Logging-while-drilling image stratum classification and identification method in complex drilling scene

Abstract

The invention relates to the technical field of logging image stratum classification and recognition, in particular to a logging-while-drilling image stratum classification and recognition method under a complex drilling scene, and the method comprises the steps: collecting electric imaging logging images of different stratums, extracting foreground images of the electric imaging logging images, and recording the foreground images as logging images; according to the distribution characteristics of the pixel points in the logging image, performing denoising processing on the logging image to obtain a denoised logging image; and carrying out enhancement processing on all the denoised logging images by adopting an image enhancement algorithm to obtain a deep learning model for carrying out classification and identification on the logging-while-drilling image strata. By analyzing the distribution characteristics of the pixel values of the pixel points in the electrical imaging logging image, the electrical imaging logging image is denoised more accurately, and the accuracy of classification and identification of the stratum to which the logging image belongs by the deep learning model obtained by neural network training is improved.

Description

Technical Field

[0001] The present application relates to the technical field of formation classification and recognition of logging images, and in particular to a formation classification and recognition method of logging while drilling images in complex drilling scenarios. Background Art

[0002] Well logging while drilling is the main means of geophysical exploration. It can send the logging images collected during the drilling process to the ground for processing in real time. Compared with conventional logging curves, it has the advantages of higher resolution and visualization. Among them, electrical imaging logging images are the most widely used of all well logging while drilling methods. It can achieve more accurate and efficient geological information detection and improve the exploration and development level of complex oil and gas reservoirs. In the prior art, convolutional neural networks are usually used to realize the formation classification and recognition of electrical imaging logging images while drilling. However, in the actual acquisition process of electrical imaging logging images, the electrical imaging logging instrument will be affected by the collision with the well wall, the electronic components and circuits of the instrument, etc., resulting in different degrees of noise in the electrical imaging logging images. Therefore, in order to reduce the influence of noise in the electrical imaging logging images on the subsequent formation classification and recognition of logging images, the collected electrical imaging logging images are usually pre-processed.

[0003] In the preprocessing process, when the electrical imaging logging images of carbonate formations with complex non-reservoir layers are used to construct their formation classification and recognition model, the existing preprocessing methods of electrical imaging logging images ignore the similarity of image distribution characteristics between the dense layer in the carbonate formation and the point noise in the electrical imaging logging images. At the same time, the image halo generated by the current focusing effect during the acquisition of the electrical imaging logging images is not considered. This is an illusion that does not exist on the actual core or well wall. This will easily cause the preprocessed electrical imaging logging images to lose more dense layer geological information and retain more illusion information that has no geological significance, thereby reducing the accuracy of the deep learning model obtained by neural network training in classifying and identifying the formation to which the logging images belong. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a method for classification and recognition of formations in logging while drilling images in complex drilling scenarios to solve the existing problems.

[0005] The present invention adopts the following technical solution for the classification and recognition of formations in a LWD image in a complex drilling scenario: An embodiment of the present application provides a method for classification and recognition of formations in logging while drilling images in a complex drilling scenario, the method comprising the following steps: S1: Collect electrical imaging logging images of different formations, and extract the foreground image of each electrical imaging logging image, which is recorded as the logging image; S2: According to the distribution characteristics of the pixels in the logging image, the logging image is denoised, specifically: S201: using a connected domain extraction algorithm and a convex hull algorithm to extract all convex hulls in each well logging image, counting the number of all pixel points in each convex hull, and obtaining all feature convex hulls in each well logging image; S202: extracting the central pixel point of each feature convex hull, and determining the density concentration of each feature convex hull by analyzing the concentrated distribution of the central pixel points of all feature convex hulls; S203: constructing a window with the central pixel point of each characteristic convex hull as the center, determining the characteristic distribution value of each characteristic convex hull by analyzing the distribution of all pixel points in the window, and determining the density characteristic value of each characteristic convex hull in combination with the density concentration, so as to screen out the dense layer convex hull from all characteristic convex hulls of each logging image, and denoising each logging image to obtain a denoised logging image; S3: All denoised logging images are enhanced using an image enhancement algorithm to obtain a deep learning model for classifying and identifying formations in LWD images.

