A method and system for identifying electrical imaging well logging information based on deep learning

Abstract

The present invention provides a method and system for recognizing well logging information from electrical imaging images based on deep learning. The method involves amplifying the acquired original electrical imaging images, then numbering them and storing them in association with the images. A generative adversarial network is used to repair the amplified images to obtain corresponding electrical imaging restoration images. The restoration images are then processed through a well logging labeling step, simultaneously obtaining a matching electrical imaging mask image. The labeled electrical imaging restoration images and the matching mask images are then used as input, processed using a predefined prediction model to obtain a well logging information prediction map. This is then combined with the matching electrical imaging restoration images for fusion calculations to obtain well logging interpretation results encompassing multiple geological types. This approach overcomes the drawbacks of existing technologies, such as reliance on the expertise of analysts and insufficient data integrity in analysis results, enabling automatic recognition of well logging electrical imaging information. This ensures recognition efficiency while improving the accuracy of well logging interpretation.

Description

Technical Field

[0001] The present invention relates to the field of well logging interpretation and identification technology, and in particular to a method and system for identifying electrical imaging image logging information based on deep learning. Background Art

[0002] Well logging data is the highest vertical resolution and most continuous geological data available to date, containing a wealth of usable information. Electrical imaging logging technology converts measured physical information such as the resistivity of the wellbore formation into electrical imaging image data, which can more intuitively reflect geological information. However, traditional well logging interpretation methods often use manual fitting or subjective expert analysis to identify the data. Manual interpretation requires a huge workload for conventional small- and medium-scale well logging interpretation, resulting in low processing efficiency. The accuracy of the identification results varies with the interpreter's cognition and proficiency, which affects the acquisition of correct results, makes it impossible to guarantee the reliability and accuracy of the data interpretation results, and cannot meet the actual engineering needs.

[0003] Furthermore, although the existing technical personnel's research on logging data recognition is gradually tending towards automation, combining manual judgment with new methods, such as automatically identifying electrical imaging images based on deep learning technology, obtaining the logging phase category corresponding to the image, and providing data support for further analysis by experts, specifically, it constructs a convolutional neural network structure and trains a deep learning model, directly inputs the collected electrical imaging images into the deep learning model for recognition, and then outputs the corresponding logging phase category information. Although this method can improve the efficiency of logging interpretation to a certain extent, it relies entirely on the deep learning model to recognize complex electrical imaging images, and is highly dependent on massive training samples and model quality, and the reliability of the results cannot be guaranteed; and after automatic recognition, only the output logging phase category results can be obtained, and the analysis data of various different logging information cannot be intuitively seen, and sufficient information cannot be provided for manual interpretation, analysis and judgment of subsequent test data, and the accuracy and reliability are insufficient.

[0004] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art known to those skilled in the art. Summary of the Invention

[0005] To solve the above problems, the present invention provides a method for identifying electrical imaging well logging information based on deep learning. In one embodiment, the method includes:

[0006] The original image arrangement step is to obtain the well logging electrical imaging image in the well logging data as the original image, amplify and number the original image, and associate and store the amplified original image and number information;

[0007] Image restoration step: using a generative adversarial network to restore the original image after amplification to obtain the corresponding electrical imaging restoration image;

[0008] The well logging marking step is to mark the well logging information on the electrical imaging restoration image based on the well logging geological type data in the required file format, and generate an electrical imaging mask image that matches the marked well logging information;

[0009] Prediction and recognition step: inputting the marked electrical imaging restoration image and the corresponding electrical imaging mask image into a pre-trained well logging information prediction model, processing the feature data of the input image based on the unique encoder-decoder network structure of the model, and obtaining a corresponding well logging information prediction map;

[0010] A fusion calculation step is performed on the well logging information prediction image and the matched electrical imaging restoration image to obtain well logging interpretation results containing multiple geological types;

[0011] The well logging information prediction model is obtained through a model training step based on a set sample electrical imaging restoration image and a set sample electrical imaging mask image.

[0012] Furthermore, in one embodiment, in the original image arrangement step, the process of performing amplification processing on the original image includes:

[0013] The collected electrical imaging original image is subjected to data enhancement processing to obtain an amplified electrical imaging original image; the data enhancement processing includes left-right mirroring, top-bottom mirroring and rotation transformation processing.

[0014] In a preferred embodiment, the image restoration step includes:

[0015] The original image after the amplification process is repaired by using the restoration network and identification network structure of the generative adversarial network, wherein the generative adversarial network includes a restoration network and two identification networks.

[0016] Specifically, in one embodiment, the two authentication networks are a global authentication network and a local authentication network;

[0017] The global identification network takes the complete image as input to identify the global consistency of the scene, and the local identification network ensures the details of the image repair area;

[0018] The local discriminator uses a set convolutional network structure to encode the original image to obtain an intermediate image whose size meets the requirements of the original image size; then the image is decoded using the set convolutional network structure to obtain a restored image after repair.

[0019] Furthermore, in one embodiment, in the well logging marking step, the well logging geological features in the electrical imaging restoration image are distinguishably marked according to different well logging geological type data, with the JSON format file as the target, and then the generated JSON files are batch converted to obtain the electrical imaging restoration image and the electrical imaging restoration image mask image with matching numbers;

[0020] Among them, different marking forms are set according to different logging geological types of data.

