Open-pit mine slope rock mass structure plane intelligent identification extraction method, device, equipment and storage medium

Abstract

The application provides an open-pit mine slope rock mass structural plane intelligent identification and extraction method, device and equipment and a storage medium, and relates to the technical field of rock mass surface feature identification. The method comprises the following steps: acquiring a high-precision three-dimensional geological outcrop model; performing two-dimensional projection on different regions of the model and labeling structural plane contours to train a semantic segmentation model; identifying a target slope projection picture by using the trained model to obtain a two-dimensional structural plane identification result; mapping a two-dimensional structural plane trace back to a three-dimensional model based on projection parameters to obtain a three-dimensional space structural plane; performing cleaning and merging processing on the three-dimensional structural plane, eliminating an out-of-plane or twisted structural plane, merging coplanar structural planes, and calculating a structural plane normal vector to obtain occurrence information such as a dip direction and a dip angle. The application solves the problems of low efficiency, high danger and dependence on manual experience of a traditional measurement method, and realizes automatic, batch and high-precision extraction of open-pit mine slope rock mass structural planes.

Description

Technical Field

[0001] This application relates to the field of rock mass surface feature recognition technology, and in particular to a method, device, equipment and storage medium for intelligent recognition and extraction of rock mass structural surfaces on open-pit mine slopes. Background Technology

[0002] With the increasing scale and depth of slope engineering mining, landslides and other disasters caused by slope instability are becoming more frequent. The stability of the slope directly affects whether such disasters will occur. Open-pit mine slopes and rock slopes have a large number of structural planes, such as joints, fissures, bedding, and faults. Their spatial combination directly affects the stability of the slope. How to accurately and quickly provide information on the structural planes on each step is a problem of great concern to researchers.

[0003] Traditional methods for identifying rock surfaces and acquiring attitude information typically involve contact measurements. Surveyors use handheld tools such as geological compasses, measuring ropes, and measuring tapes to obtain indicators such as the attitude, spacing, and trace length of accessible rock surfaces. However, in practice, traditional measurement methods are prone to significant errors due to the complexity of the surveyors and the site conditions, and are also inefficient and dangerous. Modern rock engineering projects are increasingly large-scale and fast-paced, making traditional measurement methods inadequate for modern needs. Furthermore, for complex construction sites such as steep slopes and mine slopes, traditional methods make it difficult for surveyors to reach the work area. With the rapid development of measurement technology, non-contact measurement has gained increasing popularity among researchers. Technologies such as drones, 3D laser scanning, and digital cameras can quickly acquire rock mass characterization data without contacting the rock surface. Compared to traditional methods, this is more time-saving, labor-saving, and less dangerous, making it suitable for the safe and efficient acquisition of original information on structural surfaces in modern large-scale geotechnical engineering projects.

[0004] To date, researchers have conducted some studies on structural surface acquisition. However, certain problems remain for open-pit mines, which often feature steep slopes. Two-dimensional image or point cloud interpretation relies heavily on human experience, leading to a high degree of subjectivity. Exposed features are severely damaged by the open-pit mining process, surface structural characteristics are indistinct, and the surface is often characterized by near-horizontal or vertical layered structures with indistinct curvature. Furthermore, existing techniques for identifying open-pit mine slope features suffer from low accuracy, poor performance, and slow speed, making them largely ineffective. Higher precision in extracting structural surfaces from complex rock masses is crucial for geologists to assess and analyze the rock mass; however, accurate extraction of structural surfaces from complex rock masses requires more precise technical support. Summary of the Invention

[0005] This application provides a method, device, equipment, and storage medium for intelligent identification and extraction of rock mass structural surfaces on open-pit mine slopes. By using UAV photogrammetry, deep learning semantic segmentation, and structural surface cleaning algorithms, the projected two-dimensional images are identified, structural surface contour traces are extracted, and mapped onto a three-dimensional geological model to obtain structural surfaces that conform to the characteristics of real rock masses.