[0006] Preferably, the method for obtaining all feature convex hulls in each logging image is: The number of all pixels in each convex hull is taken as the area of ​​each convex hull; The area of ​​all convex hulls in each logging image is used as the input of the threshold segmentation algorithm, and the segmentation threshold is output, which is recorded as the area threshold. All convex hulls with an area smaller than the area threshold are used as the feature convex hull of each logging image.

[0007] Preferably, extracting the central pixel point of each feature convex hull includes: Get the minimum circumscribed rectangle of each feature convex hull, and use the pixel point corresponding to the center of the minimum circumscribed rectangle as the center pixel point of each feature convex hull.

[0008] Preferably, the method for determining the density concentration of each feature convex hull is: The central pixel points of all feature convex hulls in each logging image are used as the input of the density clustering algorithm, and the local density of the central pixel points in all feature convex hulls is output as the density concentration of all feature convex hulls.

[0009] Preferably, the method for determining the feature distribution value of each feature convex hull is: The pixel values ​​of all pixels in the window of each feature convex hull are used as the input of the nearest distance algorithm, and the empirical distribution function of each feature convex hull is output; the uniform distribution function of each feature convex hull is constructed according to the pixel values ​​of all pixels in the window of each feature convex hull; The single sample test algorithm is used to calculate the statistic between the empirical distribution function and the uniform distribution function of each feature convex hull as the feature distribution value of each feature convex hull.

[0010] Preferably, the density characteristic value of each characteristic convex hull is the ratio between the density concentration and the characteristic distribution value of each characteristic convex hull.

[0011] Preferably, the process of selecting the dense layer convex hull from all feature convex hulls of each well logging image is: The density eigenvalues ​​of all feature convex hulls in each logging image are used as the input of the threshold segmentation algorithm, and the segmentation threshold is output, which is recorded as the eigenvalue threshold. The feature convex hull with a density eigenvalue greater than the eigenvalue threshold is used as the dense layer convex hull of each logging image.

[0012] Preferably, the process of performing denoising on each well logging image to obtain a denoised well logging image is: In each well logging image, all pixel points in the convex hull except the convex hull of the dense layer are used as the input of the filtering algorithm, and the denoised well logging image is output; A background image of each electrical imaging logging image is obtained, and the background image is merged with the corresponding denoised logging image to obtain a denoised logging image.

[0013] Preferably, the image enhancement algorithm is a fast Retinex image enhancement algorithm.

[0014] Preferably, the step of acquiring a deep learning model for classifying and identifying formations in logging while drilling images includes: All enhanced denoised logging images are labeled and classified according to the geological categories of carbonate formations using open source power. The labeled denoised logging images are used as the input of the neural network to output the deep learning model.

[0015] This application has at least the following beneficial effects: The present application firstly screens out characteristic convex hulls with smaller areas, thereby effectively distinguishing geological features from noise in electrical imaging logging images; further, by analyzing the distribution density of characteristic convex hulls in logging images, a density concentration is constructed, which is helpful to evaluate the significance of geological features, thereby identifying characteristic convex hulls representing dense layers; further, a window is constructed with the central pixel point of each characteristic convex hull as the center, and a density characteristic value is constructed by analyzing the distribution of all pixel points in the window and combining the density concentration, which can more accurately screen out the convex hull of the dense layer, thereby retaining important geological information in the denoising process; the present application comprehensively evaluates the density concentration and characteristic distribution value of the characteristic convex hull, accurately denoises and enhances the electrical imaging logging images, reduces the misidentification and missed identification of geological features, and enhances the adaptability and robustness of the deep learning model to complex drilling scenarios, thereby improving the accuracy of the deep learning model trained by the neural network in classifying and identifying the strata to which the logging images belong. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0017] Figure 1 A flowchart of a method for classification and identification of formations in a while drilling logging image in a complex drilling scenario provided by an embodiment of the present application; Figure 2 A schematic diagram of a logging image denoising process provided in one embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, describes in detail the specific implementation method, structure, features and effects of the LWD image formation classification and recognition method in a complex drilling scenario proposed by the present application. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0020] The following is a detailed description of a method for classification and identification of formations in LWD images in a complex drilling scenario provided by the present application in conjunction with the accompanying drawings.