[0021] In an optional embodiment, for the well logging information prediction model, its encoder includes a basic feature extraction module and a multi-scale parallel pooling module;

[0022] The basic feature extraction module performs convolution and pooling processing on the sample library image to obtain a low-level feature map that meets the requirements;

[0023] The multi-scale parallel pooling module performs multi-scale parallel pooling module processing on the low-level feature map to obtain a fusion map of feature maps of different scales, i.e., a multi-scale feature fusion map;

[0024] Its decoder performs transposed convolution processing on the multi-scale feature fusion map to obtain a decoded feature map with a resolution that meets the requirements, and splices it with the low-level feature map with the same resolution to obtain a spliced feature map.

[0025] In actual application, in one embodiment, the model training step includes the following operations:

[0026] The sample original data are processed by the operations of the original image arrangement step, the image repair step and the logging marking step respectively to obtain a sample electrical imaging restoration image and a sample electrical imaging mask image;

[0027] Divide the sample training set and sample verification set according to the requirements;

[0028] The sample electrical imaging restoration image and the sample electrical imaging mask image in the sample training set are input into the initial well logging information prediction model network, and the network evaluation index is iteratively calculated according to the set loss function;

[0029] The parameters of the prediction model are repeatedly optimized by combining the calculation results of the loss function and the matching calculation results of the sample validation set until the set requirements are met. The obtained prediction model with the optimal parameters is used as the final well logging information prediction model.

[0030] Furthermore, in one embodiment, the fusion calculation step includes:

[0031] The logging information prediction map output by the model is fused with the electrical imaging restoration image to be identified to obtain the logging information recognition result map. Then, the percentage of different types of logging information is calculated based on the logging information recognition result map, and the two are combined as the logging interpretation results.

[0032] Based on the method described in any one or more of the above embodiments, the present invention further provides a storage medium storing program codes that can implement the method described in any one or more of the above embodiments.

[0033] Based on other aspects of the method described in any one or more of the above embodiments, the present invention also provides an electrical imaging image logging information recognition system based on deep learning, which executes the method described in any one or more of the above embodiments.

[0034] Compared with the closest prior art, the present invention also has the following beneficial effects:

[0035] The present invention provides a method and system for recognizing electrical imaging well logging information based on deep learning. This method amplifies the acquired original electrical imaging images and then numbers them to support subsequent processing. This fundamentally ensures the reliability and portability of the data source. The numbering information helps to determine the matching of other image data derived from the original images with each original image in subsequent steps.

[0036] Furthermore, the solution of the present invention uses a generative adversarial network to repair the image, and processes the electrical imaging restoration image through the logging marking step, while obtaining a matching electrical imaging mask image, automatically repairing and marking the logging information in the image, and improving the integrity and intuitiveness of the image feature information;

[0037] The present invention also performs a series of processing on the input marked restored image and mask image through the logging information prediction model to obtain a logging information prediction map, thereby realizing automatic recognition of logging electrical imaging information, ensuring recognition efficiency while improving the accuracy of logging interpretation. Compared with the existing logging phase classification, the present invention can intuitively reflect the analysis and recognition data of various different logging information.

[0038] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0040] Figure 1 1 is a flow chart of a method for recognizing electrical imaging image logging information based on deep learning in one embodiment of the present invention;

[0041] Figure 2 This is a detailed diagram of the implementation process of a method for recognizing electrical imaging image logging information based on deep learning in another embodiment of the present invention;

[0042] Figure 3 This is a diagram showing the principle of image restoration in the method for recognizing electrical imaging image logging information provided by an embodiment of the present invention;

[0043] Figure 4 This is an example of conglomerate information marking in the electrical imaging image logging information recognition method provided by one embodiment of the present invention;

[0044] Figure 5 Schematic diagram of batch conversion of conglomerate logging information in the method for identifying electrical imaging image logging information provided by an embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of the conglomerate logging image repair principle in the electrical imaging image logging information recognition method provided by another embodiment of the present invention;

[0046] Figure 7 This is an example of a conglomerate logging information prediction diagram in the electrical imaging image logging information recognition method provided in one embodiment of the present invention;

[0047] Figure 8 Schematic diagram of the training principle of the well logging information prediction model in the method for recognizing well logging information from an electrical imaging image provided by an embodiment of the present invention;

[0048] Figure 9 This is an example of a conglomerate logging information recognition result diagram in the electrical imaging image logging information recognition method provided in another embodiment of the present invention;

[0049] Figure 10 It is a structural diagram of the electrical imaging image logging information recognition system based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will describe in detail the implementation methods of the present invention in conjunction with the accompanying drawings and embodiments, so that practitioners of the present invention can fully understand how the present invention applies technical means to solve technical problems and achieve the implementation process of technical effects, and can implement the present invention in accordance with the above implementation process. It should be noted that as long as no conflict exists, the various embodiments and various features of each embodiment in the present invention can be combined with each other, and the resulting technical solutions are all within the scope of protection of the present invention.

[0051] Although the flowcharts depict the operations as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. The order of the operations can be rearranged. A process can be terminated when its operations are completed, but can also have additional steps not included in the figures. A process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0052] Computer devices include user devices and network devices. User devices or clients include, but are not limited to, computers, smartphones, PDAs, and the like; network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud computing-based cloud consisting of a large number of computers or network servers. Computer devices can operate independently to implement the present invention, or they can connect to a network and interact with other computer devices in the network to implement the present invention. Networks in which computer devices reside include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, VPN networks, and the like.

