A lithology identification method of image and feature information fusion

Abstract

The present application provides a kind of lithology identification method of image and feature information fusion, comprising: step 1, original cutting image acquisition;Step 2, original image pre-processing;Step 3, cutting image segmentation;Step 4, cutting sample feature extraction;Step 5, element parameter combination;Step 6, analytic hierarchy process establishes cutting feature analysis model;Step 7, data normalization is carried out to the extracted feature;Step 8, establish lithology standard library;Step 9, determine the category to which the cutting sample belongs.The present application combines cutting image feature information with logging data, enhances the stability of the features, and improves the recognition accuracy.

Description

Technical Field

[0001] This invention belongs to the field of rock cuttings identification technology, and in particular relates to a lithology identification method that fuses image and feature information. Background Technology

[0002] In the exploration and development of oil, cuttings logging is an important component of geological logging, playing a crucial role in oil and gas displays and reservoir evaluation. Therefore, the quality of cuttings logging determines the accuracy and reliability of geological data, which in turn affects subsequent data interpretation and evaluation results.

[0003] Rock cuttings are invaluable physical data from cuttings logging. In traditional drilling processes, drill bits typically use roller cone bits and scraper bits to crush or shear the formation, resulting in relatively large rock cuttings that can be identified by the naked eye. With the development of drilling technology, PDC (Programmable Diameter) bits, with their high hardness and wear resistance, have become a widely used rock-breaking tool. However, due to the greater degree of rock fragmentation by PDC bits, the resulting rock cuttings are extremely fine, making traditional methods of geological lithology analysis using the naked eye or microscopes very difficult. Furthermore, manually observing the surface characteristics of rock cuttings with simple optical instruments is heavily influenced by subjective experience, resulting in significant differences in characteristic descriptions, long processing times, and difficulties in quantitative analysis of samples.

[0004] To address these issues, researchers proposed using computer image recognition technology to quickly extract features such as color, texture, and shape from rock debris images. Based on these features, machine learning methods, such as support vector machines and K-means clustering classifiers, are then used for lithological analysis of the rock debris. However, uncertainties during the rock debris collection process can lead to instability in the features of the collected rock debris images. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a lithology identification method that integrates image and feature information. By combining the feature information of rock cuttings images with well logging data, the stability of the features is enhanced and the identification accuracy is improved.

[0006] The method of this invention combines the feature information of rock cuttings images with well logging data, wherein the well logging data includes element combinations and drilling parameters. It introduces the hierarchical analysis method to determine feature weights and the centroid-sample point weighted distance algorithm, which makes up for the defects of manual identification methods and the inadequacy of image feature identification of rock cuttings lithology, and improves the accuracy of lithology identification.

[0007] The present invention adopts the following technical solution:

[0008] A lithology identification method that fuses image and feature information includes the following specific steps:

[0009] Step 1. Acquiring the original rock cuttings image;

[0010] The collected rock fragment samples were placed on a measurement platform, magnified by an optical microscope, photographed by a camera, and then the images were stored in a computer.

[0011] Step 2. Preprocess the original image

[0012] A median filter with a 3x3 sliding window is used to remove potential noise from the original image. Simultaneously, the median filter effectively preserves the fine edge information of the original microscopic image, preventing edge blurring and facilitating subsequent image segmentation operations.

[0013] The expression for the two-dimensional median filter function is as follows: ;

[0014] Where f and g are the original image and the processed image, respectively, and x and y represent the pixel coordinates.

[0015] After removing noise with a median filter to obtain a smooth image, the Sobel operator is used to sharpen the image. This operator contains two sets of 3x3 matrices, one horizontal and one vertical. By performing planar convolution with the image, approximate values ​​of brightness difference in the horizontal and vertical directions can be obtained, making the image clearer and highlighting more details.

[0016] The Sobel operator is as follows: ;

[0017] in, and These represent images that have undergone horizontal and vertical edge detection, respectively.

[0018] The third step is to segment the image.

[0019] The processed image is converted to grayscale and then binarized to obtain the final image. The Sobel operator is then used to calculate the values ​​of each image. and Afterwards, by The gray-level gradient of a pixel (x, y) in a microscopic image can be obtained. The gray-level amplitude image of the image is stored as a data matrix, where each value in the data matrix corresponds to the gray-level value of the pixel in this gradient image. Based on this, the watershed segmentation method is used to segment the image.

[0020] The fourth step is to extract image features from the rock cuttings samples.

[0021] Based on step three, color and texture features are extracted from each individual rock fragment sample obtained from the segmentation.

