Automatic classification and recognition method of oil exploration core images based on deep learning

Abstract

The present invention relates to the field of core image analysis, and specifically to an automatic classification and recognition method for oil exploration core images based on deep learning. Obtain a grayscale connected domain; obtain the surface uniformity and granularity according to the grayscale change of the connected domain; obtain the edge significance according to the edge gradient of the connected domain; obtain the distribution similarity in combination with the overall grayscale difference between the connected domains, and then obtain the mineral feature vector; obtain the terrain change degree according to the gradient distribution around the pixel point; obtain the porosity according to the grayscale distribution; obtain the concave-convex feature vector in combination with the granularity; obtain the clustering cluster according to the vector difference and distance difference of the pixel point, and then obtain the key area feature vector; finally obtain the classification model of the core image according to the three vectors, and classify and identify the core image according to the classification model. The present invention identifies the microstructural manifestations of the core image, and the classification model can accurately classify and identify the structure in the image.

Description

Technical Field

[0001] The present invention relates to the field of core image analysis, and in particular to a method for automatic classification and recognition of oil exploration core images based on deep learning. Background Art

[0002] Cores are an important means of obtaining information about underlying rock formations in oil exploration. Core images can be used to observe and identify rock types, particle composition, pore structure and other information. This information can be used to assess oil and gas reserves, optimize drilling parameters and improve production efficiency, thereby making more accurate decisions during the drilling process and providing an important basis for oil exploration and development. Therefore, how to accurately classify and identify core images in the process of oil exploration has become an urgent problem to be solved.

[0003] In the core identification and classification scenarios in the field of oil exploration, the most common method is currently the classification and recognition technology based on deep learning, which can automatically learn and extract the features of core images from a large amount of data, and build a corresponding model to achieve core classification and recognition tasks. However, in actual applications, model training usually requires a large amount of labeled data, but the cost of obtaining core image data is very high, and there are many core structures with different categories but similar features in actual scenarios, as well as complex core structures with gradual or mixed features, which often lead to poor performance of the model and insufficient accuracy in the classification and recognition of various cores, affecting engineers' judgment and decision-making on geology. Summary of the invention

[0004] In order to solve the technical problem that there are actually many core structures with different categories but similar features, as well as complex core structures with gradient features or mixed features, which result in the inability of conventional classification models to accurately classify various core structures, the purpose of the present invention is to provide a method for automatic classification and recognition of oil exploration core images based on deep learning, and the technical scheme adopted is as follows: a method for automatic classification and recognition of oil exploration core images based on deep learning, the method comprising: obtaining a core image in the oil exploration process; obtaining a grayscale connected domain where each pixel is located according to the grayscale difference between pixels; selecting a connected domain where a pixel is located in the core image as a reference connected domain where a reference pixel is located; obtaining the surface uniformity of the reference connected domain according to the grayscale change characteristics of the reference connected domain; obtaining the edge significance of the reference connected domain according to the gradient distribution of the edge pixels of the reference connected domain; obtaining the distribution similarity of the reference connected domain according to the difference in surface uniformity, edge significance and overall grayscale between the reference connected domain and other grayscale connected domains; According to the distribution similarity, the mineral feature vector of the reference pixel is obtained; according to the gradient distribution of the pixels in the area around the reference pixel, the terrain change degree of the reference pixel is obtained; according to the grayscale change characteristics around the reference pixel, the granularity of the reference pixel is obtained; according to the proximity of the grayscale value of the pixels around the reference pixel to the minimum grayscale value of the core image, the porosity of the reference pixel is obtained; according to the terrain change degree, the granularity and the porosity, the concave-convex feature vector of the reference pixel is obtained; according to the distance difference between the pixels, the mineral feature vector difference and the concave-convex feature vector difference, the pixels of the core image are clustered to obtain all clustering clusters; according to the mineral feature vector and the concave-convex feature vector of the pixels in each cluster, all key areas of the core image are obtained; according to the area in each key area, the regional attributes and the distance between the key areas, the key area feature vector of the reference pixel is obtained; according to the mineral feature vector, the concave-convex feature vector and the key area feature vector of the pixel, a classification model of the core image is constructed; according to the classification model, the core image is automatically classified and identified.

[0005] Furthermore, the method for obtaining the grayscale connected domain where each pixel point is located includes: randomly selecting a pixel point as the first pixel point; taking all pixel points that are interconnected with the first pixel point and whose grayscale value difference is less than a preset first threshold as the connected domain pixel point set of the first pixel point; forming the grayscale connected domain where the first pixel point is located with the area where each pixel point in the connected domain pixel point set is located; traversing each pixel point in the core image to obtain the grayscale connected domain where each pixel point is located.