[0021] An embodiment of the present application provides a method for classifying and identifying formations in logging while drilling images in a complex drilling scenario. Specifically, the following method for classifying and identifying formations in logging while drilling images in a complex drilling scenario is provided. Figure 1 , the method comprises the following steps: S1: Collect electrical imaging logging images of different formations, and extract the foreground image of each electrical imaging logging image, which is recorded as the logging image.

[0022] Electrical imaging logging images with different carbonate formation geological characteristics are collected, a total of N electrical imaging logging images are collected, and all electrical imaging logging images are mapped into grayscale images. The carbonate formation geological characteristics mainly include seven types of characteristics: bedding, dense layer, mud band, suture line, induced fracture, pyrite and flint nodule.

[0023] It should be noted that the value of the number N of electrical imaging logging image acquisition is manually set. In this embodiment, the value of N is 3000. The implementer can also set it by himself according to the specific situation. This embodiment does not impose any special restrictions.

[0024] Since geological features with low resistivity appear as low-brightness features in electrical imaging logging images, and dense layers in carbonate formations have higher resistivity values ​​in conventional logging data, dense layers have higher grayscale values ​​in electrical imaging logging images and appear as higher-brightness features.

[0025] Therefore, each electrical imaging logging image is used as the input of the threshold segmentation algorithm, and the foreground image of each electrical imaging logging image is output to characterize the image area with higher resistivity value in the electrical imaging logging image.

[0026] It should be noted that there are many commonly used threshold segmentation algorithms. In this embodiment, the Otsu threshold segmentation algorithm is used to extract the foreground image of the electrical imaging logging image. In actual application, as other implementation methods, the implementer may also use other methods. Regarding the selection of the threshold segmentation method, this embodiment does not make any special restrictions.

[0027] In addition, it should be supplemented that, in this embodiment, all threshold segmentation algorithms involved are Otsu threshold segmentation algorithms.

[0028] Among them, the specific principle of selecting the Otsu threshold segmentation algorithm will not be repeated here.

[0029] S2: De-noising the well logging image according to the distribution characteristics of the pixels in the well logging image.

[0030] In the prior art, point noise in the electrical imaging logging image is suppressed by performing median filtering on the electrical imaging logging image. However, the dense layer in the carbonate formation (dense lithological formation in the formation) is usually shown as densely packed and evenly arranged point-like layer segments in the electrical imaging logging image. Therefore, directly using the median filtering algorithm to denoise the electrical imaging logging image can easily cause the point-like geological details belonging to the dense layer in the electrical imaging logging image to be mistaken for point-like noise and filtered out.

[0031] Therefore, in order to avoid the situation where the point-like geological details belonging to the dense layer in the electrical imaging logging image are mistakenly regarded as point-like noise and filtered out during the image denoising process, the following processing is performed. The specific process is as follows: S201: using a connected domain extraction algorithm and a convex hull algorithm to extract all convex hulls in each well logging image, counting the number of all pixel points in each convex hull, and obtaining all feature convex hulls in each well logging image.

[0032] A connected domain extraction algorithm is used to extract all connected domains in each logging image, and a convex hull algorithm is used to obtain the convex hull of each connected domain in the foreground image, so as to obtain all convex hulls in the foreground image of each electrical imaging logging image, so as to ensure that all image areas with higher resistivity values ​​in the electrical imaging logging image are obtained.

[0033] It should be understood that there are many commonly used connected domain extraction algorithms. In this embodiment, a region growing algorithm is used to obtain a connected domain in a foreground image. In actual application, as other implementation methods, implementers may also use other methods such as morphological operations. Regarding the selection of a connected domain extraction algorithm, this embodiment does not impose any special restrictions.

[0034] It should be noted that there are many commonly used convex hull algorithms. In this embodiment, the Andrew algorithm is used to obtain the convex hull in the foreground image. In actual application, as other implementation methods, the implementer may also use other convex hull algorithms such as Jarvis algorithm and Craham algorithm according to specific circumstances. Regarding the selection of convex hull algorithm, this embodiment does not impose any special restrictions.

[0035] Among them, the region growing algorithm and the Andrew algorithm are both well-known technologies, and the specific process of extracting the connected domain using the region growing algorithm and obtaining the convex hull of the connected domain using the Andrew algorithm will not be described in detail.