[0022] Because the HSV color model is more natural and closer to human visual experience and color perception than the RGB color space, and experiments have shown that the HSV color histogram method is significantly superior to the RGB color histogram method. Based on the above analysis, the HSV color space model is used to extract color features, including hue, saturation, and brightness. Since images are generally stored in RGB format, a linear transformation can be used to directly convert the RGB color space to the HSV space.

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] Hue calculates: ;

[0030] S calculation: ;

[0031] V calculation: ;

[0032] The color of each pixel can be uniquely represented by the three-dimensional coordinates of the three primary colors: red, green, and blue. R, G, and B are the coordinates corresponding to these colors in the RGB color space. , , In the RGB color model , , The result after normalization , They are respectively , , The maximum and minimum values ​​in the range. for and difference, For color tone, For saturation, For brightness. Plot an HSV color histogram and use the peak values ​​of each color as analysis parameters.

[0033] As described in step 4, for the extraction of rock debris texture features, a texture model that is more easily perceived by the visual system proposed by Tamura et al. is used, employing three pieces of information—roughness, contrast, and orientation—as texture features, where:

[0034] 1) Roughness reflects the rate of change of image pixel grayscale values. To calculate roughness, the size of the neighborhood must first be selected, i.e. ,in .

[0035] The formula for calculating the neighborhood mean of each pixel in an image is: ;

[0036] in, To select the value of the power exponent of the domain size and , For pixels The size of the grayscale value, Let x and y be the mean of the neighborhood of each pixel in the image, and let i and j be the horizontal and vertical coordinates of the pixel.

[0037] Then, for each pixel, the average grayscale difference between its non-overlapping windows in the horizontal and vertical directions needs to be calculated using the following formula:

[0038] Horizontal direction: ;

[0039] Vertical direction: ;

[0040] in, , Representing pixel coordinates The average grayscale difference between non-overlapping windows in the horizontal and vertical directions. and The mean of the neighborhood of the pixel coordinates at that point. , This represents the mean value of the neighborhood of the pixel coordinates at that point.

[0041] Compare the mean difference values ​​in the horizontal and vertical directions, and select the larger value to represent the mean difference value of the current pixel's neighborhood. ;

[0042] in, The mean difference of the neighborhood for pixels with larger mean differences in both the horizontal and vertical directions. , Representing pixel coordinates ( The average grayscale difference between windows that do not overlap in the horizontal and vertical directions.

[0043] Scan each pixel of the image and select and determine an optimal size from q neighborhoods of size L. ;

[0044] in, The optimal size determined among numerous dimensions across various fields.

[0045] Scan each pixel of the image and select and determine an optimal size from q neighborhoods of size L. ,satisfy: ;

[0046] in, The difference between the mean and the mean of the region. q represents the maximum value among the differences in the mean of the neighborhood, where L is the number of neighborhood sizes.

[0047] Roughness is calculated by measuring the roughness of the entire image. It is obtained by averaging the values, and its expression is: ;

[0048] Where m and n are pixel values. This represents the roughness of the image.

[0049] 2) Contrast describes the brightness of an image. Calculating contrast requires first calculating kurtosis (fourth-order standard distance). ,in This is a fourth-order matrix representing the image's grayscale values. Let V be the variance of the image gray levels. The expression for contrast can be obtained as: ,in, This refers to the contrast of the image.

[0050] 3) Directionality is a global characteristic of a texture region, describing a state of the image, specifically a state where it is straight or approximately straight in a certain direction. First, it is necessary to calculate the magnitude and angle of the gradient vector of each pixel in the image:

[0051] First, the 3x3 matrix surrounding the pixel is compared with the matrix... sum matrix The convolution is obtained and The formula for calculating the magnitude of the gradient vector is: ;

[0052] in, and These represent the gradient changes in the horizontal and vertical directions, respectively. Let be the magnitude of the gradient vector of the pixel.

[0053] Formula for calculating the angle of the gradient vector: ;

[0054] in, The angle of the gradient vector of the pixel.

[0055] After obtaining the magnitude and angle of the gradient vector of all pixels, establish Histograms ,statistics The number of pixels exceeding a given threshold. (In the histogram) In this context, the directionality of an image can be determined by observing its peak values ​​and flatness. The overall directionality of an image can be calculated from the sharpness of the peak values ​​in the histogram. ,in, The overall orientation of the image. For function The corresponding value for obtaining the maximum value , This represents the peak value label in the histogram. Indicates the number of peaks. The range encompassed by this peak value. for The corresponding value for obtaining the maximum value , describe Histogram of distribution.

[0056] Step 5, Element Feature Extraction

[0057] By observing and analyzing the element mass fractions in the HB-X200 national standard element data table, typical elemental characteristics under specific lithologies are obtained. Then, some combinations of elements are added to highlight the differences between different lithologies, thereby improving the accuracy of the discrimination model.

