A Rock Sample Fracture Detection and Recognition Method Based on Image Visual Salience

Abstract

The present invention relates to the technical fields of oil and gas exploration and development and image processing. Specifically, it is a method for detecting and identifying rock sample fractures based on image visual saliency, which includes the following steps: S1, inputting the original core image data; S2, obtaining the fracture research area image from the original core image through image segmentation and manual trimming; S3, using morphological quotient operation processing and the dark channel prior principle to calculate and obtain the initial fracture saliency map of the fracture research area image; S4, performing superpixel segmentation on the fracture research area image, and establishing a cellular automaton update mechanism through color features and initial saliency information to realize the optimization processing of the initial fracture saliency map. The present invention introduces the visual saliency principle and the cellular automaton optimization mechanism, and realizes the rapid detection and extraction of fractures in a large number of fracture rock sample photos.

Description

Technical Field

[0001] The present invention relates to the technical fields of oil and gas exploration and development and image processing, and specifically, to a method for detecting and identifying rock sample fractures based on image visual saliency. Background Art

[0002] Natural fractures are an important structure existing in geology, with important reservoir and migration functions. They are one of the important conditions for the formation and preservation of oil and gas, have a direct impact on the productivity of oil and gas wells, determine the oil drainage capacity of the matrix and the oil supply area of the oil well, and are the focus of the research on oil and gas reservoir development plans. Therefore, in the exploration and development of fractured oil and gas reservoirs, it is necessary to comprehensively and accurately understand the development and distribution law of the fracture system.

[0003] Currently, the existing methods for detecting and identifying natural fractures in rock samples are mainly core observation methods. This method uses directional coring technology to directly observe and measure the whole core of the taken core, or observes it under a microscope through a core thin section to achieve the detection and identification of fractures, and then obtains fracture characteristic parameters. Although the core observation method is the most accurate method for detecting and identifying fractures, it is time-consuming, has a large amount of manual identification work, and extremely low efficiency. At the same time, there are also problems such as the chaotic distribution and irregular shape of core fractures, which exacerbate the difficulty of manual detection and identification of fractures (Li Shaohua et al., 2006; Li Airong and Hu Shuyong, 2021). Other methods for detecting and identifying natural fractures include production dynamic analysis methods (Jiang Ruizhong et al., 2013), in-situ stress analysis methods (Huang Tao et al., 2024), and well logging identification methods (Zhao Junlong et al., 2012), but these methods are mainly for indirectly identifying unknown underground fractures and are not applicable to the detection and identification of fractures in surface rock samples. Considering the problems such as low efficiency of manual identification of surface rock sample fractures and complex fracture distribution, it is urgent to establish an intelligent and rapid detection method for a large number of rock sample fractures by means of artificial intelligence technology.

[0004] Image visual saliency detection is to use a computer to simulate the human visual attention mechanism, calculate the importance of each part of the information in the picture, and then obtain a grayscale image with the same size as the original picture. The pixel value of each point in the grayscale image represents the saliency value of the original image pixel. The brightness is positively correlated with the size of the saliency value. This grayscale image is also called a saliency map. Through the saliency map, the saliency region of human interest can be quickly found (Yuan Ye et al., 2020). As an important carrier for carrying reservoir fracture information, the fracture region in the rock sample photo has unique visual characteristics, such as color, texture, and shape characteristics that are different from the rock background. With the help of visual saliency detection algorithms, it is expected to quickly and accurately highlight the fracture region from the complex core photo background, achieve the efficient detection and identification of rock sample fractures, reduce the manual detection workload of core fractures, significantly improve the fracture detection and identification efficiency, and support the intelligent and rapid detection of a large number of rock samples in the oilfield. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and propose a method for detecting and identifying rock sample fractures based on image visual saliency, introducing the principle of visual saliency and the optimization mechanism of cellular automata, and realizing the detection of fractures in the cores of oil and gas field reservoirs.