[0006] Furthermore, the method for obtaining the surface uniformity includes: obtaining the surface uniformity according to a surface uniformity calculation formula, and the surface uniformity calculation formula is as follows: In the formula, Indicates the surface uniformity of the reference connected domain; Represents the number of pixels in the reference connected domain; Represents the number of pixels in a preset first neighborhood of each pixel in the reference connected domain; Indicates the first The gray value of each pixel; Indicates the first The first neighborhood of the pixel The gray value of each pixel; Represents the standard deviation of the grayscale values ​​of pixels in the reference connected domain; Represents the grayscale change characteristics of the reference connected domain; represents the absolute value function.

[0007] Furthermore, the method for obtaining the edge saliency includes: calculating the gradient value of each edge pixel point in the reference connected domain, and taking the average of the gradient values ​​of all edge pixels points as the edge saliency of the reference connected domain.

[0008] Furthermore, the method for obtaining the distribution similarity includes: obtaining the distribution similarity according to a distribution similarity calculation formula, and the distribution similarity calculation formula is as follows: In the formula, Indicates the distribution similarity of the reference connected domain; Represents the number of grayscale connected domains other than the reference connected domain; Represents the reference connected domain and the The difference in surface uniformity between grayscale connected domains; Represents the reference connected domain and the The difference in edge significance between grayscale connected domains; Represents the reference connected domain and the The gray level mean difference between the gray level connected domains.

[0009] Furthermore, the method for obtaining the degree of terrain change includes: taking the reference pixel point as the starting point, establishing a Cartesian coordinate system, taking the positive direction of the y-axis as the first direction, and rotating the direction 45° clockwise as the second direction, and so on and so forth, until eight directions are selected; obtaining the degree of terrain change according to a terrain change degree calculation formula, and the terrain change degree calculation formula is as follows: In the formula, Indicates the degree of terrain change at the reference pixel; Indicates the number of directions starting from the reference pixel; Indicates the number of pixels selected in each direction starting from the reference pixel; Indicates the serial number of the reference pixel; Indicates The gradient direction angle difference between every two adjacent pixel points within a preset first number of pixel points starting from the reference pixel point in the direction; represents the minimum function; It represents the set of average changes of gradient direction angles in each direction from the first direction to the eighth direction.

[0010] Furthermore, the method for obtaining the porosity includes: obtaining the porosity according to a porosity calculation formula, and the porosity calculation formula is as follows: In the formula, Indicates the porosity of the reference pixel; Indicates the minimum grayscale value of the pixel in the core image; Indicates the number of pixels in the preset first neighborhood of the reference pixel; Indicates the reference pixel point preset second neighborhood The gray value of a pixel.

[0011] Furthermore, the method for acquiring the key area includes: acquiring the mineral feature vector according to a mineral feature vector calculation formula, and the mineral feature vector calculation formula is as follows: In the formula, The mineral feature vector representing the reference pixel; Indicates the surface uniformity of the reference pixel; Indicates the edge significance of the reference pixel; Indicates the distribution similarity of the reference connected domain; the concave-convex feature vector is obtained according to the concave-convex feature vector calculation formula, and the concave-convex feature vector calculation formula is as follows: In the formula, Represents the concave-convex feature vector of the reference pixel; Indicates the degree of terrain change at the reference pixel; Indicates the granularity of the reference pixel, and presets the grayscale change characteristics within the third neighborhood of the reference pixel; Indicates the porosity of the reference pixel point; according to the distance difference between any two pixels, the mineral feature vector difference and the concave-convex feature vector difference, the distance measurement value between any two reference pixels is obtained, and the calculation formula is as follows: In the formula, Represents pixel With pixels The distance measure between them; Represents pixel With pixels The Euclidean distance between Represents pixel With pixels The cosine similarity between the mineral feature vectors; Represents pixel With pixels The cosine similarity between the concave and convex feature vectors; clustering the core image according to the distance measurement value between the pixel points to obtain all cluster clusters in the core image; taking the cluster clusters whose mineral feature vector mean of the pixel points is greater than the preset second threshold as the mineral area, and taking the cluster cluster whose concave and convex feature vector mean of the pixel points is greater than the preset second threshold as the concave and convex area; and taking both the mineral area and the concave and convex area as the key area.

[0012] Furthermore, the method for obtaining the key area feature vector includes: if the reference pixel is in the key area, then according to the distance between the key area to which the reference pixel belongs and the closest key area, obtaining the density of the key area to which the reference pixel belongs, the calculation formula is as follows: In the formula, Indicates the density of the key area to which the reference pixel belongs; Indicates a preset second number of key areas that are closest to the key area to which the reference pixel belongs; Indicates the nearest The Euclidean distance between the centroid pixel point of the key area and the centroid pixel point of the key area to which the reference pixel belongs; the key area feature vector is obtained according to the key area feature vector calculation formula, and the key area feature vector calculation formula is as follows: In the formula, The key area feature vector representing the reference pixel; Indicates the region attribute to which the reference pixel belongs. When the reference pixel is in the key region, the value is 1, otherwise it is 0. Indicates the area of ​​the region to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0; Indicates the density of the area to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0.

