A method for identifying three-dimensional rock minerals by combining CT and electron microscopy

Abstract

The application discloses a CT and electron microscope combined three-dimensional rock mineral identification method, and relates to the technical field of rock physics and digital core. The method corrects and calibrates CT data by using high-precision mineral identification results of SEM, and establishes a mapping relationship between CT gray values and SEM gray values. Three-dimensional SEM gray value data is obtained from three-dimensional CT gray values through the mapping relationship, and high-precision reconstruction of three-dimensional mineral distribution in a rock sample is realized. The scheme combines the three-dimensional space advantage of CT and the high-precision identification advantage of SEM, makes up for the insufficient three-dimensional mineral identification precision of CT, and fills the gap that the two-dimensional representation of SEM cannot restore three-dimensional space distribution.

Description

Technical Field

[0001] This invention relates to the fields of rock physics and digital core technology, specifically a three-dimensional rock and mineral identification method combining CT and electron microscopy. Background Technology

[0002] In the fields of rock physics and digital core technology, high-precision three-dimensional mineral characterization is a crucial foundation for digital core analysis. Inaccurate mineral distribution data directly impacts the accuracy of analytical results such as rock physics parameter calculations, pore structure analysis, multiphase flow simulation, rock mechanics analysis, and reservoir evaluation. Therefore, obtaining accurate and reliable three-dimensional mineral distribution information is of paramount importance. Currently, commonly used mineral characterization methods mainly include X-ray computed tomography (CT), scanning electron microscopy (SEM), and mineral prediction using large deep learning models on CT grayscale images. While CT technology can acquire three-dimensional information, it is affected by factors such as mineral grayscale overlap, beam hardening effect, and X-ray scattering effect. This causes low-energy photons to be preferentially absorbed when photons of different energies emitted from the X-ray source penetrate the sample, resulting in a gradual increase in the average energy of the remaining rays. Consequently, the same mineral exhibits different grayscale responses inside and at the edge of the sample, producing cup-shaped artifacts, making high-precision mineral identification difficult. SEM technology can accurately identify mineral composition, but it can only acquire information from the surface or two-dimensional cross-section of the rock sample, making it difficult to obtain the three-dimensional mineral distribution inside the sample. Furthermore, mineral prediction based on large deep learning models is essentially predictive inference, lacking three-dimensional mineral calibration from actual rock cores, making it difficult to obtain reliable data. Therefore, how to accurately identify the three-dimensional mineral distribution inside rock samples is a pressing technical problem to be solved in the field of digital core research. Summary of the Invention

[0003] To address at least one of the aforementioned problems, this invention proposes a three-dimensional rock and mineral identification method combining CT and electron microscopy.

[0004] The technical solution of this invention is: a three-dimensional rock and mineral identification method combining CT and electron microscopy, comprising the following steps: S1. Core samples were taken and subjected to multi-level CT scans. The scan images were preprocessed to obtain CT grayscale images under multi-level conditions. S2. Perform SEM scanning on the core and mineral analysis simultaneously to obtain a SEM grayscale image containing mineral information. S3. Register the SEM grayscale image and the CT grayscale image to obtain an equivalent SEM grayscale image, and use a correction function to correct the registration result; S4. Based on the error between the measured grayscale value and the corrected grayscale value of the CT grayscale image under different energy level conditions, determine the weight of the CT grayscale image under different energy level conditions; combine the weights and perform weighted fusion of the CT grayscale images under different energy level conditions to obtain the fused CT grayscale image. S5. Use the registration results of SEM grayscale image and CT grayscale image to construct a training sample set to train XGBoost. After training, substitute the grayscale value of the fused CT grayscale image into the trained XGBoost model to obtain an equivalent fused SEM grayscale image. Combine the SEM grayscale image containing mineral information in S2 to obtain the mineral composition inside the rock sample.