[0006] To achieve the above purpose, the specific technical solution adopted by the present invention is as follows:

[0007] A method for detecting and identifying rock sample fractures based on image visual saliency, comprising the following steps:

[0008] S1. Input the original core image data;

[0009] S2. Obtain the fracture research area image from the original core image through image segmentation and manual trimming;

[0010] S3. Use morphological quotient operation processing and the dark channel prior principle to calculate and obtain the initial fracture saliency map of the fracture research area image;

[0011] S3.1. Morphological quotient operation processing

[0012] S3.1-1. Input the fracture research area image ;

[0013] S3.1-2. Perform multi-scale morphological closing operation processing on the image to obtain the background image ;

[0014] S3.1-3. Perform quotient operation processing on the image and the background image to obtain the quotient image ;

[0015] S3.1-4. Obtain the fracture saliency map of the quotient image through image inversion and normalization processing

[0016] S3.2. Dark channel prior calculation

[0017] S3.2-1. Construct a new atmospheric scattering model;

[0018] S3.2-2. For fracture images with uneven illumination and low illumination, generate the fracture saliency map of the dark channel prior based on the new atmospheric scattering model ;

[0019] S3.2-3. Fuse the fracture saliency map of the quotient image and the fracture saliency map of the dark channel prior to obtain the initial fracture saliency map 。

[0020] S4. Perform superpixel segmentation on the image of the crack research area, establish a cellular automaton update mechanism through color features and initial saliency information, and achieve the optimization of the initial crack saliency map;

[0021] S41. Segment the crack research area using the superpixel segmentation algorithm LSC , and then establish an influence factor matrix by using the Lab color space features of the image, the saliency map of the dark channel prior, and the saliency map of the quotient image ;

[0022] S42. Perform cross-diffusion processing on the three influence factor matrices , and to obtain a fused influence factor matrix;

[0023] S43. Perform row normalization on the influence factor matrix to obtain a normalized influence factor matrix ;

[0024] S44. Establish a confidence matrix and constrain the confidence matrix within a certain range;

[0025] S45. Based on the normalized influence factor matrix and the confidence matrix establish a synchronous update and optimization mechanism 。

[0026] In the method for detecting and identifying rock sample cracks based on image visual saliency according to the present invention, step S3.1 specifically includes the following steps:

[0027] S3.1-1: Input the image of the crack research area ;

[0028] S3.1-2: Perform multi-scale morphological closing operation on the image to obtain the background image ;

[0029] ;

[0030] wherein, represents the morphological closing operation, represents the structural element of the closing operation, represents the set of structural elements.

[0031] S3.1-3: Perform quotient operation on the image and the background image to obtain the quotient image ;

[0032] ;

[0033] In the formula, represents the quotient image the i row and j the column pixel value, represents the crack research area image the i row and j the column pixel value, represents the background image the i row and j the column pixel value.

[0034] S3.1 - 4: Obtain the crack saliency map of the quotient image ;

[0035] ;

[0036] ;

[0037] In the formula, represents the pixel value of the i row and j the column of the inverted image, represents the quotient image the i row and j the column pixel value, represents the crack saliency map the i row and j the column pixel value, represents the maximum pixel value in the inverted image, represents the minimum pixel value in the inverted image.

[0038] In the method for detecting and identifying rock sample cracks based on image visual saliency described in the present invention, in step S3.2 - 1, a new atmospheric scattering model is constructed:

[0039] ;

[0040] In the formula, represents image imaging, represents the clear image under ideal sunny weather, represents the transmittance function value, represents the atmospheric light value, represents the pixel.

[0041] In the method for detecting and identifying rock sample fractures based on image visual saliency according to the present invention, step S3.2-2 includes the following steps:

[0042] Obtain the image of the fracture research area For the R, G, and B channels, based on the new atmospheric scattering model, we have:

[0043] ;

[0044] In the formula, represents the image formation of the three channels, represents the atmospheric light value of the three channels, represents the R, G, and B channels of the image of the fracture research area ; represents the clear image of the three channels under ideal clear weather, represents the transmittance function value, represents the pixel.