[0036] Since the image area corresponding to the point-like geological features of the dense layer in the carbonate formation and the point-like noise in the electrical imaging logging image have a smaller area than the other geological feature areas in the electrical imaging logging image, the number of all pixel points in each convex hull in each logging image is counted as the area of ​​each convex hull, the area of ​​all convex hulls is used as the input of the threshold segmentation algorithm, the segmentation threshold is output, recorded as the area threshold, and all convex hulls with an area less than the area threshold are used as the feature convex hull of each logging image.

[0037] S202: extracting the central pixel point of each feature convex hull, and determining the density concentration of each feature convex hull by analyzing the concentrated distribution of the central pixel points of all feature convex hulls.

[0038] Since the dense layer in the carbonate formation appears as dense and evenly arranged point-like layer segments in the electrical imaging logging image, and the point noise in the electrical imaging logging image usually does not have this feature and is generally randomly distributed, the convex hull of the geological characteristic area belonging to the dense layer in the electrical imaging logging image has a dense and even convex hull distribution in the logging image where the convex hull is located.

[0039] Therefore, the density concentration of each characteristic convex hull can be determined by the density of the convex hull distribution within each convex hull range in the logging image and whether the distribution of the convex hull within each convex hull range is uniform, so as to evaluate whether each convex hull is a convex hull of the geological characteristic area belonging to the dense layer in the electrical imaging logging image, specifically: Get the minimum circumscribed rectangle of each feature convex hull, and use the pixel point corresponding to the center of the minimum circumscribed rectangle as the center pixel point of each feature convex hull.

[0040] The method for obtaining the minimum bounding rectangle is a well-known technique, and the specific obtaining process will not be described in detail.

[0041] The central pixel points of all feature convex hulls in each logging image are used as the input of the density clustering algorithm, and the local density of the central pixel points in all feature convex hulls is output as the density concentration of all feature convex hulls. Among them, the distance between the central pixel points is used as the metric distance of the density clustering algorithm.

[0042] It should be noted that there are many commonly used clustering algorithms. In this embodiment, the DPC density peak clustering algorithm is used to obtain the local density of each data point. In actual application, implementers can also use other clustering methods such as the DBSCAN clustering algorithm. Regarding the selection of clustering algorithms, this embodiment does not make any special restrictions.

[0043] Among them, the DPC density peak clustering algorithm is a well-known technology, and its specific principle will not be repeated here.

[0044] So far, by analyzing the distribution of the feature convex hull, the density concentration of the feature convex hull is obtained.

[0045] S203: construct a window with the central pixel point of each feature convex hull as the center, determine the feature distribution value of each feature convex hull by analyzing the distribution of all pixel points in the window, and determine the density feature value of each feature convex hull in combination with the density concentration, so as to screen out the dense layer convex hull from all the feature convex hulls of each logging image, and denoise each logging image to obtain a denoised logging image.

[0046] (1) In each well logging image, a square window is constructed with the central pixel of each feature convex hull as the center. In this embodiment, the value of the side length M of the window is 5. The implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0047] (2) Taking the pixel values ​​of all pixels within the window of each feature convex hull as the input of the nearest distance algorithm, outputting the empirical distribution function of each feature convex hull; constructing the uniform distribution function of each feature convex hull based on the pixel values ​​of all pixels within the window of each feature convex hull; (3) The single-sample Kolmogorov-Smirnov test algorithm is used to calculate the statistic between the empirical distribution function and the uniform distribution function of each characteristic convex hull as the characteristic distribution value of each characteristic convex hull; it is used to evaluate whether the distribution of the convex hull in the local image area where the convex hull is located in the electrical imaging well logging image obeys the uniform distribution. The larger the statistic, the less the distribution of the convex hull in the local image area where the convex hull is located in the electrical imaging well logging image obeys the uniform distribution.

[0048] Among them, the process of obtaining the empirical distribution function using the nearest distance algorithm, the calculation method of the uniform distribution function and the single-sample Kolmogorov-Smirnov test algorithm are all well-known technologies, and their specific principles and processes are not repeated here.

[0049] (4) Furthermore, the ratio between the density concentration and the characteristic distribution value of each characteristic convex hull is used as the density characteristic value of each characteristic convex hull to evaluate whether the convex hull has the distribution characteristics of dense layers in carbonate formations, such as dense layers with uniform arrangement of dots, in the geological characteristic area corresponding to the electrical imaging logging image.