[0053] The terms "first," "second," and the like may be used herein to describe various elements, but these elements should not be limited by these terms, and these terms are used merely to distinguish one element from another. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items. When an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or intervening elements may be present.

[0054] The terms used herein are intended only to describe specific embodiments and are not intended to limit exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms "a", "an", "an item" used herein are also intended to include the plural. It should also be understood that the terms "comprise" and / or "include" used herein specify the presence of stated features, integers, steps, operations, units and / or components, and do not preclude the presence or addition of one or more other features, integers, steps, operations, units, components and / or combinations thereof.

[0055] In the field of oil and gas exploration, well logging data offers the highest vertical resolution and greatest continuity of geological data available to date, containing a wealth of usable information. Electrical imaging logging converts measured physical information, such as the resistivity of wellbore formations, into electrical imaging data, providing a more intuitive representation of geological information. However, traditional well logging interpretation methods often rely on manual fitting or subjective expert judgment. Manual interpretation, in addition to being labor-intensive, inevitably results in accuracy that varies depending on the interpreter's knowledge and proficiency. This impacts accurate results and falls short of practical needs. To minimize the impact of human factors on geological analysis, it is necessary to combine manual judgment with novel methods.

[0056] Existing deep learning-based methods for automatic recognition of electrical imaging images primarily build a deep learning model composed of convolutional neural networks. These models are then trained by feeding electrical imaging images into the deep learning model. The images to be identified are then identified and the logging facies classification is directly output. While this method can improve the efficiency of well logging interpretation to a certain extent, its sole reliance on deep learning models to identify complex electrical imaging images does not guarantee reliable results. Furthermore, the direct output of the logging facies classification does not provide a clear understanding of the analysis and identification process for various logging information, thus failing to provide more information for subsequent manual analysis and judgment.

[0057] To solve the above problems, the present invention provides a method and system for identifying well logging information from electrical imaging images based on deep learning. The method pre-constructs a well logging information prediction model for electrical imaging images through set logic, and performs a series of processing on the collected original logging electrical imaging images according to the set strategy and then uses them as input data for the constructed model to obtain a well logging prediction map. The well logging information prediction map output by the model is then fused and calculated with the image before model prediction based on matching pixel conditions, thereby directly and automatically obtaining reliable well logging interpretation results.

[0058] Furthermore, the method for identifying well logging information in electrical imaging images provided by the present invention mainly includes the following operations: collecting and organizing the original electrical imaging images and numbering the original image data; repairing the original electrical imaging images to generate electrical imaging restoration images; marking the electrical imaging restoration images with logging information and converting them into a sample library that can be used for training; activating a pre-built deep learning network model, inputting the electrical imaging restoration images to be identified into the optimal network model for identification, and obtaining a well logging information prediction map; fusing the well logging information prediction map with the electrical imaging restoration images to be identified to obtain the well logging interpretation results. This solution can automatically predict the well logging information in the electrical imaging images and accurately and quantitatively interpret the electrical imaging images. The deep learning network model is an encoder-decoder architecture; the electrical imaging restoration images and mask images in the sample library are input into the network model for training to obtain the optimal network model.

[0059] Next, the detailed process of the method according to the embodiment of the present invention is described in detail based on the accompanying drawings. The steps shown in the flowcharts of the accompanying drawings can be executed in a computer system including, for example, a set of computer-executable instructions. Although the logical order of the steps is shown in the flowcharts, in some cases, the steps shown or described can be executed in a different order than here.

[0060] Example 1

[0061] Figure 1 The flow chart of the method for recognizing electrical imaging image logging information based on deep learning provided by the first embodiment of the present invention is shown. Figure 1 It can be seen that the method includes the following steps.

[0062] The original image arrangement step is to obtain the well logging electrical imaging image in the well logging data as the original image, amplify and number the original image, and associate and store the amplified original image and number information;

[0063] Image restoration step: using a generative adversarial network to restore the original image after amplification to obtain the corresponding electrical imaging restoration image;

[0064] The well logging marking step is to mark the well logging information on the electrical imaging restoration image based on the well logging geological type data in the required file format, and generate and store an electrical imaging mask image that matches the marked well logging information;

[0065] Prediction and recognition step: inputting the marked electrical imaging restoration image and the corresponding electrical imaging mask image into a pre-trained well logging information prediction model, processing the feature data of the input image based on the unique encoder-decoder network structure of the model, and obtaining a corresponding well logging information prediction map;

[0066] A fusion calculation step is performed on the well logging information prediction image and the matched electrical imaging restoration image to obtain well logging interpretation results containing multiple geological types;

[0067] The well logging information prediction model is obtained through a model training step based on a set sample electrical imaging restoration image and a set sample electrical imaging mask image.