[0045] Assume that the transmittance function is a fixed value and the atmospheric light value is a constant value. Take the minimum value of both sides of Equation (6), then:

[0046] ;

[0047] In the formula, represents the local area in the image, represents the element of the local area ; represents the image formation and the atmospheric light value ratio, represents the clear image and the atmospheric light value ratio, represents the transmittance function value, represents the pixel.

[0048] The dark channel prior theory believes that the dark channel is composed of the points with the minimum pixel values in the 3 color channels and can be expressed as:

[0049] ;

[0050] In the formula, represents the dark channel of the pixel ; represents the clear image of the three channels under ideal clear weather, represents the local area in the image.

[0051] Substitute Equation (8) into Equation (7) to obtain the transmittance function ; Meanwhile, by introducing a defogging degree adjustment coefficient , the estimated transmittance function is obtained:

[0052] ;

[0053] Perform the following normalization processing on Equation (9) to generate the crack saliency map of the dark channel prior ,

[0054] ;

[0055] In the formula, represents the pixel value of the crack saliency map , represents the estimated transmittance function.

[0056] In a method for detecting and identifying rock sample cracks based on image visual saliency according to the present invention, in step S3.2-3, by fusing the crack saliency map of the quotient image and the crack saliency map of the dark channel prior, the initial crack saliency map is obtained:

[0057] ;

[0058] In a method for detecting and identifying rock sample cracks based on image visual saliency according to the present invention, step S4 includes the following steps:

[0059] S41. Use the superpixel segmentation algorithm LSC to segment the crack research area , and then establish an influence factor matrix by using the Lab color space features of the image, the saliency map of the dark channel prior, and the saliency map of the quotient image respectively:

[0060] ;

[0061] In the formula, represents the -th class feature value of the superpixel represents the -th class color feature value of the superpixel represents the neighborhood set of the superpixel ; Thus, the influence factor matrix of the Lab color space features, the influence factor matrix of the dark channel prior saliency map, and the influence factor matrix of the quotient image saliency map can be obtained.

[0062] S42. Cross-diffusion processing is performed on the three impact factor matrices , and to obtain a fused impact factor matrix:

[0063] ;

[0064] ;

[0065] wherein, represents the number of iterations of the cross-diffusion processing, , and represent the row normalization processing of the three impact factor matrices. The fused impact factor matrix is expressed as follows:

[0066] ;

[0067] S43. The impact factor matrix is subjected to row standardization processing to obtain a standardized impact factor matrix :

[0068] ;

[0069] wherein, , , represents the element of the impact factor matrix F.

[0070] S44. A confidence matrix is established, and its element is:

[0071] ;

[0072] wherein, represents the element of the impact factor matrix F.

[0073] The confidence matrix is constrained within a certain range, that is, , and its element is:

[0074] ;

[0075] wherein, , , represents the element of the confidence matrix .

[0076] S45. Based on the standardized impact factor matrix and the confidence matrix Establish a synchronous update and optimization mechanism , that is:

[0077] ;

[0078] In the formula, represents the crack saliency map after the -th update and optimization. When , n ; I represents the

[0079] The present invention has the following characteristics and beneficial effects:

[0080] 1) In the regional image, the morphological quotient operation processing and the dark channel prior principle are respectively used to calculate the initial crack saliency map of the image, which can not only overcome the uneven illumination, but also effectively extract the reservoir core cracks in large quantities.

[0081] 2) By performing superpixel segmentation on the original image, and then calculating the influence factor matrix through color features and initial saliency information, a cellular automaton synchronous update mechanism is established to optimize the initial crack saliency map, realizing the rapid detection and extraction of cracks in a large number of crack rock sample photos, laying a solid foundation for the subsequent efficient and intelligent identification of crack parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is an experimental result diagram of a method for detecting and identifying rock sample cracks based on image visual saliency according to the present invention;

[0083] In the figure: (a) Core photo (with scale); (b) Extracted crack image; (c) Initial crack saliency map; (d) Optimized crack saliency map.

[0084] Figure 2 is the original core crack image in the embodiment of the present invention.