[0058] Step 6. Based on all the features extracted in Steps 4 and 5, a rock cuttings feature analysis model is established using the analytic hierarchy process (AHP).

[0059] 1) A multi-level progressive structure model was established, in which the index layer consisted of two parts: image features and element parameters. The image features included six features: hue, saturation, brightness, roughness, contrast, and orientation; the element parameters included seven features: Na+K, Ti, Na, Fe, Si, Na+Ca, and Fe+Mg.

[0060] 2) Determine the evaluation criteria for the relative importance of factors, and construct judgment matrices at each level based on these criteria.

[0061] 3) Calculate element weights. Use the root value method: Let... Judgment matrix A: Multiply the elements of each row in A: ,, Let n represent each element in an n*n matrix and find the nth root of the product of all rows. ,in, This is the product of the elements in each row of matrix A. Then... Normalization process: Obtain the weights of each influencing factor. For each influencing factor The weight.

[0062] 4) Consistency check. First, by Calculate the consistency index of the matrix, where This is a consistency index, where n is the order of the judgment matrix. Let the largest eigenvalue of the matrix be the random consistency ratio, and then calculate the consistency index of the judgment matrix: ,in, This represents the consistency index of the judgment matrix. The random consistency ratio, The average random consistency index, When n=13, the value is 1.56. If the condition is met, it indicates that the judgment matrix has no logical errors and can be used; if... If so, the judgment matrix needs to be readjusted and a consistency check needs to be performed again until... .

[0063] 5) Overall Hierarchical Ranking and Consistency Check. Calculate the comprehensive weight of each element in the indicator layer relative to the overall goal, and perform a consistency check on the overall hierarchical ranking to reduce inconsistencies in the ranking results. The weights of each influencing factor in the object layer are obtained from step 3). Influencing factors and criteria at the object level The consistency index of the single ranking of the relevant factors constituting the judgment matrix is ​​as follows: The corresponding average random consistency index Then the overall random consistency ratio of the scheme layer is: , Represents the various effects of the object layer Factor weights, when If the overall hierarchical ranking is passed, then the consistency index is considered to be valid.

[0064] Step 7: Normalize the extracted features.

[0065] To eliminate the influence of different dimensions among the indicators, deviation standardization was used to perform a linear transformation on the original characteristic data of each individual rock fragment after segmentation: ,in and These represent the maximum and minimum values ​​of the sample data, respectively. The result of the linear transformation is the multidimensional vector of the rock cuttings sample. This represents the current image feature or element feature sample value.

[0066] Step 8: Establishment of the lithology template library

[0067] Step 4 involves extracting image features from known lithological rock fragments (mudstone, sandstone, conglomerate, granite, basalt), adding the elemental features obtained in Step 5, and then performing a normalization operation on these features in Step 7 to obtain a multidimensional vector for a specific lithology.

[0068] Then, in step 6, the features are normalized to establish a standard lithology template library.

[0069] Step 9: Determine the type of rock cuttings sample.

[0070] 1) Assume a specific lithology in the template library The feature vector of the lower cuttings sample is in, These represent the values ​​for all extracted features. Calculate the centroid coordinates of n group points for this lithology: Based on the feature weights calculated using the analytic hierarchy process in step 6, it can be seen that different features have different weights. Let's assume that rock fragments have different weights depending on the feature... The weights below are Combining the Euclidean distance calculation method, for two different sample points and Calculate the weighted distance: .

[0071] 2) Calculate the different lithologies in the template library The centroid of the debris cluster below , Given a rock cuttings sample to be identified Based on the calculation method given in 1), the sample to be tested is calculated. Five different rock properties Weighted distance Find the smallest value in the set. Corresponding rock properties core That is, the sample to be tested Lithology .

[0072] 3) After determining the lithology of the points to be measured, the centroid of the group of points is re-determined using the centroid calculation formula: .

[0073] After the above steps, the lithology of rock fragments can be identified.

[0074] The beneficial effects of this invention are:

[0075] This invention provides a lithology identification method that integrates image and feature information. This method combines the feature information of rock cuttings images with well logging data, introduces the analytic hierarchy process (AHP) to determine feature weights, and employs a centroid-sample point weighted distance algorithm. This method overcomes the shortcomings of manual identification methods and the inadequacy of image feature identification of rock cuttings lithology, thereby improving the accuracy of lithology identification. Attached Figure Description

[0076] Figure 1 This is a flowchart of the process of the present invention;

[0077] Figure 2 This is a grayscale image of the original rock fragments.

[0078] Figure 3 These are images of rock debris after image preprocessing.