[0050] According to the density eigenvalue of each characteristic convex hull, it can be understood that if the density concentration of the characteristic convex hull is greater and the characteristic distribution value is smaller, the density eigenvalue is larger, indicating that the convex hull distribution in the local image area where the convex hull is located in the electrical imaging logging image is more uniform, and the area corresponding to the convex hull in the electrical imaging logging image is more likely to be a dense layer in the carbonate formation; conversely, if the density concentration of the characteristic convex hull is smaller and the characteristic distribution value is larger, the density eigenvalue is smaller, indicating that the convex hull distribution in the local image area where the convex hull is located in the electrical imaging logging image is more uneven, and the area corresponding to the convex hull in the electrical imaging logging image is more likely to be point noise.

[0051] (5) Further, the density eigenvalues ​​of all feature convex hulls in each logging image are used as the input of the threshold segmentation algorithm, and the segmentation threshold is output, which is recorded as the eigenvalue threshold. The feature convex hull with a density eigenvalue greater than the eigenvalue threshold is used as the dense layer convex hull of each logging image; it is used to characterize the convex hull of the point-like geological features of the dense layer in the carbonate formation in the electrical imaging logging image.

[0052] (6) In this embodiment, in each well logging image, the pixel points in all convex hulls except the convex hull of the dense layer are used as the input of the median filtering algorithm, and the denoised well logging image is output; The median filtering algorithm is a well-known technology, and the specific process of using the median filtering algorithm to denoise the image will not be described in detail.

[0053] Furthermore, a background image of each electrical imaging logging image is obtained, and the background image is merged with the corresponding denoised logging image to obtain a denoised logging image, so as to suppress the point noise in the electrical imaging logging image, and avoid the point noise, which is a meaningless false information, from being mistaken for geological details in the image and enhanced in the subsequent image enhancement process; at the same time, avoid the point texture detail information belonging to the dense layer in the image from being mistaken for point noise and filtered out.

[0054] It should be noted that the process of merging the logging image, ie, the foreground image, with the background image is a well-known technology, and the specific principle and process will not be described in detail.

[0055] Preferably, the schematic diagram of the logging image denoising process provided in this embodiment is as follows: Figure 2 shown.

[0056] S3: All denoised logging images are enhanced using an image enhancement algorithm to obtain a deep learning model for classifying and identifying formations in LWD images.

[0057] When using electrical imaging logging instruments to collect imaging logging data, due to the current focusing effect, the current always flows toward the area with high conductivity, resulting in the formation of a white halo outside the area with high conductivity on the electrical imaging logging image. This is an illusion of no geological significance. For example, when collecting electrical imaging logging images of geological layers with pyrite, since pyrite is an excellent conductor of electricity, a bright white halo will be formed around the area where the pyrite is located in the collected electrical imaging logging images. However, this halo in the electrical imaging logging image that has no geological significance does not exist on the actual core or well wall. If this halo in the image is not suppressed, it will affect the accurate identification of different geological features in the formation by the formation classification and recognition model of the logging while drilling image constructed later.

[0058] Therefore, the image halo suppression technology is used to enhance the electrical imaging logging image to suppress the halo area in the image and enhance the geological detail information in the image. Specifically: In this embodiment, each denoised logging image is used as an input of a fast Retinex image enhancement algorithm, and an enhanced denoised logging image is output.

[0059] Among them, the fast Retinex image enhancement algorithm is a well-known technology, and the specific process will not be repeated here.

[0060] Furthermore, the Labelme open source tool is used to label and classify all enhanced denoised logging images according to the geological category of the carbonate formation to which they belong. The labeled denoised logging images are used as the input of the convolutional neural network, and the deep learning model is output. The cross entropy loss function and the stochastic gradient descent method are used as the loss function and optimization algorithm of the convolutional neural network, respectively. The training of the deep learning model is a well-known technology, and the specific training process will not be repeated here.

[0061] So far, this embodiment uses the median filtering algorithm to complete the denoising of the electrical imaging logging image based on the dense layer convex hull obtained by the point eigenvalues ​​of the dense layer, which can avoid the situation where the point noise in the electrical imaging logging image, which is interference information with no geological significance, is mistaken for the geological details in the image and enhanced in the subsequent image enhancement process. At the same time, it avoids the situation where the point geological details belonging to the dense layer in the carbonate formation in the electrical imaging logging image are mistaken for point noise and filtered out; further, by using the image halo suppression technology to perform image enhancement processing on the denoised electrical imaging logging image, the accuracy of the deep learning model obtained by neural network training in classifying and identifying the stratum to which the logging image belongs is improved.