[0013] Furthermore, the method for obtaining the classification model of the core image includes: using the mineral feature vector, the concave-convex feature vector and the key area feature vector of each pixel point to construct a feature matrix for each pixel point; splicing the feature matrix of each pixel point in the core image according to the position of the pixel point to obtain the feature matrix of the core image; standardizing the feature matrix of the core image to obtain a standard feature matrix of the core image; inputting the standard feature matrix into a deep learning model for training to obtain a classification model of the core image.

[0014] The present invention has the following beneficial effects: the present invention obtains a core image; since the core may contain various types of minerals, the mineral feature vectors of the pixel points are analyzed; since minerals often have more similar distributions on the core surface rather than appearing alone, and the brightness of the minerals is usually more uniform than the core surface, and has a more obvious edge, the grayscale connected domain where each pixel point is located is first obtained, and the surface uniformity, edge significance and distribution similarity of the grayscale connected domain are analyzed, and then the mineral feature vectors of the reference pixel points are obtained according to the surface uniformity, edge significance and distribution similarity; since the core surface may be uneven, it will affect the illumination characteristics of the core surface, and can reflect the particle characteristics of the core surface to a certain extent, and special holes may appear at some concave and convex positions, so the concave and convex feature vectors of the core image are obtained by analyzing the terrain change degree, particle degree and porosity degree of the core image; since both the mineral feature and the concave and convex feature have regional concentration, that is, the local pixels of the mineral area and the concave and convex area are similar in performance in these two features. Since the characteristics of mineral regions and concave-convex regions are extremely important for core classification and identification, the embodiments of the present invention segment the mineral regions and concave-convex regions in the image and construct a corresponding feature matrix to represent the properties of the core image; and construct a core image classification model based on the standard feature matrix of the core image; and complete the automatic classification and identification of the core image of oil exploration based on the core image classification model. The present invention can identify the micro-structural manifestations of the core image, so that the trained classification model can accurately classify and identify the structure in the core image. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1A flow chart of a method for automatic classification and recognition of oil exploration core images based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a method for automatic classification and recognition of oil exploration core images based on deep learning proposed by the present invention in combination with the accompanying drawings and preferred embodiments. 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.

[0018] 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 invention belongs.

[0019] The following is a detailed description of a method for automatic classification and recognition of oil exploration core images based on deep learning provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows a method for automatic classification and recognition of oil exploration core images based on deep learning provided by an embodiment of the present invention. The method includes: Step S1: Acquire core images during oil exploration.

[0021] The embodiment of the present invention is mainly applied to the core identification and classification scenario in the field of oil exploration, so firstly, images of various core structures are obtained, that is, core images in the process of oil exploration are obtained. In one embodiment of the present invention, core drilling equipment is used to extract core samples from underground strata in the area to be detected, and the samples are subjected to appropriate surface treatment, such as clearing, drying and coating, and then a high-resolution camera is used to capture the surface image of the core sample, and the image is grayed to obtain the required core image.

[0022] Step S2: According to the grayscale difference between pixel points, the grayscale connected domain where each pixel point is located is obtained; the connected domain where any pixel point in the core image is located is selected as the reference connected domain where the reference pixel point is located; the surface uniformity of the reference connected domain is obtained according to the grayscale change characteristics of the reference connected domain; the edge significance of the reference connected domain is obtained according to the gradient distribution of the edge pixels of the reference connected domain; the distribution similarity of the reference connected domain is obtained according to the surface uniformity difference, edge significance difference and overall grayscale difference between the reference connected domain and other grayscale connected domains; the mineral feature vector of the reference pixel point is obtained according to the surface uniformity, the edge significance and the distribution similarity.

[0023] In the field of oil exploration, the recognition and classification of core images can provide detailed information about geology and is often used in early decision-making for oil drilling and mining. Since conventional deep learning models perform poorly when processing similar and complex core images, we need to accurately extract features from core images and make the extracted geological features better meet the needs of actual scenarios. In the oil exploration scenario, the type and content of minerals in the core affect the mechanical properties of the rock and the fluid flow characteristics, so in an embodiment of the present invention, the mineral feature vectors of the pixel points of the core image are analyzed.

[0024] In actual situations, the core may contain various types of minerals, depending on the genesis of the rock, the geological environment, and later geological processes. The brightness of these minerals is usually more uniform than that of the core surface, and has more obvious edges. In addition, minerals tend to have more similar distributions on the core surface, rather than appearing alone, so the embodiment of the present invention first obtains the grayscale connected domain where each pixel is located, and analyzes the surface uniformity, edge significance, and distribution similarity of the grayscale connected domain, and obtains the mineral feature vector of the reference pixel based on the surface uniformity, edge significance, and distribution similarity.