[0005] The beneficial effects of this invention are as follows: This invention utilizes the high-precision mineral identification results of SEM to correct and calibrate CT data, establishing a mapping relationship between CT grayscale values ​​and SEM grayscale values. Three-dimensional SEM grayscale data is obtained from the three-dimensional CT grayscale values ​​through this mapping relationship, achieving high-precision reconstruction of the three-dimensional mineral distribution within the rock sample. This scheme integrates the three-dimensional spatial advantages of CT with the high-precision identification advantages of SEM, both compensating for the insufficient accuracy of CT's three-dimensional mineral identification and filling the gap in SEM's two-dimensional representation's inability to reconstruct three-dimensional spatial distribution. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the spatial registration of the rock sample surface in this embodiment.

[0007] Figure 2 This is a schematic diagram of the area-weighted average method in this embodiment.

[0008] Figure 3 This embodiment verifies the change in the average relative error of the CT grayscale correction model.

[0009] Figure 4 This is the three-dimensional mineral distribution model in this embodiment. Detailed Implementation

[0010] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0011] A three-dimensional rock and mineral identification method combining CT and electron microscopy includes the following steps: S1. Core samples were taken and subjected to multi-level CT scans. The scan images were preprocessed to obtain CT grayscale images under multi-level conditions.

[0012] In this step, the core sample can be either a conventional cylindrical core or a cubic core. Similar to conventional CT scans, the core sample in this step needs to have relatively flat surfaces to reduce geometric artifacts. Additionally, since the core is extracted from the reservoir, it contains a significant amount of non-core impurities, such as oil and water-soluble metal ions. Therefore, it needs to be cleaned before use. The specific cleaning procedure is a standard procedure in this field and will not be described in detail here.

[0013] Different energies of X-rays have varying penetrating power, beam hardening degree, and scattering response. Therefore, in this step, a multi-energy-level CT scanner is used for scanning. In some specific embodiments, the energy levels of the CT scan can be set to 3, 4, 5, or more. Under each energy level condition, the corresponding tube voltage, tube current, and scanning time for the CT scan are set, ultimately obtaining projected images under multiple energy level conditions.

[0014] After obtaining the projected image, it is preprocessed to finally obtain the CT grayscale image. The preprocessing steps include: dark field correction and flat field correction. Then, the cone-beam reconstruction algorithm is used to reconstruct the three-dimensional digital matrix at the corresponding energy, resulting in CT grayscale images under multiple energy level conditions.

[0015] Considering that the outer wall of the core is significantly affected by factors such as beam hardening, X-ray scattering, and cone-beam artifacts, while the inner wall cross-section is less affected, several reference points are set circumferentially along the CT scan path in this step. The core is then sectioned along the reference points and the core's central axis, and the CT scan images of the core section are taken for subsequent use.

[0016] In this step, the "reference points" are several points selected along the line where the outer surface of the core coincides with the CT scan path. These reference points are usually evenly distributed, and the number of reference points can be 4 to 6. For cylindrical cores, the reference points are set on the circumference, and this circumference coincides with the incident plane of the CT scan. For square lithology, the reference points are set on the outer surface. Considering the special characteristics of square cores, usually 4 reference points are set. These four reference points are the center points of the lines where the CT scan incident plane intersects with the four sides of the square core. For cylindrical cores, the central axis is the central axis; for square cores, it is the line connecting the centers of the two non-intersecting sides of the square core and the CT scan incident plane.

[0017] S2. Perform SEM scanning on the core and mineral analysis simultaneously to obtain a SEM grayscale image containing mineral information.

[0018] To suppress the charging effect during SEM scanning of rock cores, a thin layer of carbon or gold film is sprayed onto the core surface. Backscattered electron (BSE) imaging mode is used, where the signal intensity is positively correlated with the average atomic number of the minerals, generating a grayscale image reflecting the mineral composition. The surface of the rock sample is scanned, and appropriate accelerating voltage and working distance are set, while brightness and contrast are adjusted to acquire a high-resolution SEM grayscale image.

