Automatic outdoor rock mass exploration method and system based on artificial intelligence large model

Abstract

The invention belongs to the technical field of machine learning, and discloses an open-air rock mass automatic exploration method and system based on an artificial intelligence large model. Processing a feature image of the source image, the selected reference image and camera parameters used for acquiring image data of the target open slope rock mass based on an MVS algorithm, and constructing a three-dimensional view model; after a three-dimensional view is converted into point cloud data, clustering segmentation and plane fitting are carried out on the point cloud data based on a CFSFDP algorithm, an RANSAC algorithm and normal vectors of all points, the structural surface can be accurately recognized, a rock mass volume joint number Jv is calculated, a rock mass basic quality index BQ is obtained by combining a rock hardness degree index Rc, and the rock mass basic quality index BQ is corrected. And a scientific basis is provided for a rock mass quality exploration report. According to the method, the accuracy and efficiency of rock mass quality evaluation are improved, and powerful support is provided for design and construction of rock mass engineering.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning and relates to an automated exploration method and system for open-pit rock masses based on artificial intelligence large models. Background Art

[0002] In the engineering geological investigation of slope rock masses, the accurate evaluation of rock mass quality is the key to ensuring the safety and stability of engineering. However, there are many deficiencies in the existing rock mass quality evaluation system, which seriously restricts the design and construction efficiency of slope engineering. Traditional rock mass quality evaluation mainly relies on indoor tests of rock properties and manual measurements of the development of structural fissures. This method is not only time-consuming and laborious, but also has limited accuracy. Indoor tests of rock properties require collecting a large number of samples and conducting complex test analyses, which are not only costly but also difficult to comprehensively reflect the actual situation of slope rock masses. At the same time, manual measurements of the development of structural fissures rely on the experience and judgment of engineers and are easily affected by subjective factors, resulting in inaccurate evaluation results. At the same time, with the expansion of project scale and the increase in complexity, traditional evaluation methods can no longer meet the needs of modern slope engineering. Especially in the fields of open-pit mines, highways, railways, and water conservancy projects, the stability of slopes is directly related to the safety and normal operation of the project. Therefore, a more accurate and efficient rock mass quality evaluation method is needed to cope with the complex and changeable slope engineering environment. Summary of the Invention

[0003] The purpose of the present invention is to solve the problems of time-consuming, laborious, limited accuracy, difficult to comprehensively reflect the actual situation, large subjective factor influence, and inability to meet the needs of modern engineering in the existing rock mass quality evaluation method, and to provide an automated exploration method and system for open-pit rock masses based on artificial intelligence large models.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] An automated exploration method for open-pit rock masses based on artificial intelligence large models includes:

[0006] Extract features from the collected source image data to obtain the feature image of the source image;

[0007] Based on the MVS algorithm, process the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass to construct a three-dimensional view model;

[0008] Convert the constructed three-dimensional view into three-dimensional point cloud data, and calculate the normal vectors of each point in the regional outcrop point cloud set;

[0009] Based on the CFSFDP algorithm and the RANSAC algorithm, the normal vectors of each point are used to perform clustering segmentation and plane fitting on the point cloud data, and the best-fitting plane obtained is the structural plane;

[0010] Based on geostatistics, the structural planes are distinguished to obtain the structural planes of the same group and the structural planes of different groups;

[0011] Based on the rock mass structural characteristics of the structural planes of the same group and different groups of the rock mass, and calculate the rock mass volumetric joint number Jv;

[0012] Through the rock hardness index Rc and the rock mass volumetric joint number Jv, obtain the basic rock mass quality index BQ. Considering the influence of groundwater K1, the attitude of the structural plane K2 and the initial stress state K3 of the rock mass on the basic rock mass quality index BQ, correct the basic rock mass quality index BQ to obtain the rock mass quality exploration report; the rock hardness index Rc is obtained through rock physical and mechanical tests.

[0013] A further improvement of the present invention lies in:

[0014] Furthermore, based on the MVS algorithm, process the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass to construct a three-dimensional view model. Specifically: based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass, construct a cost volume; perform regularization processing on the constructed cost volume, and based on the regularized cost volume, obtain the depth estimation value of each pixel of the collected image, and construct a three-dimensional view model.

[0015] Furthermore, based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass, construct a cost volume, specifically:

[0016] Use the pre-set reference image as the reference image, and extract the feature map of the reference image through a convolutional neural network;

[0017] For each depth plane, calculate the differentiable homography transformation matrix based on the camera parameters and the depth value, and transform the feature map of the source image to the perspective of the reference image;

[0018] Calculate the similarity between the feature map of the transformed reference image and the feature map of the source image based on the Euclidean distance;

[0019] Stack the similarity matrices under each depth plane to form a multi-dimensional cost volume.

[0020] Furthermore, for each depth plane, a differentiable homography transformation matrix is calculated based on the camera parameters and depth values, and the feature map of the source image is transformed to the perspective of the reference image, specifically as follows:

[0021] The preset scene depth range is from the minimum depth z min to the maximum depth z max , based on the internal focal length (f x , f y ) of the camera, the optical center (c x , c y ), the rotation matrix R and the translation vector t, for each pixel coordinate (x src , y src ) in the source image, according to the camera internal parameters and the depth value z selected within the preset scene depth range, the corresponding point coordinates (X, Y, Z) of this pixel in the three-dimensional space are calculated, specifically as follows:

[0022] Z = z

[0023] where the depth value z is any value within the preset scene depth range; z min ≤ z ≤ z max

[0024] Based on the rotation matrix R and the translation vector t of the camera, the three-dimensional point (X, Y, Z) is transformed to the three-dimensional space coordinates (X', Y', Z') in the perspective of the reference image, specifically as follows:

[0025]

[0026] where, is the homogeneous transformation matrix;

[0027] The transformed three-dimensional space coordinates (X', Y', Z') are perspective-projected onto the reference image plane to obtain the corresponding pixel coordinates (x', y'), specifically as follows:

[0028]

[0029] Estimate a global homography transformation matrix based on an optimization method, and this matrix approximately describes the transformation relationship from the source image to the reference image;

[0030] The depth plane is a series of parallel planes in the three-dimensional space that maintain a fixed distance from the camera optical center.