[0085] Figure 3 is the crack research area image in the embodiment of the present invention.

[0086] Figure 4 is the initial crack saliency map in the embodiment of the present invention.

[0087] Figure 5 is the optimized initial saliency map of the core crack in the embodiment of the present invention.

[0088] Figure 6 is the flow chart of performing multi-scale morphological closing operation processing on the crack research area image in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0089] The present invention will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0090] Aiming at the detection of tiny fractures at the core scale of oil and gas reservoir, the present invention introduces the principle of visual saliency and the optimization mechanism of cellular automata, and provides a method for detecting and identifying fractures in rock samples based on image visual saliency. First, for the current problems of the chaotic distribution of original fracture images and the large workload of manual identification in the fracture detection of a large number of reservoir core photos, the method uses the visual saliency detection theory to quickly and intelligently detect reservoir core fractures. During operation, according to the physical core photo with a scale ( Figure 1 Figure (a)), an image of the fracture research area with specific length and width is intercepted ( Figure 1 Figure (b)). In the area image, the morphological quotient operation and the dark channel prior principle are respectively used to calculate the initial fracture saliency map of the image ( Figure 1 Figure (c)). The results are shown in Figure 1 Figure (c), indicating that the morphological quotient operation and the dark channel prior can not only overcome uneven illumination, but also effectively extract core fractures in large quantities. Then, aiming at the problems of incomplete and discontinuous detected fractures in the initial fracture saliency map, the original image is subjected to superpixel segmentation, and the influence factor matrix is calculated through color features and initial saliency information, so as to establish a cellular automata synchronous update mechanism to optimize the initial fracture saliency map, and thus obtain the final high-quality fracture saliency map ( Figure 1 Figure (d)). The experimental results are shown in Figure 1 Figure (d), indicating that the cellular automata can optimize the initial fracture saliency map and obtain more complete core fractures. This method realizes the rapid detection and extraction of fractures in a large number of fractured rock sample photos, laying a solid foundation for the subsequent efficient and intelligent identification of fracture parameters.

[0091] The following further describes the specific processing process of the present invention in conjunction with the drawings and embodiments:

[0092] S1. Input the original core image data, as shown in Figure 2 Figure;

[0093] S2. Obtain the fracture research area image from the original core image through image segmentation and manual trimming. The fracture research area image in this embodiment is as shown in Figure 3 Figure;

[0094] S3. Use the morphological quotient operation and the dark channel prior principle to calculate and obtain the fracture research area image ( Figure 3Initial crack saliency map of ( Figure 4 ), specifically, including the following steps:

[0095] S3.1 Morphological quotient operation processing

[0096] Step 1: Input the image of the crack research area ( Figure 3 );

[0097] Step 2: As Figure 6 shown, perform multi-scale morphological closing operation on the image to obtain the background image ;

[0098] ;

[0099] In the formula, in the formula, represents the morphological closing operation processing, represents the structuring element of the closing operation, represents the set of structuring elements, where:

[0100] ;

[0101] Step 3: Perform quotient operation on the image and the background image to obtain the quotient image ;

[0102] ;

[0103] In the formula, represents the pixel value of the quotient image at the i th row and j th column, represents the pixel value of the crack research area image at the i th row and j th column, represents the pixel value of the background image at the i th row and j th column.

[0104] Step 4: Obtain the crack saliency map of the quotient image through image inversion and normalization processing ;

[0105] ;

[0106] ;

[0107] In the formula, Indicates the pixel value of the i row and the j column of the inverted image, indicates the pixel value of the row and the i column of the quotient image, j indicates the pixel value of the crack saliency map, row and the i column of the j crack saliency map, indicates the maximum pixel value in the inverted image, indicates the minimum pixel value in the inverted image.

[0108] S3.2 Dark Channel Prior Calculation

[0109] Step 1: For the crack images of uneven illumination and low illumination images, use the dark channel prior to calculate the saliency map of the crack images and extract the cracks. For low illumination images, construct a new atmospheric scattering model:

[0110] ;

[0111] In the formula, represents image imaging, represents the clear image under ideal clear weather, represents the transmittance function value, represents the atmospheric light value, represents the pixel.