[0079] Figure 4 This is a hierarchical structure diagram of lithological analysis;

[0080] Figure 5 Here is a flowchart of the Analytic Hierarchy Process (AHP).

[0081] Figure 6 This is a hue histogram obtained by converting rock cuttings from RGB color space to HSV color space using a linear transformation (this image is only used to show the results of an intermediate step in the method, demonstrating that the RGB color space can be converted to HSV color space to obtain a hue histogram).

[0082] Figure 7 The image shows the simulation results of the rock fragment lithology. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0084] like Figure 1 As shown, the present invention provides a lithology identification method based on image and feature information fusion, comprising:

[0085] (1) Obtain the original rock fragment images.

[0086] The collected rock fragment samples were placed on a measurement platform, magnified by an optical microscope, photographed by a camera, and then the images were stored in a computer.

[0087] (2) Median filtering and Sobel operator are used to preprocess the original image.

[0088] A median filter with a 3x3 sliding window is used to remove potential noise from the original image. Simultaneously, the median filter effectively preserves the fine edge information of the original microscopic image, preventing edge blurring and facilitating subsequent image segmentation operations.

[0089] The expression for the two-dimensional median filter function is as follows: ;

[0090] Where f and g are the original image and the processed image, respectively, and x and y represent the pixel coordinates.

[0091] After removing noise with a median filter to obtain a smooth image, the Sobel operator is used to sharpen the image. This operator contains two sets of 3x3 matrices, one horizontal and one vertical. By performing planar convolution with the image, approximate values ​​of brightness difference in the horizontal and vertical directions can be obtained, making the image clearer and highlighting more details.

[0092] The Sobel operator is as follows: ;

[0093] in, and These represent images that have undergone horizontal and vertical edge detection, respectively.

[0094] (3) The watershed segmentation algorithm is used to segment the rock debris image.

[0095] The processed image is converted to grayscale and then binarized to obtain the final image. The Sobel operator is then used to calculate the values ​​of each image. and Afterwards, by The gray-level gradient of a pixel (x, y) in a microscopic image can be obtained. The gray-level amplitude image of the image is stored as a data matrix, where each value in the data matrix corresponds to the gray-level value of the pixel in this gradient image. Based on this, the watershed segmentation method is used to segment the image.

[0096] (4) Extract the color and texture features of the rock fragments sample. The extracted features include hue, saturation, brightness, roughness, contrast and orientation.

[0097] Based on step three, color and texture features are extracted from each individual rock fragment sample obtained from the segmentation.

[0098] Because the HSV color model is more natural and closer to human visual experience and color perception than the RGB color space, and experiments have shown that the HSV color histogram method is significantly superior to the RGB color histogram method. Based on the above analysis, the HSV color space model is used to extract color features, including hue, saturation, and brightness. Since images are generally stored in RGB format, a linear transformation can be used to directly convert the RGB color space to the HSV space.

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] Hue calculates: ;

[0106] S calculation: ;

[0107] V calculation: ;

[0108] The color of each pixel can be uniquely represented by the three-dimensional coordinates of the three primary colors: red, green, and blue. R, G, and B are the coordinates corresponding to these colors in the RGB color space. , , In the RGB color model , , The result after normalization , They are respectively , , The maximum and minimum values ​​in the range. for and difference, For color tone, For saturation, For brightness. Plot an HSV color histogram and use the peak values ​​of each color as analysis parameters.

[0109] As described in (4), for the extraction of rock debris texture features, a texture model that is more easily perceived by the visual system proposed by Tamura et al. is used, employing three pieces of information—roughness, contrast, and orientation—as texture features, where:

[0110] 1) Roughness reflects the rate of change of image pixel grayscale values. To calculate roughness, the size of the neighborhood must first be selected, i.e. ,in .

[0111] The formula for calculating the neighborhood mean of each pixel in an image is: ;

[0112] in, To select the value of the power exponent of the domain size and , For pixels The size of the grayscale value, Let x and y be the mean of the neighborhood of each pixel in the image, and let i and j be the horizontal and vertical coordinates of the pixel.

[0113] Then, for each pixel, the average grayscale difference between its non-overlapping windows in the horizontal and vertical directions needs to be calculated using the following formula:

[0114] Horizontal direction: ;

[0115] Vertical direction: ;

[0116] in, , Representing pixel coordinates The average grayscale difference between non-overlapping windows in the horizontal and vertical directions. and The mean of the neighborhood of the pixel coordinates at that point. , This represents the mean value of the neighborhood of the pixel coordinates at that point.