[0025] Preferably, in one embodiment of the present invention, the method for obtaining the grayscale connected domain where each pixel point is located includes: randomly selecting a pixel point as a first pixel point; taking all pixels that are interconnected with the first pixel point and whose grayscale value difference is less than a preset first threshold as a connected domain pixel point set of the first pixel point, and in one embodiment of the present invention, the preset first threshold value is set to 10. It should be noted that the preset first threshold value can be set voluntarily and is not limited here.

[0026] The area where each pixel point in the connected domain pixel point set is located is formed into a grayscale connected domain where the first pixel point is located; each pixel point in the core image is traversed to obtain the grayscale connected domain where each pixel point is located.

[0027] Analyze whether the surface of the grayscale connected domain is uniform. Preferably, in one embodiment of the present invention, the method for obtaining the surface uniformity includes: obtaining the surface uniformity according to a surface uniformity calculation formula, and the surface uniformity calculation formula is as follows: In the formula, Indicates the surface uniformity of the reference connected domain; Represents the number of pixels in the reference connected domain; Represents the number of pixels in a preset first neighborhood of each pixel in the reference connected domain; Indicates the first The gray value of each pixel; Indicates the first The first neighborhood of the pixel The gray value of each pixel; Represents the standard deviation of the grayscale values ​​of pixels in the reference connected domain; Represents the grayscale change characteristics of the reference connected domain; represents the absolute value function.

[0028] In the surface uniformity calculation formula, each pixel in the reference connected domain is analyzed. The smaller the grayscale difference between the pixel and each pixel in the preset first neighborhood, the smaller the grayscale difference between the pixel and each pixel in the preset first neighborhood. The more similar the grayscale of the pixel is to the surrounding pixels, the The more uniform the grayscale distribution of the pixel points in the preset first neighborhood is, the more uniform the grayscale distribution of each pixel point in the reference connected domain is calculated; , The smaller it is, the more uniform the grayscale distribution in the reference connected domain is. At this time, the surface uniformity of the reference connected domain is greater; the grayscale discrete value in the reference connected domain The smaller it is, the more uniform the grayscale value of the reference connected domain is. At this time, the surface uniformity of the reference connected domain is greater.

[0029] In one embodiment of the present invention, the preset first neighborhood is set to be centered on each pixel point. It should be noted that the preset first neighborhood can be set by oneself and is not limited here.

[0030] Analyze whether the edge of the grayscale connected domain is significant. Preferably, in one embodiment of the present invention, the method for obtaining the edge significance includes: calculating the gradient value of each edge pixel point in the reference connected domain, and taking the average of the gradient values ​​of all edge pixels as the edge significance of the reference connected domain. The edge significance calculation formula is as follows: In the formula, Indicates the edge significance of the reference connected domain; Represents the number of edge pixels of the reference connected domain; The first The gradient value of the edge pixel.

[0031] In the edge saliency calculation formula, the larger the gradient mean of the edge pixel point is, the more obvious the edge of the edge pixel point is, that is, the higher the edge saliency of the reference connected domain is.

[0032] According to the surface feature differences between the reference connected domain and other grayscale connected domains, the surface complexity of the reference connected domain is analyzed. Preferably, in one embodiment of the present invention, the method for obtaining the distribution similarity includes: obtaining the distribution similarity according to a distribution similarity calculation formula, and the distribution similarity calculation formula is as follows: In the formula, Indicates the distribution similarity of the reference connected domain; Represents the number of grayscale connected domains other than the reference connected domain; Represents the reference connected domain and the The difference in surface uniformity between grayscale connected domains; Represents the reference connected domain and the The difference in edge significance between grayscale connected domains.

[0033] In the distribution similarity calculation formula, the smaller the surface uniformity difference, grayscale mean difference and surface significance difference between the reference connected domain and each grayscale connected domain, the more grayscale connected domains are similar to the reference connected domain, and the higher the distribution similarity of the reference connected domain.

[0034] According to the surface uniformity, the edge prominence and the distribution similarity, the mineral feature vector of the reference pixel is obtained. Preferably, in one embodiment of the present invention, the mineral feature vector is obtained according to a mineral feature vector calculation formula, and the mineral feature vector calculation formula is as follows: In the formula, The mineral feature vector representing the reference pixel; Indicates the surface uniformity of the reference pixel; Indicates the edge significance of the reference pixel; Indicates the distribution similarity of the reference connected domain.

[0035] Step S3: Cluster the pixels of the core image according to the distance difference between the pixels, the mineral feature vector difference and the concave-convex feature vector difference to obtain all clusters; obtain all key areas of the core image according to the mineral feature vector and the concave-convex feature vector of the pixels in each cluster; obtain the key area feature vector of the reference pixel according to the area in each key area, the area attribute and the distance between the key areas; construct a classification model of the core image according to the mineral feature vector, the concave-convex feature vector and the key area feature vector of the pixel; and automatically classify and identify the core image according to the classification model.