[0112] Step 2: Obtain the R, G, and B channels of the crack research area image . Based on the new atmospheric scattering model, there are:

[0113] ;

[0114] In the formula, represents the image imaging of the three channels, represents the atmospheric light value of the three channels, represents the R, G, and B channels of the crack research area image , represents the clear images of the three channels under ideal clear weather, represents the transmittance function value, represents the pixel.

[0115] Assume that the transmittance function is a fixed value and the atmospheric light value is a constant . Take the minimum value of both sides of Equation (6), then:

[0116] ;

[0117] In the formula, represents a local area in the image, represents the local area element, represents the image imaging and the atmospheric light value ratio, represents the clear image and the atmospheric light value ratio, represents the transmittance function value, represents the pixel. The dark channel prior theory believes that the dark channel is composed of the points with the smallest pixel values in the three color channels and can be expressed as:

[0118] ;

[0119] In the formula, represents the pixel dark channel, represents the clear images of the three channels under ideal clear weather, represents a local area in the image.

[0120] Substitute Equation (8) into Equation (7) to obtain the transmittance function ; At the same time, by introducing a parameter , called the defogging degree adjustment coefficient, with a range of , usually taking a value of 0.95, to obtain the estimated transmittance function :

[0121] ;

[0122] Perform the following normalization processing on Equation (9) to generate the crack saliency map of the dark channel prior ,

[0123] ;

[0124] In the formula, represents the pixel value of the crack saliency map , represents the estimated transmittance function.

[0125] Step 3: By fusing the crack saliency map of the quotient image and the crack saliency map of the dark channel prior, obtain the initial crack saliency map (as shown in Figure 4 ):

[0126] ;

[0127] S4. Perform superpixel segmentation on the crack research area image, establish a cellular automaton update mechanism through color features and initial saliency information, and realize the optimization of the initial crack saliency map ( Figure 5 ). Specifically, in order to optimize the initial crack saliency map ( Figure 4 ), a cellular automaton synchronous update mechanism will be adopted to optimize the initial saliency map, so as to obtain the final crack saliency map. First, use the superpixel segmentation algorithm LSC to segment the crack research area , and then establish a -dimensional influence factor matrix by using the Lab color space features of the image, the saliency map of the dark channel prior, and the saliency map of the quotient image , where represents the similarity value between each superpixel block and the adjacent superpixel block under the -type eigenvalue, and its calculation method is as follows:

[0128] ;

[0129] In the formula, represents the -type eigenvalue of the superpixel , represents the -type color feature value of the superpixel , and represents the neighborhood set of the superpixel .

[0130] Furthermore, the similarity value of the Lab color space feature and the -dimensional influence factor matrix , the similarity value of the dark channel prior saliency map and the -dimensional influence factor matrix , and the similarity value of the quotient image saliency map and the -dimensional influence factor matrix can be obtained.

[0131] Perform cross-diffusion processing on the three influence factor matrices , , and to obtain the fused influence factor matrix:

[0132] ;

[0133] ;

[0134] In the formula, Indicates the number of iterations of cross-diffusion processing, Indicates the transpose of a matrix, 、 and Indicate The similarity influence factor matrix of the Lab color space features, the similarity influence factor matrix of the dark channel prior saliency map, and the similarity influence factor matrix of the quotient image saliency map after 、 and Indicate The similarity influence factor matrix of the Lab color space features, the similarity influence factor matrix of the dark channel prior saliency map, and the similarity influence factor matrix of the quotient image saliency map after When 、 and . Indicates the row normalization processing of the influence factor matrix of the Lab color space features, Indicates the row normalization processing of the influence factor matrix of the dark channel prior saliency map; Indicates the row normalization processing of the influence factor matrix of the quotient image saliency map; Indicates the -dimensional influence factor matrix of the Lab color space features after normalization processing, Indicates the -dimensional influence factor matrix of the dark channel prior saliency map after normalization processing, Indicates the -dimensional influence factor matrix of the quotient image saliency map after normalization processing. The fusion influence factor matrix Is expressed as follows:

[0135] ;

[0136] Next, perform row standardization processing on the influence factor matrix to obtain the standardized influence factor matrix :

[0137] ;

[0138] In the formula, , , Indicates the element of the influence factor matrix F.