[0031] Further, regularize the constructed cost volume, and based on the regularized cost volume, obtain the depth estimation value of each pixel of the captured image, and construct a three-dimensional view model, specifically: smooth and denoise the cost volume based on a 3D convolutional neural network to obtain the regularized cost volume, apply the Softmax function to the regularized cost volume to normalize the probability value of each depth to between 0 and 1; calculate the expected value of the probability distribution of each depth to obtain the depth estimation value of each pixel point;

[0032]

[0033] where P(d k |(i,j)) represents the probability that the depth is d at the pixel (i,j); cost(d k ,(i,j)) represents the cost of the pixel (i,j) on the depth plane d k ; k' is the index variable in the summation, traversing all depth planes; d k represents the depth value of the k-th depth plane; k

[0034] The above-mentioned calculation of the expected value of the probability distribution of each depth to obtain the depth estimation value of each pixel point is specifically:

[0035] Estimated Depth(i,j) = ∑ k P(d k |(i,j)) * d k .

[0036] Further, convert the constructed three-dimensional view into three-dimensional point cloud data, and calculate the normal vectors of each point in the regional outcrop point cloud set, specifically:

[0037] For each pixel (x, y, d) in the depth map, where d is the depth value, calculate its coordinates (X, Y, Z) in three-dimensional space, specifically:

[0038] Z = d

[0039] Take the calculated three-dimensional coordinates (X, Y, Z) as the points in the point cloud data;

[0040] where f x , f y are the internal reference focal lengths of the camera; c x , c y are the optical center coordinates of the camera;

[0041] ​The normal vectors of each point in the outcrop point cloud set of the calculation area are specifically as follows: For each point in the point cloud, a neighborhood is selected based on K-nearest neighbors; a plane is fitted using the neighborhood points, and the normal vector of the plane is the normal vector of the point.

[0042] Furthermore, based on the CFSFDP algorithm and the RANSAC algorithm, the normal vectors of each point are used to perform clustering segmentation and plane fitting on the point cloud data, and the best-fitted plane obtained is the structural plane, specifically as follows:

[0043] Based on the normal vectors of each point, the angular difference between point clouds is calculated; based on the CFSFDP algorithm, the normal vectors are clustered to identify density peak points, and the density peak points are regarded as clustering centers. The sample points that are not clustering centers are sequentially assigned to the clusters of the nearest neighbors with higher local density according to the local density size;

[0044] The clustering results are projected onto a stereographic projection diagram. According to the projection results, the number and average direction values of the clustering centers are statistically analyzed to preliminarily identify the main structural plane set;

[0045] The points not assigned to the main structural plane set are regarded as noise points, and iterative calculations are performed based on the inherent characteristics of the attitude and the Fisher K value to filter out noise points; thus, the dominant attitude clustering and the corresponding point cloud are obtained; finally, through statistics in the stereographic projection diagram, the main direction of the point cloud set is the dominant attitude of this group of structural planes; the inherent characteristics of the attitude are the direction distribution characteristics of the structural planes in the geological body; after the regional search of the structural planes, the point sets of each structural plane are obtained, and then the random sample consensus RANSAC algorithm is used for rock mass structural plane fitting and transformation; when using the RANSAC algorithm, effective initial data must be selected first, and on this basis, the data set is continuously iteratively increased, and finally the final fitting result with the most data is selected; the least squares method is used to perform plane fitting on each structural plane point cloud set to obtain the best-fitted plane; according to the normal vector and position information of the fitted plane, the position and attitude of the structural plane are determined.

[0046] The structural planes are distinguished based on geostatistics to obtain structural planes of the same group and those of different groups, specifically as follows: Through geostatistical attitude statistics, the structural planes with similar dip angles and plunge angles are regarded as structural planes of the same group; the structural planes with large differences in dip angles and plunge angles are regarded as structural planes of different groups.

[0047] Furthermore, the RANSAC algorithm is specifically as follows:

[0048] Initial point cloud data is selected from the three-dimensional point cloud dataset, and its corresponding plane equation is calculated:

[0049] Ax + By + Cz + D = 0

[0050] Among them, A, B, C, and D are plane equation parameters;

[0051] Calculate the algebraic distance di from all points to this plane, select a threshold d. If di < dj, it is a valid point; otherwise, this point is not included in the model.

[0052] Repeat the above steps, calculate the iteration end parameter according to the data set, and judge whether the iteration stops according to this iteration end parameter.

[0053] Furthermore, the rock mass structure characteristics of the rock masses with the same set of structural planes and those with different sets of structural planes are specifically as follows:

[0054] Based on the rock masses with the same set of structural planes, obtain the distribution pattern of the rock masses in three-dimensional space, the extension length and connectivity of the structural planes of the same set within the rock mass, and then evaluate the potential impact of the structural planes of the same set on the stability of the rock mass under the action of groundwater, temperature change, and seismic load;

[0055] Based on the rock masses with different sets of structural planes, analyze the intersection angle, intersection form, and mechanical behavior at the intersection of the structural planes of different sets; evaluate the degree of damage to the integrity of the rock mass by the structural planes of different sets, including the size and shape of the rock mass cut into blocks and the connection state between the blocks, and then evaluate the potential threat of the structural planes of different sets to the stability of the rock mass under specific engineering conditions;

[0056] The calculated rock mass volume joint number Jv is specifically as follows:

[0057] For the structural planes of the same set, select a representative measuring line in the rock mass, measure the spacing of the structural planes on the measuring line, and add up the reciprocals of the spacings of each group of joints based on the spacing method to obtain the rock mass volume joint number Jv;

[0058] For the structural planes of different sets, calculate the contribution of each group of structural planes to the rock mass volume joint number Jv respectively, and accumulate them;

[0059] The contribution of each group of structural planes to the rock mass volume joint number Jv is calculated by the following formula:

[0060]

[0061] Among them, n is the number of groups of structural planes in the statistical area; S i is the number of structural planes per unit normal length of the i-th group of structural planes; S0 is the number of non-grouped structural planes per unit volume of the rock mass;

[0062] The rock mass basic quality index BQ is obtained through the rock hardness index Rc and the rock mass volume joint number Jv, and then the rock mass quality exploration report is obtained, specifically as follows:

[0063] BQ = 90 + 3Rc + 250Kv

[0064] Among them, Kv is the rock mass integrity coefficient; the relationship between the rock mass integrity coefficient Kv and the rock mass joint number per unit volume Jv is obtained according to the "Engineering Rock Mass Classification Standard".

[0065] After obtaining the basic quality index of the rock mass as BQ, it is also necessary to consider the corrections of groundwater K1, structural plane occurrence K2, and initial stress state K3 to BQ. The corrected index [BQ] = BQ - 100(K1 + K2 + K3), which is used for the final rock mass quality classification.

[0066] The rock mass quality exploration report includes the basic quality level of the rock mass; the basic quality level of the rock mass includes the rock hardness and the rock mass integrity, and the rock hardness and the rock mass integrity are determined through qualitative classification and quantitative indexes; the qualitative classification includes: rock weathering degree classification, rock hardness qualitative classification, structural plane bonding degree classification, rock mass integrity qualitative classification, and rock mass basic quality qualitative characteristics; the quantitative indexes include: rock hardness index Rc, rock mass integrity index, and rock mass basic quality index BQ.