[0036] The geological environment where the core is located may have complex structural deformation or weathering phenomena, resulting in unevenness on the core surface. These uneven parts will affect the illumination characteristics of the core surface and can reflect the particle characteristics of the core surface to a certain extent. Special holes may also appear in some of the uneven positions. Since the uneven characteristics of the core surface reflect the porosity, permeability and stability of the rock formation, in the embodiment of the present invention, the uneven characteristics of the core image are obtained by analyzing the degree of terrain change, particle degree and porosity of the core image.

[0037] Preferably, in one embodiment of the present invention, the method for obtaining the degree of terrain change includes: taking the reference pixel point as the starting point, establishing a Cartesian coordinate system, taking the positive direction of the y-axis as the first direction, the direction rotated 45° clockwise as the second direction, and taking the positive direction of the x-axis as the third direction, until eight directions are selected.

[0038] The terrain change degree is obtained according to the terrain change degree calculation formula, and the terrain change degree calculation formula is as follows: In the formula, Indicates the degree of terrain change at the reference pixel; Indicates the number of directions starting from the reference pixel; Indicates the number of pixels selected in each direction starting from the reference pixel; Indicates the serial number of the reference pixel; Indicates The gradient direction angle difference between every two adjacent pixel points within a preset first number of pixel points starting from the reference pixel point in the direction; represents the minimum function; It represents the set of average changes of gradient direction angles in each direction from the first direction to the eighth direction.

[0039] In the terrain change degree calculation formula, each direction is analyzed. The smaller the difference in gradient direction angle between each adjacent two pixels between the reference pixel and the preset first number of pixels, the smaller the average change in gradient direction angle. The smaller it is, the more consistent the gradient change direction in each direction is. The minimum value of the average change of the gradient direction angle in the eight directions is selected. The smaller the minimum value is, the greater the possibility of illumination change around the reference pixel point, and the smaller the degree of terrain change of the reference pixel point is.

[0040] In one embodiment of the present invention, the preset first number is set to 30. It should be noted that the preset first number can be set arbitrarily and is not limited here.

[0041] In one embodiment of the present invention, the granularity can be obtained by the grayscale change characteristic formula in the above steps. The specific calculation formula is as follows: In the formula, Indicates the granularity of the reference pixel; Indicates the number of pixels within the preset second neighborhood of the reference pixel; Indicates the number of pixels in the preset first neighborhood of the reference pixel; Indicates the reference pixel point preset second neighborhood The gray value of each pixel; The reference pixel point is preset in the second neighborhood. The pixel point is in the preset first neighborhood. Grayscale values ​​of other pixels; Indicates the standard deviation of the grayscale value in the preset second neighborhood of the reference pixel.

[0042] In the granularity calculation formula, the greater the grayscale difference between each pixel in the preset second neighborhood and other pixel points in the preset first neighborhood, the higher the prominence of the pixel in the preset second neighborhood, and the higher the granularity around the reference pixel.

[0043] Preferably, in one embodiment of the present invention, the method for obtaining the porosity comprises: obtaining the porosity according to a porosity calculation formula, and the porosity calculation formula is as follows: In the formula, Indicates the porosity of the reference pixel; Indicates the minimum grayscale value of the pixel in the core image; Indicates the number of pixels in the preset first neighborhood of the reference pixel; Indicates the reference pixel point preset second neighborhood The gray value of a pixel.

[0044] In the porosity calculation formula, the pores on the core surface do not reflect light, and the pore area often shows an extremely low grayscale value. The closer the grayscale mean value in the first neighborhood of the reference pixel point is to the grayscale minimum value of the core image, the higher the grayscale value of the core image will be. The larger it is, the greater the porosity of the reference pixel.

[0045] The concave-convex feature vector is obtained according to the concave-convex feature vector calculation formula, and the concave-convex feature vector calculation formula is as follows: In the formula, Represents the concave-convex feature vector of the reference pixel; Indicates the degree of terrain change at the reference pixel; Indicates the granularity of the reference pixel, and presets the grayscale change characteristics within the third neighborhood of the reference pixel; Indicates the porosity of the reference pixel.

[0046] In one embodiment of the present invention, the preset third neighborhood is set to be centered on the reference pixel point. It should be noted that the preset third neighborhood can be set by oneself and is not limited here.

[0047] Step S4: cluster the pixels of the core image according to the distance difference between the pixels, the mineral feature vector difference and the concave-convex feature vector difference to obtain all clustering clusters; obtain all key areas of the core image according to the mineral feature vector and the concave-convex feature vector of the pixel points in each clustering cluster; obtain the key area feature vector of the reference pixel point according to the area in each key area, the area attribute and the distance between the key areas; construct a classification model of the core image according to the mineral feature vector, the concave-convex feature vector and the key area feature vector of the pixel point; and automatically classify and identify the core image according to the classification model.