[0139] In the cellular automaton synchronous update mechanism, the next evolution state of each cell is jointly determined by its current state and the set of adjacent cells. Here, a confidence matrix is established to ensure the effectiveness and stability of the update evolution. The confidence matrix is expressed as , and its elements is:

[0140] ;

[0141] In the formula, represents the element of the influence factor matrix F.

[0142] Based on Equation (17), to ensure that the saliency value of each superpixel can be updated to a relatively stable and accurate state, the confidence matrix is constrained within a certain range, that is , where the element is:

[0143] ;

[0144] In the formula, , , represents the confidence matrix element of.

[0145] Based on the standardized influence factor matrix and the confidence matrix to establish a synchronous update and optimization mechanism , that is:

[0146] ;

[0147] In the formula, represents the crack saliency map after the th update and optimization, where when , n ; I represents the

[0148] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions in the specification are only preferred examples of the present invention and do not limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting and identifying rock sample fractures based on image visual saliency, characterized in that: It includes the following steps: S1. Input the original core image data; S2. Obtain the fracture research area image from the original core image through image segmentation and manual trimming; S3. Use morphological quotient operation processing and the dark channel prior principle to calculate and obtain the initial fracture saliency map of the fracture research area image; S4. Perform superpixel segmentation on the fracture research area image, establish a cellular automaton update mechanism through color features and initial saliency information, and realize the optimization processing of the initial fracture saliency map; The step S3 includes the following steps: S3.

1. Morphological quotient operation processing S3.1-1: Input the image of the crack research area ; S3.1-2: Perform multi-scale morphological closing operation on the image to obtain the background image ; S3.1-3: Divide the image by the background image to perform a quotient operation and obtain a quotient image ; S3.1-4: Obtain the quotient image through image inversion and normalization processing ; S3.

2. Dark channel prior calculation S3.2-1: Construct a new atmospheric scattering model; S3.2-2: For the crack images with uneven illumination and low illuminance, generate a crack saliency map of dark channel prior based on the new atmospheric scattering model ; S3.2-3: Fusion of commercial images Crack saliency map and crack saliency map of dark channel prior , to obtain the initial crack saliency map.

2. The method for detecting and identifying rock sample cracks based on image visual saliency according to claim 1, wherein: The specific steps of the step S3.1 include the following steps: S3.1-1: Input the image of the crack research area ; S3.1-2: Perform multi-scale morphological closing operation on the image to obtain the background image ; （1）； In the formula, represents morphological closing operation processing, represents the structuring element of the closing operation, represents the set of structuring elements; S3.1-3: Divide the image by the background image to obtain a quotient image ; ； In the formula, represents the quotient image the i pixel value of the j row and the column; represents the image of the crack research area i the j pixel value of the row and the column; i represents the background image j the pixel value of the row and the column. S3.1-4: Obtain the quotient image through image inversion and normalization processing of the crack significance map ; ； ； In the formula, Indicates the reverse image i Row, No. j The pixel value of the column, Representation quotient image No. i Row, No. j The pixel value of the column, Crack Significance Map No. i Row, No. j The pixel value of the column, Represents the maximum pixel value in the inverted image, Indicates the minimum pixel value in the inverted image.

3. A method for detecting and identifying rock sample fractures based on image visual saliency as described in claim 1, characterized in that: In the step S3.2-1, construct a new atmospheric scattering model: ； In the formula, represents image imaging, represents a clear image under ideal clear weather, represents the value of the transmittance function, represents the atmospheric light value, represents a pixel.