[0068] Specifically, in one embodiment, in the original image arrangement step, the process of amplifying the original image includes:

[0069] Data enhancement processing is performed on the collected raw electrical imaging images to obtain amplified raw electrical imaging images; the data enhancement processing includes left-right mirroring, top-down mirroring, and 180° rotation. Data enhancement processing of the raw images can effectively prevent overfitting. Overfitting refers to the situation where a neural network can closely fit the distribution of training data, but the fitted results have low accuracy for test data and lack generalization ability. Furthermore, as the neural network deepens, the number of parameters required to learn also increases, which can easily lead to overfitting. When the data set is small, too many parameters will fit all the characteristics of the data set rather than the commonalities between the data, resulting in overfitting. Therefore, to prevent overfitting, the present invention pre-processes the raw image information when the data set is small. The well logging electrical imaging images processed in this application are expensive to acquire and the total number of images is small, only a few dozen. Therefore, it is necessary to increase the data volume through left-right mirroring, top-down mirroring, and 180° rotation.

[0070] Then, the amplified electrical imaging original images are numbered and sorted, and the numbered amplified electrical imaging original images and a text file of the amplified electrical imaging original image number list are obtained and stored in association.

[0071] Since corresponding electrical imaging restoration images and electrical imaging mask images will be generated for each amplified electrical imaging original image in the subsequent operations of realizing logging information prediction and recognition, setting numbering information for the amplified electrical imaging original image is conducive to quickly and accurately ensuring the matching of the electrical imaging restoration image and the electrical imaging mask image with the original image. It is only necessary to ensure that the numbers are consistent between the two to ensure the accurate matching of each image file, and avoid invalid operations and interference data caused by errors in the input file.

[0072] Furthermore, the image restoration step includes:

[0073] The original image after the amplification process is repaired by using the restoration network and identification network structure of the generative adversarial network, wherein the generative adversarial network includes a restoration network and two identification networks.

[0074] The generative adversarial network uses its restoration network to repair the image and uses its identification network to discriminate the image;

[0075] Specifically, in one embodiment, the two identification networks are a global identification network and a local identification network; the global identification network ensures global consistency of the image restoration area, and the local identification network ensures details of the image restoration area.

[0076] In practical applications, the global identification network takes the complete image as input to identify the global consistency of the scene, while the local identification network ensures the details of the image restoration area;

[0077] The local discriminator uses a set convolutional network structure to encode the original image to obtain an intermediate image whose size meets the requirements of the original image size; then the image is decoded using the set convolutional network structure to obtain a restored image after repair.

[0078] Furthermore, in a preferred embodiment, in the well logging marking step, the well logging geological features in the electrical imaging restoration image are distinguishably marked according to different well logging geological type data, with the JSON format file as the target, and then the generated JSON files are batch converted to obtain the electrical imaging restoration image and the electrical imaging restoration image mask image with matching numbers;

[0079] Among them, different marking forms are set according to different logging geological types of data.

[0080] In practical application, in one embodiment, labelme software can be used to label the electrical imaging restoration image with logging information to obtain a json format file;

[0081] Among them, at least the corresponding logging geological features in the electrical imaging restoration image can be distinguished and marked according to the logging geological types such as conglomerate, hole, laminae, strips and fractures;

[0082] In practical applications, different geological types can be set to use different colors as different marking forms. For example, conglomerate is marked with red, holes are marked with green, laminae are marked with blue, stripes are marked with purple, and fractures are marked with gray.

[0083] Furthermore, the obtained json format files are batch converted to obtain a sample library that can be used for training;

[0084] The sample library comprises at least two parts, the first part is the electrical imaging restoration image, and the second part is the electrical imaging restoration image mask image; the electrical imaging restoration image and the electrical imaging restoration image mask image file numbers correspond one to one.

[0085] Specifically, the electrical imaging restoration image, the electrical imaging restoration image mask image and the label list text file, and the electrical imaging restoration image and the electrical imaging restoration image mask fusion image are associated and stored.

[0086] Furthermore, in the model training step, the network structure of the well logging information prediction model to be constructed is an encoder-decoder architecture;

[0087] For the well logging information prediction model, its encoder includes a basic feature extraction module and a multi-scale parallel pooling module;

[0088] The basic feature extraction module performs convolution and pooling processing on the sample library image to obtain a low-level feature map that meets the requirements;

[0089] The multi-scale parallel pooling module performs multi-scale parallel pooling module processing on the low-level feature map to obtain a fusion map of feature maps of different scales, i.e., a multi-scale feature fusion map;

[0090] The decoder performs transposed convolution processing on the multi-scale feature fusion map to obtain a decoded feature map with a resolution that meets the requirements, and splices the decoded feature map with the low-level feature map of the same resolution to obtain a spliced feature map.

[0091] In practical applications, the decoder can be set to perform transposed convolution processing on the multi-scale feature fusion map to obtain a feature map with a resolution of 1 / 4 of the sample library image, and the feature map is spliced with the feature map of the same resolution generated by the basic feature extraction module to obtain a spliced feature map;

[0092] Furthermore, the well logging information prediction model further performs a transposed convolution process on the obtained spliced feature map to obtain a well logging information prediction map having the same resolution as the electrical imaging restoration image in the sample library;

[0093] Then, the well logging information prediction map is spliced with the electrical imaging restoration image to obtain the electrical imaging restoration image well logging information prediction map, and output it.