[0117] Compare the mean difference values ​​in the horizontal and vertical directions, and select the larger value to represent the mean difference value of the current pixel's neighborhood. ;

[0118] in, The mean difference of the neighborhood for pixels with larger mean differences in both the horizontal and vertical directions. , Representing pixel coordinates ( The average grayscale difference between windows that do not overlap in the horizontal and vertical directions.

[0119] Scan each pixel of the image and select and determine an optimal size from q neighborhoods of size L. ;

[0120] in, The optimal size determined among numerous dimensions across various fields.

[0121] Scan each pixel of the image and select and determine an optimal size from q neighborhoods of size L. ,satisfy: ;

[0122] in, The difference between the mean and the mean of the region. q represents the maximum value among the differences in the mean of the neighborhood, where L is the number of neighborhood sizes.

[0123] Roughness is calculated by measuring the roughness of the entire image. It is obtained by averaging the values, and its expression is: ;

[0124] Where m and n are pixel values. This represents the roughness of the image.

[0125] 2) Contrast describes the brightness of an image. Calculating contrast requires first calculating kurtosis (fourth-order standard distance). ,in This is a fourth-order matrix representing the image's grayscale values. Let V be the variance of the image gray levels. The expression for contrast can be obtained as: ,in, This refers to the contrast of the image.

[0126] 3) Directionality is a global characteristic of a texture region, describing a state of the image, specifically a state where it is straight or approximately straight in a certain direction. First, it is necessary to calculate the magnitude and angle of the gradient vector of each pixel in the image:

[0127] First, the 3x3 matrix surrounding the pixel is compared with the matrix... sum matrix The convolution is obtained and The formula for calculating the magnitude of the gradient vector is: ;

[0128] in, and These represent the gradient changes in the horizontal and vertical directions, respectively. Let be the magnitude of the gradient vector of the pixel.

[0129] Formula for calculating the angle of the gradient vector: ;

[0130] in, The angle of the gradient vector of the pixel.

[0131] After obtaining the magnitude and angle of the gradient vector of all pixels, establish Histograms ,statistics The number of pixels exceeding a given threshold. (In the histogram) In this context, the directionality of an image can be determined by observing its peak values ​​and flatness. The overall directionality of an image can be calculated from the sharpness of the peak values ​​in the histogram. ,in, The overall orientation of the image. For function The corresponding value for obtaining the maximum value , This represents the peak value label in the histogram. Indicates the number of peaks. The range encompassed by this peak value. for The corresponding value for obtaining the maximum value , describe Histogram of distribution.

[0132] (5) Observe and analyze the element mass fractions in the HB-X200 national standard element data table, obtain typical element characteristics and the combination forms between elements, and finally select the element characteristics including Na+K, Ti, Na, Fe, Fe+Mg, Na+Ca, and Si.

[0133] (6) A rock debris feature analysis model was established using the analytic hierarchy process (AHP), with the target layer being lithology identification. The index layer was set as image features and element features; the object layer consisted of all features extracted in steps 4 and 5.

[0134] 1) A multi-level progressive structure model was established, in which the index layer consisted of two parts: image features and element parameters. The image features included six features: hue, saturation, brightness, roughness, contrast, and orientation; the element parameters included seven features: Na+K, Ti, Na, Fe, Si, Na+Ca, and Fe+Mg.

[0135] 2) Determine the evaluation criteria for the relative importance of factors, and construct judgment matrices at each level based on these criteria.

[0136] 3) Calculate element weights. Use the root value method: Let... Judgment matrix A: Multiply the elements of each row in A: , Let n represent each element in an n*n matrix and find the nth root of the product of all rows. ,in, This is the product of the elements in each row of matrix A. Then... Normalization process: Obtain the weights of each influencing factor. For each influencing factor The weight.

[0137] 4) Consistency check. First, by Calculate the consistency index of the matrix, where This is a consistency index, where n is the order of the judgment matrix. Let the largest eigenvalue of the matrix be the random consistency ratio, and then calculate the consistency index of the judgment matrix: ,in, This represents the consistency index of the judgment matrix. The random consistency ratio, The average random consistency index, When n=13, the value is 1.56. If the condition is met, it indicates that the judgment matrix has no logical errors and can be used; if... If so, the judgment matrix needs to be readjusted and a consistency check needs to be performed again until... .

[0138] 5) Overall Hierarchical Ranking and Consistency Check. Calculate the comprehensive weight of each element in the indicator layer relative to the overall goal, and perform a consistency check on the overall hierarchical ranking to reduce inconsistencies in the ranking results. The weights of each influencing factor in the object layer are obtained from step 3). Influencing factors and criteria at the object level The consistency index of the single ranking of the relevant factors constituting the judgment matrix is ​​as follows: The corresponding average random consistency index Then the overall random consistency ratio of the scheme layer is: , Represents the various effects of the object layer Factor weights, when If the overall hierarchical ranking is passed, then the consistency index is considered to be valid.