[0019] Subsequently, the core surface was scanned using the QEMSCAN automated mineral analysis system. Based on the QEMSCAN scan results, combined with the aforementioned BSE scan images, EDS energy dispersive spectroscopy analysis results, and the built-in mineral database, the SEM grayscale data and mineral category information were correlated to obtain an SEM grayscale image with mineral information.

[0020] As shown in S1, several core profile CT scan images were obtained. In order to correspond them, in this step, after obtaining the SEM grayscale image containing mineral information on the outer surface of the core, the core needs to be cut and the operation of this step is repeated to obtain several core profile SEM grayscale images containing mineral information.

[0021] Those skilled in the art will understand that when cutting the core, it is necessary to strictly follow the method in S1: that is, to cut using the same number of reference points at the same location plus the core axis, so that the final core profile contains SEM grayscale images of mineral information and CT scans of the core profile. Figure 1 One-to-one correspondence.

[0022] S3. Register the SEM grayscale image and the CT grayscale image to obtain an equivalent SEM grayscale image, and use a correction function to correct the registration result; S31. Establish three-dimensional coordinate systems for the SEM grayscale image and the CT grayscale image respectively, and map the SEM pixel coordinates to the CT three-dimensional coordinates through a rigid body model. In practice, feature points can be selected as positioning points. These feature points can be: 1. natural pores on the core surface or cut surface; 2. intersections of mineral grain boundaries. Using these feature points, SEM pixel coordinates can be quickly mapped to CT 3D coordinates. In the formula, Represents the CT three-dimensional coordinate system; Represents the two-dimensional coordinates along the SEM three-dimensional coordinate system. Directional scaling S u , S v The resulting three-dimensional vector (Z=0). μm , R The 3x3 rotation matrix is ​​responsible for performing rotation transformations on the source coordinates and is dimensionless.T This indicates that the translation vector translates the position after the rotation. μm .

[0023] S32. Considering that the CT three-dimensional volume pixel size is larger than the SEM three-dimensional volume pixel size, the gray values ​​corresponding to all SEM three-dimensional volume pixels within each CT three-dimensional volume pixel are counted, and the equivalent SEM gray value corresponding to each CT three-dimensional volume pixel is calculated by the area-weighted average method. Since the spatial resolution of CT is usually greater than that of SEM, meaning that the resolution of CT is lower than that of SEM, the size of CT three-dimensional voxels is significantly larger than that of SEM two-dimensional pixels. One CT voxel covers multiple SEM pixels, forming a one-to-many correspondence.

[0024] Therefore, we first count the gray values ​​of all SEM pixels within the region corresponding to each CT voxel, and then use an area-weighted average method to calculate the SEM equivalent gray value corresponding to each CT voxel location: In the formula, GSEM represents the SEM equivalent gray value of the corresponding CT voxel, which is dimensionless; A j Indicates the first j Area occupied by each SEM pixel μm 2 ; G j Indicates the first j The SEM grayscale value of each SEM pixel, dimensionless; n This indicates the number of SEM pixels within the corresponding CT voxel range; j Indicates the SEM pixel number. j =1,2,3,4… n .

[0025] S33. Obtain the mineral category with the largest proportion of QEMSCAN mineral pixels within each CT 3D volume pixel, and use this mineral category as the mineral label for that CT 3D volume pixel; through this operation, the SEM equivalent grayscale value at the CT voxel scale and the corresponding mineral category label can be established.

[0026] In particular, when two or more minerals have the same area ratio, the mineral with the closest distance between its center and the center of the CT voxel is used as the final mineral label.