[0006] Firstly, this application provides a method for intelligent identification and extraction of rock mass structural surfaces on open-pit mine slopes, including:

[0007] Obtain a three-dimensional geological outcrop model of the rock mass on the open-pit mine slope;

[0008] Two-dimensional projections are performed on different regions of the three-dimensional geological outcrop model to generate corresponding two-dimensional projection images. The structural surface contours on the two-dimensional projection images are labeled to construct a training dataset. A semantic segmentation model for intelligent recognition of structural surfaces is trained based on the training dataset.

[0009] Using the trained semantic segmentation model, the two-dimensional projection image of the target open-pit mine slope is identified to obtain the two-dimensional structural surface recognition result;

[0010] Based on the projection parameters of the two-dimensional projected image, the structural surface trace contours in the two-dimensional structural surface recognition results are extracted, and the structural surface trace contours are mapped to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface.

[0011] The obtained three-dimensional spatial structure surfaces are cleaned and merged, and the normal vectors of the processed structure surfaces are calculated to obtain the three-dimensional spatial attribute information of the dip and tilt angle of the three-dimensional spatial structure surfaces.

[0012] In one possible design, a three-dimensional geological outcrop model of the open-pit mine slope rock mass is obtained through the following method:

[0013] Using oblique photography by drones, real-world geological image data of the target open-pit mine slope rock mass are collected by flying in a terrain-following manner, and the drone flight path, heading overlap and lateral overlap are set according to the target area range;

[0014] Based on the real-scene geological image data, the three-dimensional geological outcrop model is constructed.

[0015] In one possible design, the method of generating corresponding two-dimensional projection images by performing two-dimensional projection on different regions of the three-dimensional geological outcrop model includes: dividing the three-dimensional geological outcrop model into regions according to the step height, slope normal, or spatial position, and performing two-dimensional projection on the step surfaces of each region obtained from the region division; wherein, the two-dimensional projection includes one or more of orthographic projection, oblique projection, or perspective projection, and at least two different perspectives are used to generate two-dimensional projection images for the same region.

[0016] In one possible design, the semantic segmentation model is a pixel-level classification model based on a convolutional neural network, a Transformer network, or a combination of both; during the training process, the semantic segmentation model uses a weighted cross-entropy loss function, a Dice loss function, or a combination thereof, and alleviates the sample imbalance problem by setting different loss weights for different categories.

[0017] In one possible design, after using a trained semantic segmentation model to identify a two-dimensional projection image of a target open-pit mine slope and obtain a two-dimensional structural surface identification result, the method further includes: setting a confidence threshold for the two-dimensional structural surface identification result, and identification results below the confidence threshold are not included in subsequent three-dimensional mapping processing.

[0018] In one possible design, the three-dimensional structural surfaces are cleaned and merged, specifically including:

[0019] For each set of three-dimensional structural surface points, perform plane fitting, calculate the structural surface normal vector, and calculate the distance from each point in the point set to the fitting plane. If there is a point whose distance to the fitting plane is greater than a preset first distance threshold, then the structural surface is determined to be an irregular structural surface or a twisted structural surface and is deleted.

[0020] The coplanarity of two structural surfaces is determined based on the angle between their normal vectors and the distance between their planes. When the angle between their normal vectors is less than a preset angle threshold and the distance between their planes is less than a preset second distance threshold, the two structural surfaces are determined to be coplanar. The coplanar structural surfaces are then merged. The merging process includes: constructing a local two-dimensional coordinate system, projecting the multiple structural surfaces to be merged onto a two-dimensional plane for geometric merging, generating a single closed polygon, and then converting it back to three-dimensional space.

[0021] The merged structural surfaces are fitted with planes and distances are checked again. If there is a point whose distance to the fitted plane is greater than a preset first distance threshold, the merged structural surface is deleted.