[0062] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0064] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for classification and recognition of formations in logging while drilling images under complex drilling scenarios, characterized in that: The method comprises the following steps: S1: Collect electrical imaging logging images of different formations, and extract the foreground image of each electrical imaging logging image, which is recorded as the logging image; S2: According to the distribution characteristics of the pixels in the well logging image, the well logging image is denoised. Specifically: S201: using a connected domain extraction algorithm and a convex hull algorithm to extract all convex hulls in each well logging image, counting the number of all pixel points in each convex hull, and obtaining all feature convex hulls in each well logging image; S202: extracting the central pixel point of each feature convex hull, and determining the density concentration of each feature convex hull by analyzing the concentrated distribution of the central pixel points of all feature convex hulls; S203: constructing a window with the central pixel point of each characteristic convex hull as the center, determining the characteristic distribution value of each characteristic convex hull by analyzing the distribution of all pixel points in the window, and determining the density characteristic value of each characteristic convex hull in combination with the density concentration, so as to screen out the dense layer convex hull from all characteristic convex hulls of each logging image, and denoising each logging image to obtain a denoised logging image; S3: All denoised logging images are enhanced using an image enhancement algorithm to obtain a deep learning model for classifying and identifying formations in LWD images.

2. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The method for obtaining all feature convex hulls in each logging image is as follows: The number of all pixels in each convex hull is taken as the area of ​​each convex hull; The area of ​​all convex hulls in each logging image is used as the input of the threshold segmentation algorithm, and the segmentation threshold is output, which is recorded as the area threshold. All convex hulls with an area smaller than the area threshold are used as the feature convex hull of each logging image.

3. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The step of extracting the central pixel point of each feature convex hull comprises: Get the minimum circumscribed rectangle of each feature convex hull, and use the pixel point corresponding to the center of the minimum circumscribed rectangle as the center pixel point of each feature convex hull.

4. The method for classification and identification of formations in LWD images in complex drilling scenarios according to claim 1, characterized in that: The method for determining the density concentration of each feature convex hull is: The central pixel points of all feature convex hulls in each logging image are used as the input of the density clustering algorithm, and the local density of the central pixel points in all feature convex hulls is output as the density concentration of all feature convex hulls.

5. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The method for determining the characteristic distribution value of each characteristic convex hull is: The pixel values ​​of all pixels in the window of each feature convex hull are used as the input of the nearest distance algorithm, and the empirical distribution function of each feature convex hull is output; the uniform distribution function of each feature convex hull is constructed according to the pixel values ​​of all pixels in the window of each feature convex hull; The single sample test algorithm is used to calculate the statistic between the empirical distribution function and the uniform distribution function of each feature convex hull as the feature distribution value of each feature convex hull.

6. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The density characteristic value of each feature convex hull is the ratio between the density concentration and the characteristic distribution value of each feature convex hull.

7. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The process of selecting the dense layer convex hull from all the characteristic convex hulls of each logging image is as follows: The density eigenvalues ​​of all feature convex hulls in each logging image are used as the input of the threshold segmentation algorithm, and the segmentation threshold is output, which is recorded as the eigenvalue threshold. The feature convex hull with a density eigenvalue greater than the eigenvalue threshold is used as the dense layer convex hull of each logging image.

8. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The process of performing denoising on each well logging image to obtain a denoised well logging image is as follows: In each well logging image, all pixel points in the convex hull except the convex hull of the dense layer are used as the input of the filtering algorithm, and the denoised well logging image is output; A background image of each electrical imaging logging image is obtained, and the background image is merged with the corresponding denoised logging image to obtain a denoised logging image.

9. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The image enhancement algorithm is a fast Retinex image enhancement algorithm.

10. The method for classification and identification of formations in logging while drilling images in a complex drilling scenario according to claim 1, characterized in that: The method of obtaining a deep learning model for classifying and identifying formations in logging while drilling images includes: All enhanced denoised logging images are labeled and classified according to the geological categories of carbonate formations using open source power. The labeled denoised logging images are used as the input of the neural network to output the deep learning model.

Images