[0027] S34. Several reference points are set circumferentially along the CT scan path. A circular area with a radius of the reference points as the first threshold is taken as the reference area. Based on the deviation of CT 3D volume pixels with the same equivalent SEM gray value outside and inside the reference area, considering spatial position and CT scan energy level, a correction function is established, and the corrected CT gray value outside the reference area is obtained based on the correction function. CT scan results are affected by factors such as beam hardening, X-ray scattering, cone-beam geometry, and equipment parameters, causing the CT grayscale value response of the same mineral in the core edge region to deviate from its true value, resulting in bias. This bias directly leads to errors in analysis based on CT grayscale values. Therefore, it is necessary to correct the CT value bias in the edge region at different energies to provide a basis for obtaining a unified corrected CT grayscale value through subsequent multi-energy fusion. The grayscale value in the core center region, due to its shorter X-ray path and less scattering interference, is largely unaffected by the above effects and can accurately reflect mineral characteristics, serving as a standard reference. Therefore, the CT grayscale data for each energy are processed as follows: After completing the spatial registration of CT and SEM data, the deviation of samples with the same SEM equivalent gray value at each spatial location from the reference area is statistically analyzed within a circular region with the radius around the reference point as the first threshold. Then, a correction function for CT gray value as spatial location parameters and scanning parameters change is established.

[0028] , In the formula: C k Indicates the first k The grayscale correction factor for energy CT is dimensionless. x , y , z Represents spatial coordinates, dimensionless; U k This indicates the CT tube voltage, in V; I k This represents the CT tube current, in A; k Indicate the energy number: and k = 1, 2, ..., n; G k UI represents the original CT grayscale value corresponding to the k-th energy CT. This indicates the correction of CT grayscale values. UI .

[0029] The original CT grayscale value of each energy is corrected by a correction function to obtain a corrected CT grayscale value, so that regions with the same mineral composition characteristics can obtain a consistent CT grayscale response, thereby improving the accuracy of the correspondence between CT grayscale values ​​and real mineral composition.

[0030] S35. Obtain the CT grayscale image of the corresponding section of the SEM scan as the section CT grayscale image; and register the section CT grayscale image and the section SEM grayscale image according to steps S31~33. S36. Establish the correspondence between the CT grayscale image of the cross section and the grayscale value of the equivalent SEM of the cross section, and use it as a verification dataset to verify the accuracy of the CT grayscale correction model established in S34. S37. Record the average relative error of all points where the relative deviation between the real CT gray value and the corrected CT gray value is greater than the second threshold. When the average relative error is greater than the third threshold, adjust the correction function until the average relative error is no greater than the third threshold.

[0031] The measured CT grayscale values ​​at each point of the newly added cross-section are used as the true CT values. The relevant parameters are input into the calibration model to obtain the corrected CT value at that point. The accuracy of the model is evaluated by comparing the deviations between the two values. To rigorously assess the calibration model's ability to handle significant deviations, validation focuses on regions with large deviations. First, the percentage relative deviation between the calibrated CT value and the true CT value at each validation point is calculated: Collect all points whose relative deviation is greater than the second threshold, and let the number of these points be M. Calculate the average relative error of these points. E k and with E k A value less than or equal to the third threshold is used as the criterion for determining the accuracy of the calibration model. The formula is as follows: In the formula: Indicates the first k The first energy CT under the first i The relative deviation of each screening point is dimensionless. G true,i,k Indicates the first k The first energy CT under the first i The measured CT grayscale values ​​corresponding to each validation sample UI ; G pred,i,k Indicates the first k The first energy CT grayscale correction model obtained the first i CT values ​​of each sample UI ; E k This represents the average relative error (%), and is dimensionless. M Indicates the number of samples participating in the validation; dimensionless. i Indicates the sample number, 1, 2, 3, 4, ... M .

[0032] If the average relative error between the calibration model's calibration result and the actual CT measurement result exceeds the third threshold, the calibration function parameters are readjusted based on the error distribution. The verification and optimization process is repeated until the average relative error... E k The value is less than or equal to the third threshold to ensure a stable and reliable CT grayscale value correction model for each energy.

[0033] In this step, the first threshold is typically one-twentieth to one-tenth of the core diameter (cylindrical core) or one-tenth to one-fifth of the shorter side length of the plane containing the reference point. The second threshold is typically set to 1%; and the third threshold is typically set to 3%.