[0094] Specifically, in one embodiment, the model training step includes the following operations:

[0095] The sample original data are processed by the operations of the original image arrangement step, the image repair step and the logging marking step respectively to obtain a sample electrical imaging restoration image and a sample electrical imaging mask image;

[0096] Divide the sample training set and sample verification set according to the needs;

[0097] The sample electrical imaging restoration image and the sample electrical imaging mask image in the sample training set are input into the initial well logging information prediction model network, and the network evaluation index is iteratively calculated according to the set loss function;

[0098] The parameters of the prediction model are repeatedly optimized by combining the calculation results of the loss function and the matching calculation results of the sample validation set until the set requirements are met. The obtained prediction model with the optimal parameters is used as the final well logging information prediction model.

[0099] In a specific embodiment, the fusion calculation step includes:

[0100] The logging information prediction map output by the model is fused with the electrical imaging restoration image to be identified to obtain the logging information recognition result map. Then, the percentage of different types of logging information is calculated based on the logging information recognition result map, and the two are combined as the logging interpretation results.

[0101] Attachment Figure 2 A detailed flowchart of the execution flow of a method for recognizing electrical imaging well logging information based on deep learning in one embodiment of the present invention is provided. The process of implementing well logging information recognition is described in detail below with reference to the detailed flowchart, including the following steps:

[0102] P100: Collect and organize the original images of electro-imaging and number the original image data;

[0103] P101: Performing restoration processing on the original electrographic image to generate an electrographic restoration image;

[0104] P102: Marking the electrical imaging restored image with logging information and converting it into a sample library that can be used for training;

[0105] P103: Construct a well logging information prediction model. The network structure of this model is an encoder-decoder architecture.

[0106] P104: Input the restored images and mask images in the sample library into the network model for training to obtain the optimal network model;

[0107] P105: Input the electrical imaging restoration image to be identified into the optimal network model for identification, and obtain the well logging information prediction map;

[0108] P106: The logging information prediction map is fused with the electrical imaging restoration image to be identified to obtain the logging interpretation result.

[0109] In an embodiment of the present invention, in P100, the original electrical imaging images are collected and organized, and data enhancement processing is performed on the collected original electrical imaging images. The data enhancement processing includes three transformations: left-right mirroring, top-bottom mirroring, and 180° rotation, to obtain amplified original electrical imaging images; the amplified original electrical imaging images are numbered and organized to obtain numbered original electrical imaging images and a text file of amplified original electrical imaging image number list.

[0110] In the embodiment of the present invention, in P101, a generative adversarial network is used to repair the original image of the electrical imaging to generate a restored image of the electrical imaging, such as Figure 3As shown. The generative adversarial network consists of a restoration network and two discrimination networks. The restoration network is used to restore the image that enters it, while the discrimination network is used to distinguish the image. The goal of the discrimination network is to classify the restored image and the real image as accurately as possible, while the restoration network is to fool the discrimination network as much as possible, making it unable to accurately distinguish the restored image from the real image, thereby improving the quality of image restoration. The two discrimination networks are a global discrimination network and a local discrimination network. The global discrimination network ensures the global consistency of the image restoration area, and the local discrimination network ensures the details of the image restoration area.

[0111] Specifically, in one embodiment, the local discriminator network architecture can be as follows: To generate a portion of an image, a 12-layer convolutional network is used to encode the original image, resulting in an image 1 / 16th the size of the original image. This image is then decoded using a 4-layer convolutional network to obtain a restored image. The global discriminator takes the full image as input and identifies global consistency in the scene, while the local discriminator observes only a 1 / 4th the size of the original image centered on the padded area to identify local consistency.

[0112] On the other hand, in other embodiments of the present invention, during the image restoration process, image data restoration can also be achieved based on a loss function. The loss function is usually set to use the square of the difference between the restored image and the image before restoration, that is, E(x; x0) = ||(x-x0)|| 2 , where the image before restoration is x0 and the image after restoration is x. The network parameters are adjusted through multiple iterations to make the loss function converge. In this application, the loss function is set to E(x; x0) = ||(x-x0)⊙m|| 2 , where the mask image is m. The role of the mask image is that the loss function only calculates the error of the image part except the blank strip, and the blank strip part does not participate in the calculation.

[0113] In the embodiment of the present invention, in P102, labelme software is used to mark the electrical imaging restoration image with logging information. The logging information includes conglomerate, holes, laminae, stripes, and fractures. Conglomerate is marked in red, holes in green, laminae in blue, stripes in purple, and fractures in gray. Figure 4 To mark the conglomerate information. After marking, a json format file is obtained. The json format file is batch converted to obtain four files, namely, the electrical imaging restoration image, the electrical imaging restoration image mask image, the label list text file, and the electrical imaging restoration image and the electrical imaging restoration image mask fusion image. Figure 5As shown, the electrical imaging restoration image and the electrical imaging restoration image mask image are taken as the sample library for training the network, and the electrical imaging restoration image and the electrical imaging restoration image mask image file numbers correspond one to one.

[0114] Among them, the mask image is usually used to reflect the blank strip part in the well logging electrical imaging image with blank strips. In actual application, the mask image is extracted by the following operation: the RGB pixel value of the blank strip part of the electrical imaging image is set to (0, 0, 0), and the image pixel value of the non-blank strip part is set to (255, 255, 255), so that the mask image of the well logging image can be generated.