[0067] Among them, the rock weathering degree classification, rock hardness qualitative classification, and structural plane bonding degree classification are determined on-site by professionals; the rock hardness index Rc is obtained through rock physical and mechanical tests; the other items are analyzed by software according to the selected judgment criteria, and finally the rock mass quality evaluation conclusion is given.

[0068] An open-pit rock mass automatic exploration system based on an artificial intelligence large model includes:

[0069] A feature extraction module, which extracts features from the collected source image data to obtain the feature image of the source image.

[0070] A construction module, which processes the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass based on the MVS algorithm to construct a three-dimensional view model.

[0071] A calculation module, which converts the constructed three-dimensional view into three-dimensional point cloud data and calculates the normal vectors of each point in the regional outcrop point cloud set.

[0072] A first acquisition module, which performs clustering segmentation and plane fitting on the point cloud data based on the CFSFDP algorithm and the RANSAC algorithm using the normal vectors of each point, and the obtained best-fitting plane is the structural plane.

[0073] A discrimination module, which discriminates the structural plane based on geostatistics to obtain the same-group structural planes and non-same-group structural planes.

[0074] The second acquisition module, which calculates the rock mass joint number Jv based on the rock mass structural characteristics of the same group of structural planes and non - same - group structural planes of the rock mass.

[0075] The third acquisition module, which obtains the basic rock mass quality index BQ through the rock hardness index Rc and the rock mass joint number Jv, and considers the influence of groundwater K1, the attitude of structural planes K2 and the initial stress state K3 of the rock mass on the basic rock mass quality index BQ, corrects the basic rock mass quality index BQ, and obtains the rock mass quality exploration report; the rock hardness index Rc is obtained through rock physical and mechanical tests.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] The present invention extracts features from the source image, and constructs an accurate three - dimensional view model by using the MVS algorithm in combination with camera parameters. By converting to three - dimensional point cloud data and calculating the normal vector, it provides a basis for subsequent clustering segmentation and plane fitting. The application of the CFSFDP algorithm and the RANSAC algorithm makes the identification of structural planes more accurate, effectively distinguishing the same - group and non - same - group structural planes. Combining with the geostatistical method further improves the accuracy of rock mass structural feature analysis. By calculating the rock mass joint number Jv and combining with the rock hardness index Rc, the basic rock mass quality index BQ is obtained, and the basic rock mass quality index BQ is corrected, providing a scientific basis for the rock mass quality exploration report. The present invention improves the accuracy and efficiency of rock mass quality evaluation, and provides strong support for the design and construction of rock mass engineering. Description of the Drawings

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0079] Figure 1 It is a flow chart of the automated exploration method for open - pit rock mass based on the artificial intelligence large - model of the present invention;

[0080] Figure 2 It is a structural diagram of the automated exploration system for open - pit rock mass based on the artificial intelligence large - model of the present invention;

[0081] Figure 3 It is a framework diagram of the automated exploration method for open - pit rock mass based on the artificial intelligence large - model of the present invention. Detailed Embodiments

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0083] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0084] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0085] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, or the orientations or positional relationships in which the products of the invention are usually placed during use, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation of the present invention. In addition, terms such as "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0086] In addition, if the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0087] In the description of the embodiments of the present invention, it should also be noted that unless otherwise clearly specified and limited, if terms such as "set", "installed", "connected", "coupled" are understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0088] The following further describes the present invention in detail with reference to the accompanying drawings:

[0089] SeeFigure 1 , the present invention discloses an automated exploration method for open-pit rock masses based on an artificial intelligence large model, including:

[0090] S101: Extract features from the collected source image data to obtain the feature image of the source image;

[0091] S102: Process the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass based on the MVS algorithm to construct a three-dimensional view model;

[0092] Based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass, construct a cost volume; perform regularization processing on the constructed cost volume, and based on the regularized cost volume, obtain the depth estimation value of each pixel of the collected image, and construct a three-dimensional view model.

[0093] The construction of the cost volume based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass is specifically as follows:

[0094] Use the preset reference image as the reference image, and extract the feature map of the reference image through a convolutional neural network;

[0095] For each depth plane, calculate a differentiable homography transformation matrix based on the camera parameters and the depth value, and transform the feature map of the source image to the perspective of the reference image;

[0096] Calculate the similarity between the feature map of the transformed reference image and the feature map of the source image based on the Euclidean distance;

[0097] Stack the similarity matrices under each depth plane to form a multi-dimensional cost volume.

[0098] For each depth plane, calculating a differentiable homography transformation matrix based on the camera parameters and the depth value, and transforming the feature map of the source image to the perspective of the reference image is specifically as follows:

[0099] Preset the scene depth range as the minimum depth z min and the maximum depth z max , based on the internal reference focal length (f x , f y ) of the camera, the optical center (c x , c y ), the rotation matrix R and the translation vector t, for each pixel coordinate (x src , y src), according to the camera's internal parameters and the depth value z selected within the preset scene depth range, calculate the corresponding point coordinates (X, Y, Z) of this pixel in three-dimensional space, specifically as follows:

[0100] Z = z

[0101] where the depth value z is any value within the preset scene depth range; z min ≤ z ≤ z max

[0102] Based on the rotation matrix R and translation vector t of the camera, transform the three-dimensional point (X, Y, Z) to the three-dimensional space coordinates (X', Y', Z') in the perspective of the reference image, specifically as follows:

[0103]

[0104] where, is the homogeneous transformation matrix;

[0105] Perspectively project the transformed three-dimensional space coordinates (X', Y', Z') onto the reference image plane to obtain the corresponding pixel coordinates (x', y'), specifically as follows:

[0106]

[0107] Estimate a global homography transformation matrix based on an optimization method, and this matrix approximately describes the transformation relationship from the source image to the reference image;

[0108] The depth plane is a series of parallel planes in three-dimensional space that maintain a fixed distance from the camera optical center.

[0109] Regularize the constructed cost volume, and based on the regularized cost volume, obtain the depth estimate value of each pixel of the acquired image, and construct a three-dimensional view model;

[0110] Based on a 3D convolutional neural network, smooth and denoise the cost volume to obtain the regularized cost volume. Apply the Softmax function to the regularized cost volume to normalize the probability value of each depth to between 0 and 1; calculate the expected value of the probability distribution of each depth to obtain the depth estimate value of each pixel point;

[0111]

[0112] where, P(d k |(i, j)) represents the probability that the depth is d at the pixel (i, j) k ; cost(d k , (i, j)) represents the pixel (i, j) on the depth plane d kThe cost on; k' is the index variable in the summation, traversing all depth planes; d k represents the depth value of the k-th depth plane;

[0113] The above-mentioned calculation of the expected value of the probability distribution at each depth to obtain the depth estimate value of each pixel point is specifically as follows:

[0114] Estimated Depth(i,j) = ∑ k P(d k |(i,j)) * d k .