[0034] S4. Based on the error between the measured grayscale value and the corrected grayscale value of the CT grayscale image under different energy level conditions, determine the weight of the CT grayscale image under different energy level conditions; combine the weights and perform weighted fusion of the CT grayscale images under different energy level conditions to obtain the fused CT grayscale image. S41. Based on the relative error between the measured gray value and the corrected gray value of each CT voxel under different energy level conditions, calculate the weight of different energy levels under the CT voxel conditions using the normalization method. S42. The fused CT grayscale value of each CT voxel is calculated by weighted summation to obtain the fused CT grayscale image.

[0035] Based on the validation results of the grayscale correction models for each energy CT in step six, determine the weights of different energy CT at different spatial locations in multi-energy fusion correction. For the first... k The validation error of a single-energy CT scan is denoted as . E k Let the first i The spatial coordinates of the verification points are as follows: , No. k The single-point relative error of an energy at this point is: Using the spatial coordinates of the verification point and CT energy sequence number k Using the input as the basis and the corresponding single-point relative error as the output, a spatial error distribution model for the k-th energy CT is established, and its expression is: In the formula, Indicates the first k A single energy CT in spatial location Single-point relative error, % Indicates the first k A spatial error distribution model for energy CT; Represents the spatial coordinates of a CT voxel; dimensionless. k Indicates the energy number, and k =1,2,3,…, n .

[0036] Because X-rays of different energies have varying penetrating power, beam hardening degree, and scattering response, the reliability of CT scans at different spatial locations of rock samples varies. A smaller local prediction error indicates that the corrected grayscale value of the CT scan at that voxel location is closer to the true CT grayscale response, resulting in higher data reliability and a larger corresponding fusion weight. Conversely, a larger local prediction error corresponds to a smaller fusion weight.

[0037] The inverse normalization method for local prediction error is used to determine the first... k The energy CT at the first i The fusion weight under each CT voxel is expressed as follows: In the formula, This represents the fusion weight of the k-th energy CT on the i-th voxel, which is dimensionless; This represents the prediction error of the k-th energy CT on the i-th voxel.

[0038] After determining the fusion weights of each energy CT at each CT voxel location, the preliminary corrected CT grayscale values ​​obtained in step five are applied to the original multi-energy CT three-dimensional grayscale data of the entire rock sample. For any CT voxel, its original CT grayscale values ​​under different energy conditions are read. G k And based on the spatial coordinates of the voxel , No. k The tube voltage corresponding to each energy CT U k and tube current I k Calculate the preliminary corrected CT grayscale value at the corresponding energy. Subsequently, utilizing the local fusion weights of each voxel at different energies, a voxel-by-voxel weighted fusion was performed on the energy-corrected CT grayscale values ​​to obtain the unique multi-energy fusion-corrected CT grayscale value for that voxel. Its expression is: In the formula: This represents the CT grayscale value after multi-energy fusion correction. UI .

[0039] S5. Using the registration results of SEM grayscale images and CT grayscale images, a training sample set is constructed to train XGBoost. After training, the grayscale values ​​of the fused CT grayscale image are substituted into the trained XGBoost model to obtain an equivalent fused SEM grayscale image. Combined with the SEM grayscale image containing mineral information from S2, the mineral composition inside the rock sample is obtained. In this step, XGBoost is a common machine learning model in this field, so its specific architecture will not be described in detail here.

[0040] The training set is constructed using the registration results of SEM grayscale image and CT grayscale image in S3. The CT grayscale value in the CT grayscale image is used as input and the corresponding SEM equivalent grayscale value is used as output to train the XGBoost model, thereby constructing a nonlinear mapping relationship between CT grayscale value and SEM equivalent grayscale value.

[0041] After obtaining the nonlinear mapping relationship, the fused CT gray value in the fused CT gray image obtained in S4 is used as input and substituted into the trained XGBoost model to obtain the equivalent fused SEM gray value corresponding to the fused CT gray value; all the equivalent fused SEM gray values ​​in the core are combined according to their positional relationship to obtain the equivalent fused SEM gray image.

[0042] Finally, using the SEM grayscale image containing mineral information in S2, mineral labels are assigned to the equivalent fused SEM grayscale image, thus obtaining a three-dimensional mineral composition distribution model.