[0115] In the embodiment of the present invention, the network model used in P103 is an encoder-decoder deep learning network architecture, and the encoder includes a basic feature extraction module and a multi-scale parallel pooling module. The basic feature extraction module performs convolution and pooling processing on the sample library image, and after performing the first convolution group operation on the input sample library image, a resolution of the original image is obtained. Figure 1 / 2 feature map, and then perform the second convolution group operation to obtain the original resolution Figure 1 / 4 feature map, and then perform the third convolution group operation to obtain the original resolution Figure 1 / 8 feature map, and finally perform the fourth convolution group operation to obtain the original resolution Figure 1 / 16 feature map, i.e., a low-level feature map with an output stride of 16. This low-level feature map is processed by a multi-scale parallel pooling module, which employs four parallel convolution operations: the first convolution operation uses a 1×1 convolution kernel, the second convolution operation uses a 3×3 convolution kernel with a dilation rate of rate = 6, the third convolution operation uses a 3×3 convolution kernel with a dilation rate of rate = 12, and the fourth convolution operation uses a 3×3 convolution kernel with a dilation rate of rate = 18. After performing the four convolution operations on the image, four feature maps of different scales are obtained. These four feature maps are fused to form a multi-scale feature fusion map.

[0116] The decoder performs transposed convolution processing on the multi-scale feature fusion map, and obtains a resolution of the original Figure 1 A feature map of 1 / 4, i.e., a feature map with a decoder output stride of 4, is spliced with a feature map of the same resolution generated by the basic feature extraction module to obtain a spliced feature map. The spliced feature map is subjected to transposed convolution processing to obtain a well logging information prediction map with the same resolution as the sample library image.

[0117] In the embodiment of the present invention, in P104, the electrical imaging restoration image and the mask image in the sample library are input into the network model for training, such as Figure 5, the training process of the network model includes:

[0118] (1) Data reading: The data in the sample library is divided into two parts: training set and validation set. The number of training set accounts for 87% of the total number of sample library data, and the number of validation set accounts for 13% of the total number of sample library data.

[0119] (2) supervised learning process, such as Figure 9 As shown, the training set data is input into the network model described in P103, the input data is processed through the encoder-decoder network architecture, and the loss function of the network model is iteratively calculated on the input data.

[0120] (3) The loss function selects the mean intersection-over-union (mIOU) as the evaluation index of the model. The confusion matrix is calculated through training. The rows of the confusion matrix are the true values of the logging information class, and the columns are the predicted values of the logging information class. The sum of each row is the number of true samples of the logging information class, and the sum of each column is the number of samples predicted to be the logging information class. The calculation formula of mIOU is as follows:

[0121]

[0122] Among them, k is the number of information categories, i represents the true value, j represents the predicted value, and p ij It means predicting i as j.

[0123] (4) In order to find the best model, the validation set is used to verify the accuracy of the output label of the current model after each epoch, and to adjust the parameters of the network model during the next epoch training.

[0124] (5) Repeat the above process until the set number of training times or the loss function no longer changes. The network model parameters at this time are the optimal network model parameters.

[0125] (6) The optimal network model is obtained and the training is completed.

[0126] In the embodiment of the present invention, before executing P105, the original image of the electric imaging to be identified is first input into the generative adversarial network, and the original image of the electric imaging to be identified is repaired to generate a restored image of the electric imaging to be identified, such as Figure 6 Then the electrical imaging restoration image to be identified is input into the optimal network model trained by P104 to obtain the well logging information prediction map, as shown in Figure 7 As shown in the figure, it can be seen that the conglomerate information in the electrical imaging restoration image to be identified can be effectively identified by the network model.

[0127] In the embodiment of the present invention, in P106, the well logging information prediction image obtained in P105 is fused with the electrical imaging restoration image to be identified, such as Figure 8 As shown in the figure, the logging information recognition result diagram is obtained. The percentage of the five types of logging information in the logging information recognition result diagram, namely gravel, holes, laminae, stripes, and fractures, is calculated. In specific applications, the areas of the identified gravel, holes, laminae, stripes, and fractures can be calculated through contour monitoring to obtain their percentages. The logging information recognition results can then be quantified to obtain the logging interpretation results.

[0128] The accompanying drawings only directly illustrate the method by taking the electrical imaging image containing gravel as an example. The automatic recognition of the logging information of holes, laminae, stripes and fractures is also included in the method.

[0129] For simplicity of description, the aforementioned method embodiments are described as a series of actions. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, as certain steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also be aware that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily required for the present invention.

[0130] By adopting the scheme described in one or more of the above embodiments of the present invention, compared with the existing logging image analysis technology, not only the automatic recognition of logging electrical imaging information is realized, but also the limitations of the available data of logging phase category information are overcome. Furthermore, the present invention performs fusion calculations based on the logging information prediction map automatically identified and output by the model, which can automatically mark the logging information and generate logging data calculation results corresponding to different logging geological types. The traditional manual qualitative interpretation is transformed into computer quantitative judgment, which improves the automation performance of the electrical imaging image interpretation process. On the basis of improving the efficiency of electrical imaging logging information analysis, it further improves the accuracy and comprehensiveness of information analysis, providing convenient and reliable information for subsequent manual observation and analysis.

[0131] It should be pointed out that in other embodiments of the present invention, the method can also obtain a new deep learning-based electrical imaging image logging information recognition method by combining one or several of the above embodiments, so as to realize the recognition, analysis and application of various geological information contained in the logging electrical imaging.