4. A method for detecting and identifying rock sample fractures based on image visual saliency as claimed in claim 1, characterized in that: The step S3.2-2 includes the following steps: Obtain the image of the fracture research area For the R, G, and B channels, based on the new atmospheric scattering model, there are: ； In the formula, represents the image imaging of three channels, represents the atmospheric light values of three channels, represents the image of the crack research area for the R, G, and B channels, represents the clear images of three channels under ideal clear weather, represents the transmittance function value, represents a pixel; Assume the transmittance function is a fixed value and the atmospheric light value is a constant. Taking the minimum value on both sides of Equation (6), we get: ； In the formula, represents a local area in the image, represents the local area of the elements, represents the image formation and the atmospheric light value ratio, represents the clear image and the atmospheric light value ratio, represents the transmittance function value, represents a pixel; The dark channel prior theory believes that the dark channel is composed of the points with the minimum pixel values in the three color channels and can be expressed as: ； In the formula, represents the dark channel of a pixel , represents the clear images of three channels under ideal clear weather, represents a local area in the image; Substituting Equation (8) into Equation (7) gives the transmittance function ; meanwhile, by introducing the defogging degree adjustment coefficient , the estimated transmittance function is obtained as follows: ； Perform the following normalization on Equation (9) to generate the crack saliency map of the dark channel prior , ； In the formula, represents the pixel value of the crack significance map , represents the estimated transmittance function.

5. A method for detecting and identifying rock sample fractures based on image visual saliency as described in claim 1, characterized in that: In the step S3.2-3, by fusing the crack saliency map of the quotient image and the crack saliency map of the dark channel prior , an initial crack saliency map is obtained as follows: 。 6. The method for detecting and identifying rock sample fractures based on image visual saliency according to claim 1, wherein: The step S4 includes the following steps: S41. Use the superpixel segmentation algorithm LSC to segment the crack research area , and then establish an influence factor matrix by using the Lab color space features of the image, the saliency map of the dark channel prior, and the quotient image saliency map respectively : ； Wherein, represents each superpixel block and the adjacent superpixel blocks at the similarity value under the class eigenvalue, represents the class eigenvalue of the superpixel ; represents the class color eigenvalue of the superpixel ; represents the neighborhood set of the superpixel ; Thus, the similarity value of the Lab color space feature and the influence factor matrix of dimension , the similarity value of the dark channel prior saliency map and the influence factor matrix of dimension , and the similarity value of the quotient image saliency map and the influence factor matrix of dimension can be obtained; S42. Cross-diffusion processing is performed on the three impact factor matrices , and to obtain a fused impact factor matrix: ； ； Wherein, represents the number of iterations of cross-diffusion processing, , and represent the row normalization processing of three influence factor matrices; represents the transpose of a matrix, , and represent the similarity influence factor matrices of the Lab color space features, the similarity influence factor matrices of the dark channel prior saliency maps, and the similarity influence factor matrices of the quotient image saliency maps after times of cross-diffusion processing, , and represent the similarity influence factor matrices of the Lab color space features, the similarity influence factor matrices of the dark channel prior saliency maps, and the similarity influence factor matrices of the quotient image saliency maps after times of cross-diffusion processing; represents the influence factor matrix of the -dimensional Lab color space features after normalization processing, represents the influence factor matrix of the -dimensional dark channel prior saliency maps after normalization processing, represents the influence factor matrix of the -dimensional quotient image saliency maps after normalization processing; Fusion impact factor matrix It is expressed as follows: ； S43. Perform row normalization on the impact factor matrix to obtain the normalized impact factor matrix : ； In the formula, , , represent the elements of the influence factor matrix F; S44. Establish a confidence matrix , whose elements are as follows: ； In the formula, represents the element of the influence factor matrix F; The confidence matrix is constrained within the following range, i.e., , where the element is: ； In the formula, , , represents the elements of the confidence matrix ; S45. Establish a synchronous update and optimization mechanism based on the standardized impact factor matrix and the new confidence matrix , that is: , namely: ； In the formula, represents the crack significance map after the -th update and optimization, where when ; I represents n the identity matrix of order

Images