[0043] To facilitate a better understanding of the present invention by those skilled in the art, specific test examples are provided below.

[0044] (1) A cube specimen with a side length of 10 mm was drilled and finely ground from a full-diameter sandstone core to ensure that all opposite faces were flat and parallel to reduce geometric artifacts. After being washed with organic solvent and dried to constant weight, the specimen was fixed on a multi-energy X-ray CT sample stage using a low-density clamp. Three energy levels were set for CT scanning, denoted as follows: E 1, E 2, E 3. Among them, E 1 corresponds to a tube voltage of 80kV and a tube current of 80μA; E 2 corresponds to a tube voltage of 110kV and a tube current of 100μA; E 3. Corresponding tube voltage 140kV and tube current 120μA. A suitable metal filter was selected to pre-harden the X-ray. The sample stage rotated 360° at a constant speed with a small angular step, and the detector continuously acquired transmission projection images from various angles. After scanning, dark field correction, flat field correction, and scattering correction were performed on the projection data. A cone-beam reconstruction algorithm was used to reconstruct three sets of 3D CT grayscale data volumes with a voxel size of 4μm, denoted as follows: G 1, G 2, G 3.

[0045] Using the center point of the intersection line between the CT scan incident plane of the cube rock sample and the four sides of the square rock core as the reference point, the two relative reference points and the center line of the cube rock sample are used as cross-sections to obtain two CT gray value cross-sections inside the cube rock sample that pass through the center line, and obtain its CT gray value data.

[0046] (2) The six outer surfaces of the cubic rock sample after CT scanning were scanned by SEM-BSE to obtain two-dimensional SEM grayscale images of the corresponding surfaces. The SEM pixel size was set to 1 μm. Subsequently, the same scanned area was identified by the QEMSCAN automatic mineral analysis system to obtain the mineral category information corresponding to each SEM pixel, forming a mineral category map corresponding to the SEM grayscale image.

[0047] After completing the scanning of the six outer surfaces, the center point of the intersection line between the CT scan incident plane of the cubic rock sample and the four sides of the square core was used as the reference point. The two opposite reference points and the center line of the cubic rock sample were then used as cross-sections for laser cutting, dividing it into four equal cuboids, thus obtaining two new internal sections. SEM-BSE scanning and QEMSCAN mineral identification were also performed on these new internal sections to obtain SEM grayscale data and pixel-level mineral category data.

[0048] (3) Register the SEM grayscale image and the CT grayscale image, and use the correction function to correct the registration result to obtain the equivalent SEM grayscale image corresponding to the CT grayscale image.

[0049] Since the CT voxel size is larger than the SEM pixel size, an area-weighted average method is used to calculate the SEM equivalent gray value for SEM pixels falling within the same CT voxel range. Figure 2 As shown. Then, the CT gray values ​​corresponding to samples with the same SEM equivalent gray values ​​are statistically analyzed, and a circular area with the center of the six outer surface points of the rock sample as the center and one-tenth of the side length as the radius is used as the reference area to establish a CT gray value correction model to correct the original CT gray value data.

[0050] In the formula: x , y , z These are spatial coordinates, dimensionless; U k CT tube voltage, V; I k CT tube current, A; The kth one; CT value corrected by the correction function UI .

[0051] Simultaneously, the area percentage of mineral categories identified by QEMSCAN within the CT voxel range is statistically analyzed. If quartz pixels account for 67.5%, feldspar pixels for 20.0%, calcite pixels for 7.5%, and clay mineral pixels for 5.0% within the voxel, then quartz, with the largest area percentage, is identified as the mineral label for that CT voxel. This establishes a "SEM equivalent grayscale value - mineral label" dataset at the CT voxel scale.

[0052] The configuration process is as follows Figure 1 As shown, Figure 1 In this context, 'a' represents a SEM scan image containing mineral information; Figure 1 In this context, 'b' represents the CT grayscale image.