[0132] It should be noted that, based on the method in any one or more of the above-mentioned embodiments of the present invention, the present invention also provides a storage medium, which stores program code that can implement the method as described in any one or more of the above-mentioned embodiments. When the code is executed by the operating system, it can implement the electrical imaging image logging information recognition method based on deep learning as described above.

[0133] Example 2

[0134] The methods disclosed in the above embodiments of the present invention are described in detail. The methods of the present invention can be implemented using various devices or systems. Therefore, based on other aspects of the methods described in any one or more of the above embodiments, the present invention further provides a deep learning-based electrical imaging image logging information recognition system for executing the deep learning-based electrical imaging image logging information recognition method described in any one or more of the above embodiments. Specific embodiments are provided below for detailed description.

[0135] Specifically, Figure 10 FIG. 1 shows a schematic diagram of the structure of the electrical imaging image logging information recognition system based on deep learning provided in an embodiment of the present invention. Figure 10 As shown, the system includes:

[0136] an original image arrangement module configured to obtain a well logging electrical imaging image from the well logging data as an original image, amplify and number the original image, and associate and store the amplified original image and number information;

[0137] An image restoration module is configured to use a generative adversarial network to restore the original image after the amplification process to obtain a corresponding electrical imaging restoration image;

[0138] A well logging marking module is configured to mark the well logging information on the electrical imaging restoration image based on the well logging geological type data in accordance with the required file format, and to generate and store the electrical imaging mask image that matches the marked well logging information;

[0139] A prediction and recognition module is configured to input the marked electrical imaging restoration image and the corresponding electrical imaging mask image into a pre-trained well logging information prediction model, and process the feature data of the input image based on the unique encoder-decoder network structure of the model to obtain a corresponding well logging information prediction map;

[0140] a fusion calculation module configured to perform fusion calculation on the well logging information prediction image and the matched electrical imaging restoration image to obtain well logging interpretation results containing multiple geological types;

[0141] The well logging information prediction model is obtained by training using a model training module based on a set sample electrical imaging restoration image and a sample electrical imaging mask image.

[0142] Furthermore, in one embodiment, the original image arrangement module performs an amplification process on the original image by:

[0143] The collected electrical imaging original image is subjected to data enhancement processing to obtain an amplified electrical imaging original image; the data enhancement processing includes left-right mirroring, top-bottom mirroring and rotation transformation processing.

[0144] In one embodiment, the image restoration module is specifically configured as follows:

[0145] The original image after the amplification process is repaired by using the restoration network and identification network structure of the generative adversarial network, wherein the generative adversarial network includes a restoration network and two identification networks.

[0146] Furthermore, in one embodiment, the image restoration module sets its two identification networks of the generative adversarial network to be a global identification network and a local identification network;

[0147] The global identification network takes the complete image as input to identify the global consistency of the scene, and the local identification network ensures the details of the image repair area;

[0148] The local discriminator uses a set convolutional network structure to encode the original image to obtain an intermediate image whose size meets the requirements of the original image size; then the image is decoded using the set convolutional network structure to obtain a restored image after repair.

[0149] Specifically, in one embodiment, the well logging marking module is configured as follows:

[0150] Taking the json format file as the target, the logging geological features in the electrical imaging restoration image are distinguished and marked according to different logging geological type data, and then the generated json files are batch converted to obtain the electrical imaging restoration image and the electrical imaging restoration image mask image with matching numbers;

[0151] Among them, different marking forms are set according to different logging geological types of data.

[0152] Specifically, in one embodiment, the model training module is configured as follows:

[0153] The encoder of the well logging information prediction model is configured to include a basic feature extraction module and a multi-scale parallel pooling module;

[0154] The basic feature extraction module performs convolution and pooling processing on the sample library image to obtain a low-level feature map that meets the requirements;

[0155] The multi-scale parallel pooling module performs multi-scale parallel pooling module processing on the low-level feature map to obtain a fusion map of feature maps of different scales, i.e., a multi-scale feature fusion map;

[0156] A decoder is set to perform transposed convolution processing on the multi-scale feature fusion map to obtain a decoded feature map with a resolution that meets the requirements, and the decoded feature map is spliced with the low-level feature map with the same resolution to obtain a spliced feature map.

[0157] Furthermore, the model training module obtains the well logging information prediction model through the following operations:

[0158] The sample original data are processed by the operations of the original image arrangement step, the image repair step and the logging marking step respectively to obtain a sample electrical imaging restoration image and a sample electrical imaging mask image;

[0159] Divide the sample training set and sample verification set according to the needs;

[0160] The sample electrical imaging restoration image and the sample electrical imaging mask image in the sample training set are input into the initial well logging information prediction model network, and the network evaluation index is iteratively calculated according to the set loss function;

[0161] The parameters of the prediction model are repeatedly optimized by combining the calculation results of the loss function and the matching calculation results of the sample validation set until the set requirements are met. The obtained prediction model with the optimal parameters is used as the final well logging information prediction model.

[0162] In one embodiment, the fusion calculation module is specifically configured as follows:

[0163] The logging information prediction map output by the model is fused with the electrical imaging restoration image to be identified to obtain the logging information recognition result map. Then, the percentage of different types of logging information is calculated based on the logging information recognition result map, and the two are combined as the logging interpretation results.