[0115] S103: Convert the constructed three-dimensional view into three-dimensional point cloud data and calculate the normal vectors of each point within the regional outcrop point cloud set;

[0116] For each pixel (x, y, d) in the depth map, where d is the depth value, calculate its coordinates (X, Y, Z) in three-dimensional space, specifically as follows:

[0117] Z = d

[0118] Take the calculated three-dimensional coordinates (X, Y, Z) as the points in the point cloud data;

[0119] where, f x , f y is the internal reference focal length of the camera; c x , c y is the optical center coordinate of the camera;

[0120] The above-mentioned calculation of the normal vectors of each point within the regional outcrop point cloud set is specifically as follows: For each point in the point cloud, select a neighborhood based on K nearest neighbors; use the neighborhood points to fit a plane, and the normal vector of the plane is the normal vector of this point.

[0121] S104: Based on the CFSFDP algorithm and the RANSAC algorithm, use the normal vectors of each point to perform clustering segmentation and plane fitting on the point cloud data, and the obtained best-fitting plane is the structural plane;

[0122] Based on the normal vectors of each point, calculate the angular difference between the point clouds; cluster the normal vectors based on the CFSFDP algorithm, identify the density peak points, and regard the density peak points as the clustering centers, and assign the sample points that are not clustering centers to the clusters belonging to the nearest neighbors with higher local density in turn according to the local density size;

[0123] Project the clustering result onto the stereographic projection map, and according to the projection result, count the number and average direction value of the clustering centers to initially identify the main structural plane set;

[0124] Points not assigned to the main set of structural planes are regarded as noise points. Iterative calculations are performed based on the inherent characteristics of the attitude and the Fisher K value to filter out noise and miscellaneous points. Subsequently, the dominant attitude clustering and the corresponding point cloud are obtained. Finally, by performing statistics in the stereographic projection diagram, the main direction of the point cloud set is the dominant attitude of this set of structural planes. The inherent characteristics of the attitude refer to the directional distribution characteristics of the structural planes in the geological body. After the regional search of the structural planes, the point sets of each structural plane are obtained, and then the random sample consensus (RANSAC) algorithm is used for the fitting and transformation of the rock mass structural planes. When using the RANSAC algorithm, effective initial data must be selected first. On this basis, the data set is continuously iteratively increased, and finally the final fitting result with the most data is selected. The least squares method is used to perform plane fitting on each structural plane point cloud set to obtain the best fitting plane. Based on the normal vector and position information of the fitting plane, the position and attitude of the structural plane are determined.

[0125] The RANSAC algorithm is specifically as follows:

[0126] Select initial point cloud data from the three-dimensional point cloud data set and calculate its corresponding plane equation:

[0127] Ax + By + Cz + D = 0

[0128] where A, B, C, and D are the parameters of the plane equation.

[0129] Calculate the algebraic distance di from all points to this plane. Select a threshold d. If di < dj, it is an effective point; otherwise, this point is not included in the model.

[0130] Repeat the above steps. Calculate the iteration end parameter according to the data set, and judge whether the iteration stops according to this iteration end parameter.

[0131] S105: Distinguish the structural planes based on geostatistics to obtain the same-group structural planes and non-same-group structural planes;

[0132] Through geostatistical attitude statistics, the structural planes with similar dip angles and plunge angles are regarded as the same-group structural planes; the structural planes with large differences in dip angles and plunge angles are regarded as non-same-group structural planes.

[0133] S106: Based on the rock mass structural characteristics of the same-group structural planes and non-same-group structural planes of the rock mass, calculate the rock mass volumetric joint count Jv;

[0134] Based on the same-group structural planes of the rock mass, obtain the distribution pattern of the rock mass in three-dimensional space, the extension length and connectivity of the same-group structural planes in the rock mass, and further evaluate the potential impact of the same-group structural planes on the stability of the rock mass under the action of groundwater, temperature change, and seismic load.

[0135] Based on the non - parallel structural planes in rock masses, analyze the intersection angles, intersection forms, and mechanical behaviors at the intersections between non - parallel structural planes; evaluate the degree of damage to the integrity of rock masses by non - parallel structural planes, including the size and shape of the blocks into which the rock mass is cut and the connection state between the blocks, and further evaluate the potential threats of non - parallel structural planes to the stability of rock masses under specific engineering conditions;

[0136] The calculated rock mass volumetric joint number Jv is specifically as follows:

[0137] For parallel structural planes, select representative survey lines in the rock mass, measure the spacing of the structural planes on the survey lines, and obtain the rock mass volumetric joint number Jv by summing the reciprocals of the spacings of each group of joints based on the spacing method;

[0138] For non - parallel structural planes, calculate the contributions of each group of structural planes to the rock mass volumetric joint number Jv respectively and accumulate them;

[0139] The contributions of each group of structural planes to the rock mass volumetric joint number Jv are calculated by the following formula:

[0140]

[0141] where n is the number of groups of structural planes in the statistical area; S i is the number of structural planes per unit normal length of the i - th group of structural planes; S0 is the number of non - grouped structural planes per unit volume of the rock mass.

[0142] S107: Obtain the basic quality index BQ of the rock mass through the rock hardness index Rc and the rock mass volumetric joint number Jv, consider the influence of groundwater K1, the attitude of structural planes K2, and the initial stress state K3 of the rock mass on the basic quality index BQ of the rock mass, and correct the basic quality index BQ of the rock mass to obtain the rock mass quality exploration report; the rock hardness index Rc is obtained through rock physical and mechanical tests.

[0143] The process of obtaining the basic quality index BQ of the rock mass through the rock hardness index Rc and the rock mass volumetric joint number Jv, and then obtaining the rock mass quality exploration report is specifically as follows:

[0144] BQ = 90 + 3Rc + 250Kv

[0145] where Kv is the rock mass integrity coefficient; the relationship between the rock mass integrity coefficient Kv and the rock mass volumetric joint number Jv is obtained according to the "Engineering Rock Mass Classification Standard".

[0146] Integrity degree of rock mass <![CDATA[Jv range (bars / m 3 )]]> Range of Kv Intact <3 >0.75 Relatively intact 3～10 0.75～0.55 Relatively fractured 10～20 0.55～0.35 Fractured 20～35 0.35～0.15 Highly fractured >35 <0.15

[0147] After obtaining the basic rock mass quality index BQ, it is also necessary to consider the correction of BQ by groundwater K1, the occurrence of structural planes K2, and the initial stress state K3. The corrected index [BQ] = BQ - 100(K1 + K2 + K3) is used for the final rock mass quality classification.