[0053] Finally, after multiple adjustments to the correction function, the result is as follows: Figure 3 As shown, from Figure 3 It can be seen that by adjusting the correction function multiple times, the average relative error can be effectively reduced, making the correction function more accurate.

[0054] (4) Based on the error of the measured gray value and the corrected gray value of the CT gray image under different energy level conditions, determine the weight of the CT gray image under different energy level conditions; combine the weights and perform weighted fusion of the CT gray images under different energy level conditions to obtain the fused CT gray image.

[0055] After calibration, the percentage of relative deviation between the measured CT data of the newly added cross-section and the data corrected by the CT grayscale correction model at the corresponding location is calculated, and data with a deviation greater than 1% are counted; then the average relative error of the counted data is calculated to obtain... E k The result was >3%, indicating the validation failed. Therefore, the corrected model parameters were refitted, and the validation and optimization process was repeated multiple times. The validation results showed: E 1 = 2.1%, E 2 = 1.5%, E 3 = 2.8% Mean relative error E k If the error is below 3%, the CT grayscale correction model passes validation. The fusion weights for each energy CT value are then calculated based on the validation error. Substituting the above error values, we get: w 1 = 0.317, w 2 = 0.444, w 3 = 0.239 The validated CT grayscale correction models for each energy level were used to correct the entire rock sample's 3D CT data, obtaining the corrected 3D CT grayscale values ​​for each energy level. For any CT voxel, the corrected CT grayscale values ​​from the three energy levels were weighted and fused: Therefore, voxel-by-voxel calculations were performed on the entire rock sample to obtain a fused CT grayscale image of the same size and resolution as the original CT data.

[0056] (5) Use the registration results of SEM grayscale image and CT grayscale image to construct a training sample set to train XGBoost. After training, substitute the grayscale value of the fused CT grayscale image into the trained XGBoost model to obtain an equivalent fused SEM grayscale image. Combine the SEM grayscale image containing mineral information in S2 to obtain the mineral composition inside the rock sample.

[0057] Further multi-energy fusion correction of CT grayscale values and spatial coordinates Using the corresponding SEM equivalent grayscale value as input, an XGBoost mapping model between the two is established.

[0058] The mapping model was used to predict the entire three-dimensional CT data and obtain the equivalent grayscale distribution of three-dimensional SEM inside the rock sample.

[0059] Based on the "SEM equivalent grayscale value - mineral label" dataset obtained in step (2), mineral classification is performed on the three-dimensional SEM equivalent grayscale data to obtain a three-dimensional mineral composition distribution model inside the rock sample. The identification results are as follows: Figure 4 As shown in the figure, the rock sample is mainly composed of quartz, feldspar, and calcite, with quartz accounting for 58.6% of the volume, feldspar for 27.3%, and calcite for 11.5%. The remainder consists of pores and a small amount of other minerals. The results indicate that this invention can achieve three-dimensional reconstruction of the internal mineral composition of rock samples based on multi-energy CT, SEM grayscale, and QEMSCAN mineral calibration information, obtaining high-precision three-dimensional mineral distribution information. This solves the problem of obtaining high-precision three-dimensional mineral distribution information in the field of digital core research.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A three-dimensional rock and mineral identification method combining CT and electron microscopy, characterized in that, Includes the following steps: S1. Core samples were taken and subjected to multi-level CT scans. The scan images were preprocessed to obtain CT grayscale images under multi-level conditions. S2. Perform SEM scanning on the core and mineral analysis simultaneously to obtain a SEM grayscale image containing mineral information. S3. Register the SEM grayscale image and the CT grayscale image to obtain an equivalent SEM grayscale image, and use a correction function to correct the registration result; S4. Based on the error between the measured grayscale value and the corrected grayscale value of the CT grayscale image under different energy level conditions, determine the weight of the CT grayscale image under different energy level conditions. By combining weights, CT grayscale images under different energy level conditions are weighted and fused to obtain a fused CT grayscale image; S5. Use the registration results of SEM grayscale image and CT grayscale image to construct a training sample set to train XGBoost. After training, substitute the grayscale value of the fused CT grayscale image into the trained XGBoost model to obtain an equivalent fused SEM grayscale image. Combine the SEM grayscale image containing mineral information in S2 to obtain the mineral composition inside the rock sample.

2. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 1, characterized in that, In S1, after the CT scan is completed, the projection data is corrected for dark field and flat field, and then the cone-beam reconstruction algorithm is used to reconstruct the three-dimensional digital matrix under the corresponding energy to obtain the CT grayscale image.

3. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 1, characterized in that, In S2, the SEM grayscale image containing mineral information is obtained through the following steps: After obtaining the SEM grayscale data, the SEM scanning area is scanned using the QEMSCAN automatic mineral analysis system. The SEM grayscale data and mineral category information are then correlated by combining the backscattered electron image, EDS energy dispersive spectroscopy analysis results, and mineral database.

4. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 3, characterized in that, In S2, when performing SEM scanning, the outer surface of the core is first scanned, and then several profiles are set inside the core. SEM scanning is then performed along the profiles to obtain the profile SEM grayscale image. The profiles pass through at least one reference point and the core axis at the same time.

5. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 4, characterized in that, In S3, registration includes the following steps: S31. Establish three-dimensional coordinate systems for the SEM grayscale image and the CT grayscale image respectively, and map the SEM pixel coordinates to the CT three-dimensional coordinates through a rigid body model. S32. Considering that the CT three-dimensional volume pixel size is larger than the SEM three-dimensional volume pixel size, the gray values ​​corresponding to all SEM three-dimensional volume pixels within each CT three-dimensional volume pixel are counted, and the equivalent SEM gray value corresponding to each CT three-dimensional volume pixel is calculated by the area-weighted average method. S33. Obtain the mineral category with the largest proportion of QEMSCAN mineral pixels within each CT 3D volume pixel, and use this mineral category as the mineral label for that CT 3D volume pixel; S34. Set several reference points circumferentially along the CT scan path. Take the circular area with the radius around the reference points as the first threshold as the reference area. Based on the deviation of CT three-dimensional volume pixels with the same equivalent SEM gray value outside and inside the reference area, consider the spatial position and CT scan energy level, establish a correction function, and obtain the corrected CT gray value outside the reference area based on the correction function.

6. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 5, characterized in that, Also includes: S35. Obtain the CT grayscale image of the corresponding section of the SEM scan as the section CT grayscale image; and register the section CT grayscale image and the section SEM grayscale image according to the steps of S31~33 and calculate the equivalent SEM grayscale value of the section. S36. Establish the correspondence between the CT grayscale image of the cross section and the grayscale value of the equivalent SEM of the cross section, and use it as a verification dataset to verify the accuracy of the CT grayscale correction model established in S34. S37. Record the average relative error of all points where the relative deviation between the real CT gray value and the corrected CT gray value is greater than the second threshold. When the average relative error is greater than the third threshold, adjust the correction function until the average relative error is no greater than the third threshold.

7. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 6, characterized in that, S4 includes the following steps: S41. Based on the relative error between the measured gray value and the corrected gray value of each CT voxel under different energy level conditions, calculate the weight of different energy levels under the CT voxel conditions using the normalization method. S42. The fused CT grayscale value of each CT voxel is calculated by weighted summation to obtain the fused CT grayscale image.

8. The three-dimensional rock and mineral identification method combining CT and electron microscopy according to claim 1, characterized in that, S5 includes the following sub-steps: Establish an XGBoost model, using the fused CT grayscale value from the fused CT grayscale image as input and the equivalent SEM grayscale value from the equivalent SEM grayscale image in S3 as output, and train the XGBoost model; substitute the fused grayscale value from the fused CT grayscale image in S4 into the trained XGBoost model to obtain the equivalent fused SEM grayscale image corresponding to the fused CT grayscale image; combine the SEM grayscale image containing mineral information to identify the mineral category corresponding to the CT voxels in the fused CT grayscale image.

Images