[0048] The above steps can obtain the mineral features and concave-convex features of all pixel points of the core image, and both the mineral features and the concave-convex features have regional concentration, that is, the local pixels of the mineral region and the concave-convex region have similar performances in these two features. Since the characteristics of the mineral region and the concave-convex region are extremely important for core classification and identification, the mineral region and the concave-convex region in the image are segmented in the embodiment of the present invention and the corresponding feature matrix is ​​constructed to represent the properties of the core image.

[0049] Preferably, in one embodiment of the present invention, the method for obtaining the key area includes: according to the above steps, the mineral feature vector of the pixel point in the core image can be obtained. With the concave and convex feature vector .

[0050] According to the distance difference between any two pixels, the mineral feature vector difference and the concave-convex feature vector difference, the distance measurement value between any two pixels is obtained. The calculation formula is as follows: In the formula, Represents pixel With pixels The distance measure between them; Represents pixel With pixels The Euclidean distance between Represents pixel With pixels The cosine similarity between the mineral feature vectors; Represents pixel With pixels The cosine similarity between the concave and convex feature vectors.

[0051] In the distance measurement value calculation formula, the smaller the distance between pixels, the smaller the distance measurement value between pixels; the larger the cosine similarity of the mineral feature vectors and the cosine similarity of the concave-convex feature vectors between pixels, the more similar the surface features between pixels are. At this time, the two pixels are considered to be of the same type, and the distance measurement value between pixels is smaller.

[0052] The core image is clustered according to the distance measurement value between the pixels to obtain all cluster clusters in the core image. In the embodiment of the present invention, the DBSCAN clustering algorithm is used to perform density clustering on the core image, wherein the DBSCAN clustering algorithm is well known to technicians in this field and is not limited or elaborated here.

[0053] The clusters whose mineral feature vector mean of the pixel points is greater than the preset second threshold are taken as mineral regions, and the clusters whose concave-convex feature vector mean of the pixel points is greater than the preset second threshold are taken as concave-convex regions. In one embodiment of the present invention, the mean of the mineral feature vector and the mean of the concave-convex feature vector are normalized, and the preset second threshold is set to 0.7. It should be noted that the preset second threshold can be set voluntarily and is not limited here.

[0054] Both the mineral area and the concave-convex area are regarded as key areas.

[0055] The key area feature vector is constructed using the features of the key area itself. Preferably, in one embodiment of the present invention, the method for obtaining the key area feature vector includes: if the reference pixel point is in the key area, then according to the distance between the key area to which the reference pixel point belongs and the closest key area, the density of the key area to which the reference pixel point belongs is obtained, and the calculation formula is as follows: In the formula, Indicates the density of the key area to which the reference pixel belongs; Indicates a preset second number of key areas that are closest to the key area to which the reference pixel belongs; Indicates the nearest The Euclidean distance between the centroid pixel of a key area and the centroid pixel of the key area to which the reference pixel belongs.

[0056] In the density calculation formula, the closer the distance between the key area to which the reference pixel belongs and the closest key area is, the greater the density of the key area to which the reference pixel belongs is.

[0057] The key area feature vector is obtained according to the key area feature vector calculation formula, and the key area feature vector calculation formula is as follows: In the formula, The key area feature vector representing the reference pixel; Indicates the region attribute to which the reference pixel belongs. When the reference pixel is in the key region, the value is 1, otherwise it is 0. Indicates the area of ​​the region to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0; Indicates the density of the area to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0.

[0058] Preferably, in one embodiment of the present invention, the method for obtaining the classification model of the core image includes: using the mineral feature vector, the concave-convex feature vector and the key area feature vector of each pixel point to construct a feature matrix of each pixel point; splicing the feature matrix of each pixel point in the core image according to the position of the pixel point to obtain the feature matrix of the core image. The calculation method of the feature matrix is ​​a technical means well known to those skilled in the art, and is not limited or elaborated here.

[0059] The feature matrix of the core image is standardized to obtain a standard feature matrix of the core image. It should be noted that the standardization process is a technical means well known to those skilled in the art and will not be described in detail here. The standard feature matrix is ​​input into the deep learning model for training to obtain a classification model of the core image. It should be noted that the classification model acquisition method is a technical means well known to those skilled in the art and will not be described in detail here.

[0060] After the trained core image classification model is obtained, the image to be classified is input into the model for automatic classification and identification to obtain accurate core type judgment results. Engineers can conduct precise geology analysis based on the results and make more accurate decisions during oil drilling and mining.

[0061] At this point, the automatic classification and recognition of oil exploration core images is completed.