[0148] The rock mass quality exploration report includes the basic quality level of the rock mass. The basic quality level of the rock mass includes the rock hardness and the rock mass integrity, which are determined through qualitative classification and quantitative indicators. The qualitative classification includes: rock weathering degree classification, rock hardness qualitative classification, structural plane combination degree classification, rock mass integrity qualitative classification, and basic quality qualitative characteristics of the rock mass. The quantitative indicators include: rock hardness index Rc, rock mass integrity index, and basic rock mass quality index BQ.

[0149] Among them, the rock weathering degree classification, rock hardness qualitative classification, and structural plane combination degree classification are determined on-site by professionals. The rock hardness index Rc is obtained through rock physical and mechanical tests. The others are all analyzed by software according to the selected judgment criteria, and finally the rock mass quality evaluation conclusion is given.

[0150] See Figure 2 , the present invention discloses an automated exploration system for open-pit rock masses based on an artificial intelligence large model, including:

[0151] A feature extraction module that extracts features from the collected source image data to obtain the feature image of the source image.

[0152] A construction module that processes the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass based on the MVS algorithm to construct a three-dimensional view model.

[0153] A calculation module that converts the constructed three-dimensional view into three-dimensional point cloud data and calculates the normal vectors of each point in the regional outcrop point cloud set.

[0154] A first acquisition module that performs clustering segmentation and plane fitting on the point cloud data based on the CFSFDP algorithm and the RANSAC algorithm with the normal vectors of each point, and the obtained best-fitting plane is the structural plane.

[0155] A discrimination module that discriminates the structural planes based on geostatistics to obtain the same-group structural planes and non-same-group structural planes.

[0156] A second acquisition module that calculates the rock mass volume joint number Jv based on the rock mass structural characteristics of the same-group structural planes and non-same-group structural planes of the rock mass.

[0157] A third acquisition module, which acquires the basic rock mass quality index BQ through the rock hardness index Rc and the rock mass joint number per unit volume Jv, and corrects the basic rock mass quality index BQ by considering the influence of groundwater K1, the occurrence of structural planes K2 and the initial stress state K3 of the rock mass on the basic rock mass quality index BQ, so as to obtain a rock mass quality exploration report; the rock hardness index Rc is obtained through rock physical and mechanical tests.

[0158] Embodiment:

[0159] Refer to Figure 3 , the present invention discloses an automated exploration method for open-pit rock masses based on an artificial intelligence large model, including:

[0160] Step 1: Data collection

[0161] According to project conditions, various different rock mass types, typical, outcropping complete, and statistically significant slopes are selected on-site or according to existing three-dimensional scene models. An unmanned aerial vehicle equipped with an optical lens or a laser scanner and with a network rtk module is used to collect high-resolution images and data of these slopes at a close distance (10 - 30m).

[0162] These data cover the spatial position, shape, texture and other geological features of the rock mass. The collected data will be used to generate a three-dimensional slope real scene model and a geological information map, providing a basis for subsequent platform analysis work.

[0163] Step 2: On-site investigation and laboratory tests

[0164] To ensure that the evaluation results meet the requirements of national standards and guarantee the integrity and reliability of the data, professional geological technicians will conduct on-site investigations and sampling tests on the target slope rock mass, aiming at:

[0165] Determine the lithology, observe the rock composition on-site and refer to relevant materials to name the rock;

[0166] Qualitatively identify the rock hardness, and divide the hardness type according to the information feedback from hammering, knife scratching, immersion, etc.;

[0167] Qualitatively divide the rock weathering degree, and divide the weathering type according to the rock structure damage, mineral composition and color change, etc.;

[0168] Qualitatively divide the degree of structural plane combination, and divide the combination quality according to the information such as the opening degree, roughness, cementation state and filling material properties of the structural plane;

[0169] Collect rock samples for strength tests or collect existing rock physical and mechanical test results to obtain quantitative indicators for classifying rock hardness, quantitatively classify hardness types, and use them in the calculation of the basic quality index BQ of rock masses;

[0170] Select a small number of points on site and conduct manual measurements in strict accordance with the measurement methods required by the specifications. The content includes: the attitudes of strata and joint fissures, the number of developed groups of joint fissures, spacing, fissure width, etc. Organize and calculate the measurement results of the volumetric joint number Jv of the rock mass, and form a comparison and verification with the automatic calculation results of the platform.

[0171] In addition, on-site investigations can also improve other detailed information that may be missed or difficult to capture by drones, and improve the accuracy of the final 3D model.

[0172] Step 3: Model reconstruction and automatic structural interpretation

[0173] Using 3D image modeling algorithms, generate a 3D model with position attributes from the images and coordinate data obtained by drones; using machine learning algorithms, automatically identify and classify rock mass features, and extract and calculate evaluation parameters.

[0174] Step 3.1: Multi-view stereo reconstruction

[0175] MVSNet based on deep learning is an end-to-end method for estimating depth view by view from multi-view images and fusing them to generate a point cloud model. This method mainly includes four steps: feature extraction, cost volume construction, cost volume regularization, and depth regression.

[0176] Among them, the purpose of feature extraction is to extract useful feature information from the input multi-view images, and these feature information are crucial for subsequent depth estimation and 3D reconstruction.

[0177] The process is as follows: Input multiple images taken from different perspectives. Commonly used feature extraction methods include SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), etc. These methods can extract feature points that are invariant to scale, rotation, and illumination changes. In the era of deep learning, convolutional neural networks are also widely used in feature extraction. By training the CNN, more robust and effective feature representations can be automatically learned.

[0178] Output the feature maps corresponding to each image, and these feature maps contain the key feature information in the images and are used for subsequent processing.

[0179] The purpose of cost volume construction is to construct a cost volume based on the extracted feature maps to represent the matching cost between feature points from different perspectives.

[0180] The process is as follows: Input the feature maps of the reference image and the source image, as well as the camera parameters. Through differentiable homography transformation, the feature map of the source image is transformed into the perspective of the reference image to form multiple hypothetical depth planes. For each depth plane, calculate the similarity (such as correlation, distance, etc.) between the feature map of the reference image and the transformed feature map of the source image to form a cost volume. Each element in the cost volume represents the matching cost between the feature points at the corresponding positions under a specific depth hypothesis.

[0181] Output a multi-dimensional cost volume, whose dimensions usually include the number of depth hypotheses, the number of channels of the feature map, the height and width of the feature map.

