image; The reconstructed wellbore image is evaluated based on the peak signal-to-noise ratio and the similarity metric.

6. A super-resolution reconstruction device for wellbore electrical imaging, characterized in that, include: The acquisition unit is used to acquire multiple wellbore electrical imaging data segments, wherein the wellbore electrical imaging data segments are obtained by cutting the dataset of well logging electrical imaging; The processing unit is used to downsample multiple wellbore electrical imaging data segments to obtain multiple training data sets, wherein each training data set corresponds one-to-one with a wellbore electrical imaging data segment. The training unit is used to input multiple sets of training data into the adversarial network for iterative training until the changes in the Wasserstein distance, the perceptual loss function, and the discriminator loss function are less than the corresponding thresholds, thereby obtaining a generator model. The adversarial network includes a generator network and a discriminator network, and the generator model is the trained generator network. The reconstruction unit is used to input the wellbore electrical imaging data fragment to be reconstructed into the generator model to obtain a reconstructed wellbore image; The training unit includes a training module, a first calculation module, a second calculation module, a third calculation module, a fourth calculation module, an adjustment module, and a repetition module. The training module is used to execute training steps, and inputs multiple sets of training data into the adversarial network for one iteration of training. The first calculation module is used to perform a first calculation step, calculating the Wasserstein distance according to a first formula, wherein the first formula is: , for and All possible joint distributions of the combination The set of data, where x is the input training data set and y is the image data output by the generator network; The second calculation module is used to perform the second calculation step, calculating the value of the perceptual loss function according to the second formula, which is: ,in, , , ; The third calculation module is used to perform a third calculation step, calculating the discriminator loss function value according to a third formula, which is: ; The fourth calculation module is used to perform a fourth calculation step, which involves calculating the absolute value of the difference between the Wasserstein distance and the Wasserstein distance of the previous iteration to obtain the change value of the Wasserstein distance, calculating the absolute value of the difference between the perceptual loss function value and the perceptual loss function value of the previous iteration to obtain the change value of the perceptual loss function value, and calculating the absolute value of the difference between the discriminator loss function value and the discriminator loss function value of the previous iteration to obtain the change value of the discriminator loss function value. The adjustment module is used to perform adjustment steps, adjusting the parameters of the generator network when any one of the changes in the Wasserstein distance, the changes in the perceptual loss function, and the changes in the discriminator loss function is greater than or equal to the corresponding threshold. The repetition module is used to sequentially execute the training step, the first calculation step, the second calculation step, the third calculation step, the fourth calculation step, and the adjustment step at least once, until the change value of the Wasserstein distance, the change value of the perceptual loss function, and the change value of the discriminator loss function are less than the corresponding thresholds, thereby obtaining the generator model.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method of any one of claims 1 to 5.

8. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method of any one of claims 1 to 5 when it runs.

9. An image reconstruction system, characterized in that, include: The wellbore electro-imaging apparatus, one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing any one of claims 1 to 5.

Images