[0022] In one possible design, a plane fitting is performed on each set of points on the 3D structural surface, the normal vector of the structural surface is calculated, and the distance from each point in the set to the fitted plane is calculated, specifically including:

[0023] Obtain the structural surface point set, which consists of all points on the structural surface, and the spatial coordinates of the points are represented as follows:

[0024] P i =(x i ,y i ,z i ), i=1,2,…,N (1)

[0025] Among them, P i Let x be the spatial coordinates of the i-th point. i ,y i ,z i These are the X-axis, Y-axis, and Z-axis coordinates of the i-th point, respectively, and N is the total number of points contained in the structural surface;

[0026] The geometric center of the point set of the structural surface is calculated using the following formula:

[0027] , , (2)

[0028] (3)

[0029] in, The geometric center of the set of points on the structural surface. , , Let X, Y, and Z represent the arithmetic mean of the X, Y, and Z coordinates of all points, respectively.

[0030] Subtract the geometric center from each point to construct the decentralized point set matrix A, which is represented as:

[0031] (4)

[0032] in, It is an N×3 matrix;

[0033] Performing singular value decomposition on the point set matrix A yields:

[0034] (5)

[0035] in, C is an N×N orthogonal matrix; C is an N×3 diagonal matrix whose diagonal elements are singular values, namely the first singular value σ1, the second singular value σ2, and the third singular value σ3, where σ1≥σ2≥σ3≥0; V is a 3×3 orthogonal matrix; T is the matrix transpose operator;

[0036] The right singular vector corresponding to the third singular value σ3 The normal vector of the plane that best fits the set of structural points , is represented as:

[0037] (6)

[0038] in, , , These are the components of the normal vector along the X, Y, and Z axes, respectively.

[0039] Using the normal vector And the geometric center P, construct the plane equation of the structural surface, expressed as:

[0040] (7)

[0041] in, , and Let be the coordinate variable of any point in the plane equation;

[0042] Alternatively, the planar equation of the structural surface can be equivalently expressed as:

[0043] (8)

[0044] (9)

[0045] in, This is the constant term in the plane equation;

[0046] Calculate any point P i (x i ,y i ,z i The vertical distance from the plane of the structure The calculation formula is:

[0047] (10).

[0048] Secondly, this application provides an intelligent identification and extraction device for rock mass structural surfaces on open-pit mine slopes, the device comprising:

[0049] The data acquisition module is configured to acquire a three-dimensional geological outcrop model of the open-pit mine slope rock mass;

[0050] The model building module is configured to perform two-dimensional projection on different regions of the three-dimensional geological outcrop model, generate corresponding two-dimensional projection images, annotate the structural surface contours on the two-dimensional projection images, build a training dataset, and train a semantic segmentation model for intelligent recognition of structural surfaces based on the training dataset.

[0051] The structural surface recognition module is configured to use a trained semantic segmentation model to recognize the two-dimensional projection image of the target open-pit mine slope and obtain the two-dimensional structural surface recognition result.

[0052] The trace contour mapping module is configured to extract the structural surface trace contour from the two-dimensional structural surface recognition result based on the projection parameters of the two-dimensional projected image, and map the structural surface trace contour to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface.

[0053] The structural surface information calculation module is configured to clean and merge the obtained three-dimensional spatial structural surfaces, and calculate the normal vectors of the processed structural surfaces to obtain the three-dimensional spatial attribute information of the three-dimensional spatial structural surfaces, including their dip and tilt angles.

[0054] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes as described in the first aspect and various possible designs of the first aspect.

[0055] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes as described in the first aspect and various possible designs of the first aspect.

[0056] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes as described in the first aspect and various possible designs of the first aspect.