[0164] In the deep learning-based electrical imaging image logging information recognition system provided by an embodiment of the present invention, each module or unit structure can operate independently or in combination according to actual analysis and calculation requirements to achieve corresponding technical effects.

[0165] It should be understood that the embodiments disclosed herein are not limited to the specific structures, processing steps, or materials disclosed herein, but should extend to equivalent substitutions of these features understood by those skilled in the relevant art. It should also be understood that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting.

[0166] The phrase "one embodiment" mentioned in the specification means that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Therefore, the phrase "one embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0167] Although the embodiments disclosed herein are as described above, the contents described herein are merely embodiments for facilitating understanding of the present invention and are not intended to limit the present invention. Any person skilled in the art may make any modifications and variations in the form and details of the embodiments without departing from the spirit and scope of the present invention. However, the scope of patent protection of the present invention shall remain subject to the scope defined by the appended claims.

Claims

1. A method for identifying electrical imaging logging information based on deep learning, characterized in that: The method comprises: The original image arrangement step is to obtain the well logging electrical imaging image in the well logging data as the original image, amplify and number the original image, and associate and store the amplified original image and number information; Image restoration step: using a generative adversarial network to restore the original image after amplification to obtain the corresponding electrical imaging restoration image; The well logging marking step is to mark the well logging information on the electrical imaging restoration image based on the well logging geological type data in the required file format, and generate an electrical imaging mask image that matches the marked well logging information; Prediction and recognition step: inputting the marked electrical imaging restoration image and the corresponding electrical imaging mask image into a pre-trained well logging information prediction model, processing the feature data of the input image based on the encoder-decoder network structure of the model to obtain a corresponding well logging information prediction map; A fusion calculation step is performed on the well logging information prediction image and the matched electrical imaging restoration image to obtain well logging interpretation results containing multiple geological types; The encoder of the well logging information prediction model includes a basic feature extraction module and a multi-scale parallel pooling module; The basic feature extraction module performs convolution and pooling processing on the sample library image to obtain a low-level feature map that meets the requirements; The multi-scale parallel pooling module performs multi-scale parallel pooling module processing on the low-level feature map to obtain a fusion map of feature maps of different scales, i.e., a multi-scale feature fusion map; The decoder performs transposed convolution processing on the multi-scale feature fusion map to obtain a decoded feature map with a resolution that meets the requirements, and splices the decoded feature map with the low-level feature map of the same resolution to obtain a spliced feature map; The well logging information prediction model is obtained by training based on a set sample electrical imaging restoration image and a sample electrical imaging mask image through a model training step; The model training step includes the following operations: The sample original data are processed by the operations of the original image arrangement step, the image repair step and the logging marking step respectively to obtain a sample electrical imaging restoration image and a sample electrical imaging mask image; Divide the sample training set and sample verification set according to the needs; The sample electrical imaging restoration image and the sample electrical imaging mask image in the sample training set are input into the initial well logging information prediction model network, and the network evaluation index is iteratively calculated according to the set loss function; The parameters of the prediction model are repeatedly optimized by combining the calculation results of the loss function and the matching calculation results of the sample validation set until the set requirements are met. The obtained prediction model with the optimal parameters is used as the final well logging information prediction model.

2. The method according to claim 1, wherein In the original image arrangement step, the process of amplifying the original image includes: The collected electrical imaging original image is subjected to data enhancement processing to obtain an amplified electrical imaging original image; the data enhancement processing includes left-right mirroring, top-bottom mirroring and rotation transformation processing.

3. The method according to claim 1, wherein The image restoration step includes: The original image after the amplification process is repaired by using the restoration network and identification network structure of the generative adversarial network, wherein the generative adversarial network includes a restoration network and two identification networks.

4. The method according to claim 3, wherein The two identification networks are the global identification network and the local identification network; The global identification network takes the complete image as input to identify the global consistency of the scene, and the local identification network ensures the details of the image repair area; The local identification network uses a set convolutional network structure to encode the original image to obtain an intermediate image whose size meets the requirements of the original image size; then the image is decoded using the set convolutional network structure to obtain a restored image after repair.

5. The method according to claim 1, wherein In the well logging marking step, the well logging geological features in the electrical imaging restoration image are distinguished and marked according to different well logging geological type data, with the JSON format file as the target. Then, the generated JSON files are batch converted to obtain the electrical imaging restoration image and the electrical imaging restoration image mask image with matching numbers. Among them, different marking forms are set according to different logging geological types of data.

6. The method according to claim 1, wherein The model training step includes the following operations: The sample original data are processed by the operations of the original image arrangement step, the image repair step and the logging marking step respectively to obtain a sample electrical imaging restoration image and a sample electrical imaging mask image; Divide the sample training set and sample verification set according to the needs; The sample electrical imaging restoration image and the sample electrical imaging mask image in the sample training set are input into the initial well logging information prediction model network, and the network evaluation index is iteratively calculated according to the set loss function; The parameters of the prediction model are repeatedly optimized by combining the calculation results of the loss function and the matching calculation results of the sample validation set until the set requirements are met. The obtained prediction model with the optimal parameters is used as the final well logging information prediction model.

7. A storage medium, characterized in that: The storage medium stores program code that can implement the method according to any one of claims 1 to 6.

8. An electrical imaging image logging information recognition system based on deep learning, characterized in that: The system executes the method according to any one of claims 1 to 6.

Images