[0182] The purpose of cost volume regularization is to regularize the cost volume to reduce the influence of noise and outliers and improve the accuracy and robustness of depth estimation.

[0183] The process is as follows: Input the cost volume; use a 3D convolutional neural network (3D CNN) to regularize the cost volume. 3D CNN can effectively capture the spatial information in the cost volume and smooth and denoise the cost volume through operations such as convolution and pooling. Commonly used network structures include 3D U-Net, etc. These network structures can effectively extract the deep features in the cost volume while maintaining the spatial resolution.

[0184] Output the regularized cost volume or probability, which contains the probability distribution of each pixel point under different depth hypotheses.

[0185] The purpose of depth regression is to regress the depth value of each pixel point based on the regularized cost volume or probability volume.

[0186] The process is as follows: Input the regularized cost volume or probability volume, apply the Softmax function to the probability volume to normalize the probability value of each depth to between 0 and 1. By calculating the expected value of the probability distribution of each depth, the depth estimation value of each pixel point is obtained. Optionally, structures such as residual networks can be used to further optimize and adjust the preliminary depth estimation value.

[0187] Output a depth map, which contains the depth value of each pixel point in the input image.

[0188] Step 3.2: Automatic interpretation of structural planes

[0189] (1) Calculate the normal vectors of each point in the regional outcrop point cloud set.

[0190] According to the normal vectors of each point obtained above, perform similarity judgment and clustering on the point cloud to determine that the main point cloud set is the same group of structural planes, and conduct statistical analysis in the stereographic projection diagram to obtain that the main direction of the point cloud set is the dominant occurrence of this group of structural planes.

[0191] Cluster and segment the point cloud set, perform plane fitting, and the best-fitting plane obtained is the structural plane, and the attitude of the corresponding structural plane is calculated through the normal vector.

[0192] Calculate geometric information of the structural plane such as trace length and spacing.

[0193] (2) Identification of structural planes in the same group is achieved by the Clustering with Fast Search and Find of Density Peaks (CFSFDP) algorithm for clustering analysis of the normal vectors. The CFSFDP algorithm is based on the local density and controlled distance of each discontinuity point, and realizes by measuring the similarity of all discontinuity point data. The number and average direction value of the potential discontinuity point set can be obtained through the stereographic projection diagram.

[0194] Since there are non-discontinuity points in the point cloud, this method can detect whether each point belongs to a given structural plane set. If a point is not assigned to the main structural plane set, it is regarded as a noise point. Use the inherent characteristics of the attitude (conforming to the Fisher distribution) to iteratively calculate the Fisher K value to filter out noise points, and then obtain the dominant attitude clustering and the matching point cloud. Finally, the main direction of the point cloud set is obtained through statistical analysis in the stereographic projection diagram, which is the dominant attitude of this group of structural planes.

[0195] For the plane fitting of the structural plane, after the regional search of the structural plane, the point sets of each structural plane can be obtained, and then the random sample consensus (RANSAC) algorithm is used for the fitting and transformation of the rock mass structural plane. When using the RANSAC algorithm, first select effective initial data, and on this basis, continuously iterate to increase the data set. Finally, the model with the most data is selected as the final model.

[0196] The process of RANSAC plane fitting is as follows:

[0197] ① Select initial point cloud data in the three-dimensional point cloud dataset and calculate its corresponding plane equation:

[0198] Ax + By + Cz + D = 0

[0199] Where: A, B, C, and D are the parameters of the plane equation.

[0200] ② Calculate the algebraic distance di from all points to this plane, select a threshold d, if di < dj, it is a valid point, otherwise this point is not included in the model;

[0201] ③ Repeat the above steps, calculate the iteration end parameter according to the dataset, and judge whether the iteration stops according to this iteration end parameter.

[0202] (3) Identification of structural planes in different groups is data collection and multi-angle dataset construction:

[0203] Images of the rock are collected from different angles. These image data can reflect the local terrain geometric information and can be used to construct an image dataset containing multi-scale features.

[0204] The collected images are preprocessed using image processing techniques, including steps such as image enhancement, grayscale processing, smoothing, and binarization, in order to improve the quality of the images. The intensity of the feature signals is quantified using metrics such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), and the scale space theory is combined to analyze the influence of different-scale features on the recognition accuracy.

[0205] Deep learning model training:

[0206] The preprocessed multi-scale dataset is used to train the deep learning model. Usually, a convolutional neural network (CNN) or a more advanced deep convolutional neural network (DCNN) is used for image feature extraction and recognition.

[0207] For image features at different scales, the model needs to have the ability to recognize multi-scale features to adapt to complex environmental noise.

[0208] Application and verification: Through the above method, the model's ability to recognize the joint number of rock masses can be enhanced. This method minimizes the influence of environmental noise on the recognition results, thereby improving the recognition accuracy and robustness. Finally, the effectiveness of this method in recognizing the joint number of new projects is verified through experiments, and its performance in improving the recognition accuracy of geological discontinuities in complex environments is analyzed. Combining with actual rock joint data, the practicability and accuracy of the model are verified.

[0209] Step 4: Quality assessment system

[0210] An evaluation framework and process are formulated according to national standard requirements. The basic quality level of rock masses is determined by two factors: the rock hardness and the rock mass integrity, and these two factors should be determined through two methods: qualitative classification and quantitative indicators. Specifically, it includes:

[0211] (1) Qualitative classification: ① Rock weathering degree classification, ② Qualitative classification of rock hardness, ③ Classification of structural plane combination degree, ④ Qualitative classification of rock mass integrity, ⑤ Qualitative characteristics of rock mass basic quality;

[0212] (2) Quantitative indicators: ⑥ Rock hardness index, ⑦ Rock mass integrity index, ⑧ Rock mass basic quality index (BQ).

[0213] The classification and determination criteria for all the above items are consistent with the relevant standards. Among them, ① the classification of rock weathering degree, ② the qualitative classification of rock hardness, and ③ the classification of the bonding degree of structural planes are determined on-site by professionals and input into the software platform; ⑥ the rock hardness index is obtained through rock physical and mechanical tests and then input into the software platform; for other items, the software automatically calculates and analyzes according to the selected determination criteria, and finally gives the conclusion of rock mass quality assessment.

[0214] Step 5: Output of result documents

[0215] The platform can customize the content and style of the report according to the evaluation results to meet the needs of different users. At the same time, it provides a report export function, allowing users to export the report in multiple formats (such as PDF, Excel, etc.). The report content includes:

[0216] (1) Statistical indicators: including basic indicators such as the number of fissures and the extension length, as well as calculations and analyses according to industry standard formulas, such as the linear fissure rate, attitude distribution, recognition confidence rate, etc.