[0062] In summary, the core image in the process of oil exploration is obtained; the grayscale connected domain where each pixel is located is obtained according to the grayscale difference between the pixels; the connected domain where a pixel is located in the core image is selected as the reference connected domain where the reference pixel is located; the surface uniformity of the reference connected domain is obtained according to the grayscale change characteristics of the reference connected domain; the edge significance of the reference connected domain is obtained according to the gradient distribution of the edge pixels of the reference connected domain; the distribution similarity of the reference connected domain is obtained according to the surface uniformity difference, edge significance difference and overall grayscale difference between the reference connected domain and other grayscale connected domains; the mineral feature vector of the reference pixel is obtained according to the surface uniformity, the edge significance and the distribution similarity; the terrain change degree of the reference pixel is obtained according to the gradient distribution of the pixels in the area around the reference pixel; the reference pixel is obtained according to the grayscale change characteristics around the reference pixel. The particle degree of the pixel point is considered; the porosity of the reference pixel point is obtained according to the proximity of the gray value of the pixel points around the reference pixel point to the minimum gray value of the core image; the concave-convex feature vector of the reference pixel point is obtained according to the terrain change degree, the particle degree and the porosity; the pixel points of the core image are clustered according to the distance difference between the pixel points, the mineral feature vector difference and the concave-convex feature vector difference to obtain all clustering clusters; all key areas of the core image are obtained according to the mineral feature vector and the concave-convex feature vector of the pixel points in each clustering cluster; the key area feature vector of the reference pixel point is obtained according to the area in each key area, the area attribute and the distance between the key areas; a classification model of the core image is constructed according to the mineral feature vector, the concave-convex feature vector and the key area feature vector of the pixel point; the core image is automatically classified and identified according to the classification model.

[0063] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. 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.

[0064] 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.

Claims

1. A method for automatic classification and recognition of oil exploration core images based on deep learning, characterized in that: The method comprises: obtaining a core image in the process of oil exploration; obtaining a grayscale connected domain where each pixel point is located according to the grayscale difference between pixel points; selecting a connected domain where a pixel point is located in the core image as a reference connected domain where a reference pixel point is located; obtaining the surface uniformity of the reference connected domain according to the grayscale change characteristics of the reference connected domain; obtaining the edge significance of the reference connected domain according to the gradient distribution of edge pixel points of the reference connected domain; obtaining the distribution similarity of the reference connected domain according to the surface uniformity difference, edge significance difference and overall grayscale difference between the reference connected domain and other grayscale connected domains; obtaining the mineral feature vector of the reference pixel point according to the surface uniformity, edge significance and distribution similarity; obtaining the terrain change degree of the reference pixel point according to the gradient distribution of pixel points in the area around the reference pixel point; obtaining the granularity of the reference pixel point according to the grayscale value of the pixel points around the reference pixel point; obtaining the granularity of the reference pixel point according to the grayscale value of the pixel points around the reference pixel point According to the degree of proximity to the minimum grayscale value of the core image, the porosity of the reference pixel is obtained; according to the degree of terrain change, the particle degree and the porosity, the concave-convex feature vector of the reference pixel is obtained; according to the distance difference between the pixels, the mineral feature vector difference and the concave-convex feature vector difference, the pixels of the core image are clustered to obtain all cluster clusters; according to the mineral feature vector and the concave-convex feature vector of the pixel in each cluster, all key areas of the core image are obtained; according to the area, regional attributes and distance between key areas in each key area, the key area feature vector of the reference pixel is obtained; according to the mineral feature vector, the concave-convex feature vector and the key area feature vector of the pixel, a classification model of the core image is constructed; according to the classification model, the core image is automatically classified and identified; the method for obtaining the key area includes: obtaining the mineral feature vector according to the mineral feature vector calculation formula, and the mineral feature vector calculation formula is as follows: In the formula, The mineral feature vector representing the reference pixel; Indicates the surface uniformity of the reference pixel; Indicates the edge significance of the reference pixel; Indicates the distribution similarity of the reference connected domain; the concave-convex feature vector is obtained according to the concave-convex feature vector calculation formula, and the concave-convex feature vector calculation formula is as follows: In the formula, Represents the concave-convex feature vector of the reference pixel; Indicates the degree of terrain change at the reference pixel; Indicates the granularity of the reference pixel, and presets the grayscale change characteristics within the third neighborhood of the reference pixel; Indicates the porosity of the reference pixel point; according to the distance difference between any two pixels, the mineral feature vector difference and the concave-convex feature vector difference, the distance measurement value between any two reference pixels is obtained, and the calculation formula is as follows: In the formula, Represents pixel With pixels The distance measure between them; Represents pixel With pixels The Euclidean distance between Represents pixel With pixels The cosine similarity between the mineral feature vectors; Represents pixel With pixels The cosine similarity between the concave and convex feature vectors; clustering the core image according to the distance measurement value between the pixel points to obtain all cluster clusters in the core image; taking the cluster clusters whose mineral feature vector mean of the pixel points is greater than the preset second threshold as the mineral area, and taking the cluster cluster whose concave and convex feature vector mean of the pixel points is greater than the preset second threshold as the concave and convex area; and taking both the mineral area and the concave and convex area as the key area.

2. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1 is characterized in that: The method for obtaining the grayscale connected domain where each pixel point is located includes: randomly selecting a pixel point as a first pixel point; taking all pixel points that are interconnected with the first pixel point and whose grayscale value difference is less than a preset first threshold as a connected domain pixel point set of the first pixel point; forming the grayscale connected domain where the first pixel point is located with the area where each pixel point in the connected domain pixel point set is located; traversing each pixel point in the core image to obtain the grayscale connected domain where each pixel point is located.

3. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1 is characterized in that: The method for obtaining the surface uniformity includes: obtaining the surface uniformity according to a surface uniformity calculation formula, and the surface uniformity calculation formula is as follows: In the formula, Indicates the surface uniformity of the reference connected domain; Represents the number of pixels in the reference connected domain; Represents the number of pixels in a preset first neighborhood of each pixel in the reference connected domain; Indicates the first The gray value of each pixel; Indicates the first The first neighborhood of the pixel The gray value of each pixel; Represents the standard deviation of the grayscale values ​​of pixels in the reference connected domain; Represents the grayscale change characteristics of the reference connected domain; represents the absolute value function.

4. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1 is characterized in that: The method for obtaining the edge saliency comprises: calculating the gradient value of each edge pixel point in the reference connected domain, and taking the average of the gradient values ​​of all edge pixels points as the edge saliency of the reference connected domain.

5. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1, characterized in that: The method for obtaining the distribution similarity includes: obtaining the distribution similarity according to a distribution similarity calculation formula, and the distribution similarity calculation formula is as follows: In the formula, Indicates the distribution similarity of the reference connected domain; Represents the number of grayscale connected domains other than the reference connected domain; Represents the reference connected domain and the The difference in surface uniformity between grayscale connected domains; Represents the reference connected domain and the The difference in edge significance between grayscale connected domains; Represents the reference connected domain and the The gray level mean difference between the gray level connected domains.

6. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1, characterized in that: The method for obtaining the degree of terrain change includes: taking the reference pixel point as the starting point, establishing a Cartesian coordinate system, taking the positive direction of the y-axis as the first direction, and rotating the direction 45° clockwise as the second direction, and continuing to rotate by analogy until eight directions are selected; obtaining the degree of terrain change according to a terrain change degree calculation formula, and the terrain change degree calculation formula is as follows: In the formula, Indicates the degree of terrain change at the reference pixel; Indicates the number of directions starting from the reference pixel; Indicates the number of pixels selected in each direction starting from the reference pixel; Indicates the serial number of the reference pixel; Indicates The gradient direction angle difference between every two adjacent pixel points within a preset first number of pixel points starting from the reference pixel point in the direction; represents the minimum function; represents an exponential function with a natural constant as base; It represents the set of average changes of gradient direction angles in each direction from the first direction to the eighth direction.

7. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1, characterized in that: The method for obtaining the porosity includes: obtaining the porosity according to a porosity calculation formula, and the porosity calculation formula is as follows: In the formula, Indicates the porosity of the reference pixel; Indicates the minimum grayscale value of the pixel in the core image; Indicates the number of pixels in the preset first neighborhood of the reference pixel; Indicates the reference pixel point preset second neighborhood The gray value of a pixel.

8. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1, characterized in that: The method for obtaining the key area feature vector includes: if the reference pixel point is in the key area, then according to the distance between the key area to which the reference pixel point belongs and the closest key area, the density of the key area to which the reference pixel point belongs is obtained, and the calculation formula is as follows: In the formula, Indicates the density of the key area to which the reference pixel belongs; Indicates a preset second number of key areas that are closest to the key area to which the reference pixel belongs; Indicates the nearest The Euclidean distance between the centroid pixel point of the key area and the centroid pixel point of the key area to which the reference pixel belongs; the key area feature vector is obtained according to the key area feature vector calculation formula, and the key area feature vector calculation formula is as follows: In the formula, The key area feature vector representing the reference pixel; Indicates the region attribute to which the reference pixel belongs. When the reference pixel is in the key region, the value is 1, otherwise it is 0. Indicates the area of ​​the region to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0; Indicates the density of the area to which the reference pixel belongs. When the reference pixel is not in the key area, the value is 0.

9. The method for automatic classification and recognition of oil exploration core images based on deep learning according to claim 1, characterized in that: The method for obtaining the classification model of the core image includes: using the mineral feature vector, the concave-convex feature vector and the key area feature vector of each pixel point to construct a feature matrix of each pixel point; splicing the feature matrix of each pixel point in the core image according to the position of the pixel point to obtain the feature matrix of the core image; standardizing the feature matrix of the core image to obtain a standard feature matrix of the core image; inputting the standard feature matrix into a deep learning model for training to obtain a classification model of the core image.

Images