[0139] (7) Use deviation standardization to perform linear transformation on the original feature data of the segmented rock cuttings to obtain the multidimensional vector of the rock cuttings sample.

[0140] To eliminate the influence of different dimensions among the indicators, deviation standardization was used to perform a linear transformation on the original characteristic data of each individual rock fragment after segmentation: ,in and These represent the maximum and minimum values ​​of the sample data, respectively. The result of the linear transformation is the multidimensional vector of the rock cuttings sample.

[0141] (8) Perform steps 4 and 5 on the rock fragments of known lithology to extract the image features of the rock fragments, and then normalize the features in step 7 to design a test program and establish a lithology standard library.

[0142] Step 4 involves extracting image features from known lithological rock fragments (mudstone, sandstone, conglomerate, granite, basalt), adding the elemental features obtained in Step 5, and then performing a normalization operation on these features in Step 7 to obtain a multidimensional vector for a specific lithology.

[0143] Then, in step 6, the features are normalized to establish a standard lithology template library.

[0144] (9) Calculate the centroid of the scatter point set under different lithologies. Based on the feature weights calculated by the hierarchical analysis method in (6), for a given rock cutting sample to be identified, calculate the weighted Euclidean distance from the sample to be tested to the centroid coordinates of different lithologies by the weighted Euclidean distance calculation method. Then compare the distances, and the one with the smallest value is the category to which the sample point belongs.

[0145] 1) Assume a specific lithology in the template library The feature vector of the lower cuttings sample is in, These represent the values ​​for all extracted features. Calculate the centroid coordinates of n group points for this lithology: Based on the feature weights calculated using the analytic hierarchy process in step 6, it can be seen that different features have different weights. Let's assume that rock fragments have different weights depending on the feature... The weights below are Combining the Euclidean distance calculation method, for two different sample points and Calculate the weighted distance: .

[0146] 2) Calculate the different lithologies in the template library The centroid of the debris cluster below , Given a rock cuttings sample to be identified Based on the calculation method given in 1), the value of the sample to be tested is calculated. Five different rock properties Weighted distance Find the smallest value in the set. Corresponding rock properties core That is, the sample to be tested The lithology it belongs to is k.

[0147] 3) After determining the lithology of the points to be measured, the centroid of the group of points is re-determined using the centroid calculation formula: .

[0148] After the above steps, the lithology of rock fragments can be identified.

[0149] Example

[0150] like Figure 2-3 As shown, taking the basic drilling data of the Archean strata in the Bohai Bay Basin as an example, 1551 samples of five types of lithology—mudstone, sandstone, conglomerate, granite, and basalt—were randomly divided to obtain 1080 training samples and 471 test samples. Feature extraction and model building were performed on these samples, and a Matlab testing program was designed. The accuracy rate of the method was 94.46%, which is a good result. Figure 7 As shown, the method of the present invention is feasible under this model and the effect is quite ideal.

[0151] Figure 4 , Figure 6 In this three-layer structure diagram, the object layer consists of hue, saturation, brightness, roughness, contrast, orientation, Na+K, Ti, Na, Fe, Si, Na+Ca, and Fe+Mg; the index layer consists of image features and elemental parameters; and the target layer is the identification of rock debris lithology.

[0152] Figure 5 This includes constructing a judgment matrix from the existing hierarchical structure, calculating indicator weights, performing consistency checks, and ranking the hierarchical structure.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A lithology identification method that fuses image and feature information, characterized in that, Includes the following steps: Step 1. Acquiring the original rock cuttings image; Step 2. Preprocessing the original image; Step 3. Rock cuttings image segmentation; Step 4. Feature extraction of rock cuttings samples; Including 1) Extraction of rock debris color features: Since images are generally saved in RGB format, the RGB color space is directly converted to HSV space through linear transformation, thereby drawing HSV color histograms and taking their respective peak values ​​as analysis parameters. 2) Extraction of rock debris texture features: For the extraction of rock debris texture features, a texture model that is easier for visual systems to perceive, as proposed by Tamura, is used, and roughness, contrast, and orientation are used as texture features. Step 5. Combining element parameters; By observing and analyzing the element mass fractions in the HB-X200 national standard element data table, typical elemental characteristics under specific lithologies are obtained, and some combination forms between elements are added to highlight the differences between various lithologies. Step 6. Establish a rock fragment characteristic analysis model using tomographic analysis; Step 7. Normalize the extracted features; Step 8. Establish a lithological standard library; Step 9. Determine the type of rock cuttings sample; include: 1) Assume a specific lithology in the template library Feature vector of the cuttings sample ,in, Given the values ​​for all extracted features, calculate the centroid coordinates of the n group points for this lithology: Based on the feature weights calculated using the analytic hierarchy process in step 6, different features have different weights. Let's assume that rock fragments are among the features... The weights below are Combining the Euclidean distance calculation method, for two different sample points and Calculate the weighted distance: ; 2) Calculate the different lithologies in the template library The centroid of the debris cluster below , Given a rock cuttings sample to be identified Based on the calculation method given in 1), the rock cuttings sample to be identified is calculated. Five different rock properties Weighted distance: Find the smallest value in the set. , Corresponding rock properties core That is, the sample to be tested Lithology ; 3) After determining the lithology of the points to be measured, the centroid of the group of points is re-determined using the centroid calculation formula: .

2. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 1 includes: placing the collected rock fragment sample on a measurement platform, first magnifying it with an optical microscope, then taking a picture of the sample with a camera, and finally storing the image in a computer.

3. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 2 includes: using a median filter with a sliding window of 3*3 to remove noise in the original image. At the same time, using a median filter can effectively preserve the edge information of the original microscopic image, prevent edge blurring, and facilitate subsequent image segmentation operations. The expression for the two-dimensional median filter function is as follows: ; in, For the original image, The processed image, All represent pixel coordinates; After removing noise with a median filter to obtain a smooth image, the Sobel operator is used to sharpen the image. This operator contains two sets of 3×3 matrices, one horizontal and one vertical. By convolving them with the image in a plane, approximate values ​​of brightness difference in the horizontal and vertical directions are obtained, making the image to be processed clearer and highlighting more details. The Sobel operator is as follows: ; in, and These represent images that have undergone horizontal and vertical edge detection, respectively.

4. The lithology identification method based on image and feature information fusion according to claim 3, characterized in that, Step 3 includes: converting the processed image to grayscale and binarizing it to obtain an image, and then calculating the Sobel operator respectively. and Afterwards, by The gray-level gradient of the pixel (x,y) in the microscopic image is obtained. The gray-level amplitude image of the image is stored as a data matrix, where each value of the data matrix is ​​the gray-level value corresponding to the pixel in the gradient image. Based on this, the watershed segmentation method is used to segment the image.

5. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 4 specifically involves: Based on step 3, color and texture features are extracted for each individual rock fragment sample obtained from the segmentation. Since the HSV color model is natural, it is closer to human visual experience and color perception than the RGB color space. Furthermore, experiments have verified that the HSV color space model is used to extract color features, including three color attributes: hue, saturation, and brightness. Since images are generally stored in RGB format, a linear transformation can be used to directly convert the RGB color space to the HSV color space: Where R, G, and B are the coordinates corresponding to the RGB color space. , , In the RGB color model , , The result after normalization; , They are respectively , , The maximum and minimum values ​​in the range. for and The difference, Hue calculates: S calculation: calculate: ; For color tone, For saturation, Brightness; Plot the HSV color histogram and take the peak values ​​of each color as the analysis parameters; For the extraction of rock debris texture features, a texture model that is more easily perceived by the visual system proposed by Tamura is used, employing three pieces of information—roughness, contrast, and orientation—as texture features, where: 1) Roughness reflects the rate of change of image pixel grayscale values. To calculate roughness, the size of the neighborhood must first be selected, i.e. ,in ; The formula for calculating the neighborhood mean of each pixel in an image is: in, To select the value of the power exponent of the domain size and , For pixels The size of the grayscale value, Let x and y be the neighborhood mean of each pixel in the image, and let i and j be the horizontal and vertical coordinates of the pixel. Then, for each pixel, the average grayscale difference between its non-overlapping windows in the horizontal and vertical directions needs to be calculated using the following formula: Horizontal direction: Vertical direction: in, , Representing pixel coordinates The average grayscale difference between non-overlapping windows in the horizontal and vertical directions. and The mean of the neighborhood of the pixel coordinates at that point. , This represents the neighborhood mean of the pixel coordinates at that point; Compare the mean difference values ​​in the horizontal and vertical directions, and select the larger value to represent the mean difference value of the current pixel's neighborhood: in, The mean difference of the neighborhood for pixels with larger mean differences in both the horizontal and vertical directions. , Representing pixel coordinates ( The average grayscale difference between windows that do not overlap in the horizontal and vertical directions; The optimal size determined from among numerous dimensions across various fields; Scan every pixel of the image and select and determine an optimal size from a range of sizes, L. ,satisfy: in, The difference between the mean and the mean of the region. The maximum value among the differences in the neighborhood mean, where L is the number of neighborhood sizes; Roughness is calculated by measuring the roughness of the entire image. It is obtained by averaging the values, and its expression is: Where m and n are both pixel values. The roughness of the image; 2) Contrast describes the brightness of an image; calculating contrast requires first calculating kurtosis. ,in The image grayscale is a fourth-order matrix. Let the variance of the image gray levels be the expression for the contrast ratio: ,in, This refers to the image contrast. 3) Directionality is a global characteristic of a texture region, describing a state of the image, specifically a state where it is straight or approximately straight in a certain direction. First, it is necessary to calculate the magnitude and angle of the gradient vector of each pixel in the image: First, the 3x3 matrix surrounding the pixel is compared with the matrix... sum matrix The convolution is obtained and The formula for calculating the magnitude of the gradient vector is: ,in, and These represent the gradient changes in the horizontal and vertical directions, respectively. The magnitude of the gradient vector of the pixel; Formula for calculating the angle of the gradient vector: in, The angle of the gradient vector of the pixel; After obtaining the magnitude and angle of the gradient vectors of all pixels, a description is established. Histogram of distribution ,statistics The number of pixels greater than a given threshold in the histogram In this context, one can determine whether an image has a clear directionality by observing its peaks and flatness, and calculate the overall directionality of the image from the sharpness of the peaks in the histogram. ,in, The overall orientation of the image. For function The corresponding value for obtaining the maximum value , This represents the peak value label in the histogram. Indicates the number of peaks. The range encompassed by this peak value. for The corresponding value for obtaining a large value , To describe Histogram of distribution.

6. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 6 includes: 1) Defining the types and quantities of target layer, indicator layer and object layer: taking rock debris lithology identification as the analysis target, the indicator layer is set as image features and element features, and the object layer is the concretization of image features and element features, thereby establishing a multi-level progressive structure model for lithology identification, which serves as the model basis for subsequent steps of analysis; 2) Determine the evaluation criteria for the relative importance of factors: Determine the relative importance of each influencing factor according to the meaning of the evaluation scale, and construct judgment matrices at each level based on the relative importance of factors; 3) Calculate element weights using the root-based method: Using the root value method: Let the n×n judgment matrix A be: , This means multiplying the elements of each row in matrix A for each element in the n*n matrix: And find the nth root of the product of all rows: ,in, The product of the elements in each row of matrix A, then... Normalization process: Obtain the weights of each influencing factor. For each influencing factor The weights; 4) By Calculate the consistency index of the matrix, where This is a consistency index, where n is the order of the judgment matrix. Let the largest eigenvalue of the matrix be the random consistency ratio, and then calculate the consistency index of the judgment matrix: ,in, This represents the consistency index of the judgment matrix. The random consistency ratio, The average random consistency index, When n=13, the value is 1.

56. If the matrix is ​​found to be logically correct, then the condition is met; if... If this happens, the judgment matrix needs to be readjusted, and a consistency check needs to be performed again until... ; 5) Hierarchical single ranking and consistency verification: Calculate the comprehensive weight of each element in the indicator layer relative to the overall goal, and perform a consistency verification on the overall hierarchical ranking to reduce inconsistencies in the overall ranking results. Obtain the weights of each influencing factor in the object layer from step 3). Influencing factors and criteria at the object level The consistency index of the single ranking of the relevant factors constituting the judgment matrix is ​​as follows: Its corresponding average random consistency index Then the overall ordering consistency ratio of the scheme layer is: , Represents the various effects of the object layer Factor weights, when If the overall hierarchical ranking is passed, then the consistency index is considered to be valid.

7. The lithology identification method based on image and feature information fusion according to claim 6, characterized in that, The types and quantities of target layer, indicator layer, and object layer are clearly defined. Specifically, a multi-level progressive structure is established. Among them, image features include six features: hue, saturation, brightness, roughness, contrast, and orientation; elemental parameters include seven features: Na+K, Ti, Na, Fe, Si, Na+Ca, and Fe+Mg.

8. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 7 involves applying deviation normalization to perform a linear transformation on the image features and elemental features of the segmented rock fragments. Where max and min represent the maximum and minimum values ​​of the sample data, respectively. The result of the linear transformation is the multidimensional vector of the rock cuttings sample. This represents the current image feature or element feature sample value.

9. The lithology identification method based on image and feature information fusion according to claim 1, characterized in that, Step 8 involves performing the operation of step 4 on rock fragments of known lithology, extracting the image features of the rock fragments under that lithology, adding the element features obtained in step 5, performing data normalization on these features to obtain a multidimensional vector under a specific lithology, and establishing a standard lithology template library.

Images