[0217] (2) Visual reports: Provide multiple types of visual reports according to the requirements of current business operations, such as rose diagrams, pole diagrams, density diagrams, etc., which can intuitively display information such as the distribution of fissures and structural planes, attitude characteristics, and recognition results.

[0218] (3) Conclusion of rock mass quality classification: According to the analysis and calculation results of a certain target rock mass input, output the quality classification conclusion according to the requirements of national standards, including a detailed description of the analysis process.

[0219] The terminal device provided by the embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in each of the above method embodiments. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in each of the above device embodiments.

[0220] The computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention.

[0221] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0222] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0223] The memory can be used to store the computer program and / or module. By running or executing the computer program and / or module stored in the memory, and by invoking the data stored in the memory, the processor implements various functions of the terminal device.

[0224] If the modules / units integrated in the terminal device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above various method embodiments. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0225] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An automated exploration method for open-pit rock masses based on artificial intelligence large models, characterized in that, Including: Performing feature extraction on the collected source image data to obtain the feature image of the source image; Based on the MVS algorithm, processing the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass to construct a three-dimensional view model; Converting the constructed three-dimensional view into three-dimensional point cloud data, and calculating the normal vectors of each point within the regional outcrop point cloud set; Based on the CFSFDP algorithm and the RANSAC algorithm, clustering and segmenting the point cloud data with the normal vectors of each point and performing plane fitting, and the obtained best-fitting plane is the structural plane; Differentiating the structural plane based on geostatistics to obtain the same-group structural planes and non-same-group structural planes; Based on the rock mass structural characteristics of the same-group structural planes and non-same-group structural planes of the rock mass, and calculating the rock mass volumetric joint number Jv; Through the rock hardness index Rc and the rock mass volumetric joint number Jv, obtaining the basic quality index BQ of the rock mass, considering the influence of groundwater K1, structural plane attitude K2, and the initial stress state K3 of the rock mass on the basic quality index BQ of the rock mass, and correcting the basic quality index BQ of the rock mass to obtain a rock mass quality exploration report; the rock hardness index Rc is obtained through rock physical and mechanical tests.

2. The automated exploration method for open-pit rock mass based on artificial intelligence large model according to claim 1, characterized in that The processing of the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass based on the MVS algorithm to construct a three-dimensional view model is specifically: based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass, constructing a cost volume; performing regularization processing on the constructed cost volume, and based on the regularized cost volume, obtaining the depth estimation value of each pixel of the collected image, and constructing a three-dimensional view model.

3. The automated exploration method for open-pit rock masses based on the large artificial intelligence model according to claim 2, characterized in that, The constructing a cost volume based on the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass is specifically: Taking the preset reference image as the reference image, and extracting the feature map of the reference image through a convolutional neural network; For each depth plane, calculating a differentiable homography transformation matrix based on the camera parameters and the depth value, and transforming the feature map of the source image to the perspective of the reference image; Calculating the similarity between the feature map of the transformed reference image and the feature map of the source image based on the Euclidean distance; Stacking the similarity matrices under each depth plane to form a multi-dimensional cost volume.

4. The automated exploration method for open-pit rock mass based on the artificial intelligence large model according to claim 3, wherein The calculating a differentiable homography transformation matrix based on the camera parameters and the depth value for each depth plane, and transforming the feature map of the source image to the perspective of the reference image is specifically: The preset scene depth range is from the minimum depth z min to the maximum depth z max . Based on the internal focal length (f x , f y ) and optical center (c x , c y ) of the camera, the rotation matrix R, and the translation vector t, for each pixel coordinate (x src , y src ) in the source image, according to the internal parameters of the camera and the depth value z selected within the preset scene depth range, calculate the corresponding point coordinates (X, Y, Z) of this pixel in three-dimensional space, specifically as follows: Among them, the depth value z is any value within the preset scene depth range; z min ≤z≤z max Based on the rotation matrix R and translation vector t of the camera, transforming the three-dimensional point (X, Y, Z) to the three-dimensional space coordinates (X', Y', Z') in the perspective of the reference image, specifically: Among them, is a homogeneous transformation matrix; Perspectively projecting the transformed three-dimensional space coordinates (X', Y', Z') onto the reference image plane to obtain the corresponding pixel coordinates (x', y'), specifically: Estimating a global homography transformation matrix based on an optimization method, and this matrix approximately describes the transformation relationship from the source image to the reference image; The depth plane is a series of parallel planes in three-dimensional space that maintain a fixed distance from the camera optical center.

5. The automated exploration method for open-pit rock mass based on the large artificial intelligence model according to claim 4, characterized in that Regularize the constructed cost volume, and based on the regularized cost volume, obtain the depth estimation value of each pixel in the acquired image, and construct a three-dimensional view model. Specifically: Smooth and denoise the cost volume based on a 3D convolutional neural network to obtain the regularized cost volume, apply the Softmax function to the regularized cost volume to normalize the probability value of each depth to between 0 and 1; Calculate the expected value of the probability distribution of each depth to obtain the depth estimation value of each pixel point. where, P(d k |(i,j)) represents the probability of depth d k at pixel (i,j); cost(d k ,(i,j)) represents the cost of pixel (i,j) on depth plane d k ; k' is the index variable in the summation, traversing all depth planes; d k represents the depth value of the k-th depth plane; The calculation of the expected value of the probability distribution of each depth to obtain the depth estimation value of each pixel point is specifically as follows: Estimated Depth(i,j)=∑ k P(d k |(i,j))*d k 。 6. The automated exploration method for open-pit rock mass based on the large artificial intelligence model according to claim 5, wherein, Convert the constructed three-dimensional view into three-dimensional point cloud data, and calculate the normal vectors of each point in the regional outcrop point cloud set. Specifically: For each pixel (x, y, d) in the depth map, where d is the depth value, calculate its coordinates (X, Y, Z) in three-dimensional space. Specifically: Take the calculated three-dimensional coordinates (X, Y, Z) as the points in the point cloud data. Among them, f x , f y is the internal focal length of the camera; c x , c y is the optical center coordinate of the camera; The calculation of the normal vectors of each point in the regional outcrop point cloud set is specifically as follows: For each point in the point cloud, select a neighborhood based on K nearest neighbors; Use the neighborhood points to fit a plane, and the normal vector of the plane is the normal vector of the point.

7. The automated exploration method for open-pit rock mass based on the large artificial intelligence model according to claim 6, wherein Based on the CFSFDP algorithm and the RANSAC algorithm, use the normal vectors of each point to perform clustering segmentation and plane fitting on the point cloud data. The obtained best-fitting plane is the structural plane. Specifically: Calculate the angular difference between point clouds based on the normal vectors of each point; Cluster the normal vectors based on the CFSFDP algorithm, identify the density peak points, and regard the density peak points as the clustering centers. Assign the sample points that are not clustering centers to the clusters of the nearest neighbors with higher local density in turn according to the local density size. Project the clustering result onto a stereographic projection diagram. According to the projection result, count the number and average direction value of the clustering centers, and preliminarily identify the main structural plane set. Regard the points not assigned to the main structural plane set as noise points, and perform iterative calculations based on the inherent characteristics of the attitude and the Fisher K value to filter out noise clutter points; Then obtain the dominant attitude clustering and the matching point cloud; Finally, through statistics in the stereographic projection diagram, the main direction of the point cloud set is the dominant attitude of this group of structural planes; The inherent characteristics of the attitude are the direction distribution characteristics of the structural planes in the geological body; After the regional search of the structural planes, obtain the point sets of each structural plane, and then use the random sample consensus RANSAC algorithm for rock mass structural plane fitting and transformation; When using the RANSAC algorithm, first select effective initial data, continuously iterate to increase the data set on this basis, and finally select the final fitting result with the most data; Use the least squares method to perform plane fitting on each structural plane point cloud set to obtain the best-fitting plane; Determine the position and attitude of the structural plane according to the normal vector and position information of the fitting plane. The structural planes are distinguished based on geostatistics to obtain the structural planes of the same group and those of different groups, specifically: through geostatistical dip statistics, the structural planes with similar direction angles and dip angles are regarded as the structural planes of the same group; the structural planes with significantly different direction angles and dip angles are regarded as the structural planes of different groups.

8. The automated exploration method for open-pit rock masses based on the large artificial intelligence model according to claim 7, characterized in that The RANSAC algorithm is specifically as follows: Select initial point cloud data from the three-dimensional point cloud dataset and calculate its corresponding plane equation: Ax + By + Cz + D = 0 where A, B, C, and D are the parameters of the plane equation; Calculate the algebraic distance di from all points to this plane, select a threshold d, if di < dj, it is a valid point, otherwise this point is not included in the model; Repeat the above steps, calculate the iteration end parameter according to the dataset, and judge whether the iteration stops according to this iteration end parameter.

9. The automated exploration method for open-pit rock mass based on the large artificial intelligence model according to claim 8, characterized in that The rock mass structural characteristics of the structural planes of the same group and those of different groups in the rock mass are specifically as follows: Based on the structural planes of the same group in the rock mass, obtain the distribution pattern of the rock mass in three-dimensional space, the extension length and connectivity of the structural planes of the same group in the rock mass, and then evaluate the potential impact of the structural planes of the same group on the stability of the rock mass under the action of groundwater, temperature change, and seismic load; Based on the structural planes of different groups in the rock mass, analyze the intersection angles, intersection forms, and mechanical behaviors at the intersections of the structural planes of different groups; evaluate the degree of damage to the integrity of the rock mass by the structural planes of different groups, including the size and shape of the blocks into which the rock mass is cut and the connection state between the blocks, and then evaluate the potential threat of the structural planes of different groups to the stability of the rock mass under specific engineering conditions; The calculated rock mass volume joint number Jv is specifically as follows: For the structural planes of the same group, select a representative survey line in the rock mass, measure the spacing of the structural planes on the survey line, and sum the reciprocals of the spacing of each group of joints based on the spacing method to obtain the rock mass volume joint number Jv; For the structural planes of different groups, calculate the contribution of each group of structural planes to the rock mass volume joint number Jv and accumulate them; The contribution of each group of structural planes to the rock mass volume joint number Jv is calculated by the following formula: where n is the number of joint sets within the statistical region; S i is the number of joints per unit normal length of the i-th joint set; S0 is the number of non-grouped joints per unit volume of rock mass; The basic quality index BQ of the rock mass is obtained through the rock hardness index Rc and the rock mass volume joint number Jv, and then the rock mass quality exploration report is obtained, specifically: BQ = 90 + 3Rc + 250Kv where Kv is the rock mass integrity coefficient; according to the "Engineering Rock Mass Classification Standard", the relationship between the rock mass integrity coefficient Kv and the rock mass volume joint number Jv is obtained; After obtaining the basic quality index of the rock mass as BQ, it is also necessary to consider the correction of BQ by groundwater K1, structural plane occurrence K2, and initial stress state K3. The corrected index [BQ] = BQ - 100(K1 + K2 + K3), which is used for the final rock mass quality classification; The rock mass quality exploration report includes the basic quality level of the rock mass; the basic quality level of the rock mass includes the rock hardness degree and the rock mass integrity degree, and the rock hardness degree and the rock mass integrity degree are determined through qualitative classification and quantitative indexes; the qualitative classification includes: rock weathering degree classification, rock hardness degree qualitative classification, structural plane combination degree classification, rock mass integrity degree qualitative classification and rock mass basic quality qualitative characteristics; the quantitative indexes include: rock hardness degree index Rc, rock mass integrity degree index and rock mass basic quality index BQ; Among them, the rock weathering degree classification, the rock hardness degree qualitative classification, and the structural plane combination degree classification are determined on-site by professionals; the rock hardness degree index Rc is obtained through rock physical and mechanical tests; the others are all analyzed by software according to the selected judgment criteria, and finally the rock mass quality evaluation conclusion is given.

10. An automated exploration system for open-pit rock masses based on artificial intelligence large models, characterized in that, Including: A feature extraction module, which extracts features from the collected source image data to obtain the feature image of the source image; A construction module, which processes the feature image of the source image, the selected reference image, and the camera parameters used for collecting the image data of the target open-pit slope rock mass based on the MVS algorithm to construct a three-dimensional view model; A calculation module, which converts the constructed three-dimensional view into three-dimensional point cloud data and calculates the normal vectors of each point in the regional outcrop point cloud set; A first acquisition module, which based on the CFSFDP algorithm and the RANSAC algorithm, clusters and segments the point cloud data with the normal vectors of each point and performs plane fitting, and the obtained best-fitting plane is the structural plane; A discrimination module, which discriminates the structural plane based on geostatistics to obtain the same-group structural planes and non-same-group structural planes; A second acquisition module, which based on the rock mass structural characteristics of the same-group structural planes and non-same-group structural planes of the rock mass and calculates the rock mass volume joint number Jv; A third acquisition module, which obtains the rock mass basic quality index BQ through the rock hardness degree index Rc and the rock mass volume joint number Jv, considers the influence of groundwater K1, structural plane occurrence K2 and the initial stress state K3 of the rock mass on the rock mass basic quality index BQ, corrects the rock mass basic quality index BQ, and obtains the rock mass quality exploration report; the rock hardness degree index Rc is obtained through rock physical and mechanical tests.

Images