[0057] The intelligent identification and extraction method, device, equipment, and storage medium for open-pit mine slope rock mass structure planes provided in this application have at least the following beneficial effects:

[0058] Based on the acquisition of a three-dimensional geological outcrop model of the rock mass, this application performs dimensionality reduction on the three-dimensional geological model through screenshots, utilizes a semantic segmentation model to achieve rapid and large-scale recognition processing, and extracts traces to map them onto the three-dimensional geological model. This process enhances the dimensionality of the structural surface results identified in the image, resulting in structural surfaces with three-dimensional spatial attributes. This method is more comprehensively applicable to the extraction of rock slope structural surfaces in complex scenarios such as mines. In practical applications, it can provide data support for rock slope stability. This invention is simple to implement, has significant effects, and meets the requirements of the application. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] Figure 1 A schematic diagram illustrating the implementation principle of an intelligent identification and extraction method for rock mass structure surfaces on open-pit mine slopes, provided in this application embodiment;

[0061] Figure 2 A flowchart illustrating an intelligent identification and extraction method for rock mass structure surfaces on open-pit mine slopes, provided in this application embodiment;

[0062] Figure 3 A flowchart illustrating the process of obtaining a three-dimensional geological outcrop model of an open-pit mine slope rock mass, as provided in this application embodiment.

[0063] Figure 4 A flowchart illustrating the training process of the semantic segmentation model provided in this application embodiment;

[0064] Figure 5 A flowchart of the structural surface cleaning algorithm provided in the embodiments of this application;

[0065] Figure 6 A flowchart of the singular value decomposition method provided in the embodiments of this application;

[0066] Figure 7 The following is an example of the intelligent identification and extraction method for three-dimensional structural surfaces of rock mass on open-pit mine slopes provided in this application embodiment;

[0067] Figure 8 A structural diagram of the intelligent identification and extraction device for rock mass structure surfaces on open-pit mine slopes provided in this application embodiment.

[0068] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0070] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0071] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0072] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0073] This application provides an intelligent identification and extraction method for rock mass structural surfaces on open-pit mine slopes. This method is applicable to engineering scenarios such as open-pit mine slopes and steep rock slopes. Based on acquiring high-precision three-dimensional geological outcrop model data of the rock mass, and using a semantic segmentation model, it can semi-automatically extract specified structural surfaces through simple manual interaction. Different structural surface extraction methods are more comprehensively applicable to the extraction of rock mass slope structural surfaces in complex scenarios such as mines. At the same time, the calculation of the attitude of the extracted structural surfaces can provide data support for the stability of the rock mass slope. This invention is simple to implement, flexible to operate, and has significant effects, meeting the application requirements.

[0074] like Figure 1 The diagram illustrates the implementation principle of an intelligent identification and extraction method for rock mass structural surfaces on open-pit mine slopes. This method utilizes a UAV equipped with a high-precision lens as the data acquisition hardware. High-precision image data of the open-pit mine slope rock mass is acquired using this hardware, and a high-precision three-dimensional geological outcrop model is constructed based on this image data. The constructed high-precision three-dimensional geological outcrop model undergoes model projection processing, and a training dataset is created based on the projection results. This training dataset is input into a semantic segmentation model to complete model training and optimization. After structural surface identification using the trained semantic segmentation model, trace extraction is performed on the identification results to initially obtain the bench structural surfaces of the mining area. For the initially obtained bench structural surfaces, optimization processing steps are sequentially performed, including merging similar structural surfaces and removing dissimilar structural surfaces. Through this series of optimization operations, invalid and duplicate structural surface information is eliminated, ultimately obtaining a large number of high-accuracy structural surfaces. Based on these high-accuracy structural surfaces, key spatial attribute information of the rock mass structural surfaces, such as the orientation and average spacing of the structural surfaces, can be further calculated accurately.

[0075] Specifically, such as Figure 2 As shown, the intelligent identification and extraction method for rock mass structure surfaces on open-pit mine slopes specifically includes the following steps S10-S50.

[0076] S10: Obtain a three-dimensional geological outcrop model of the rock mass on the open-pit mine slope.

[0077] In some embodiments, such as Figure 3 As shown, the three-dimensional geological outcrop model of the open-pit mine slope rock mass can be obtained through the following steps S101-S102.

[0078] S101: Use drones for data collection. Set the drone flight altitude according to the accuracy required for the actual application, set the drone flight path according to the size of the target area, use terrain-following oblique photography, set the drone's forward overlap and lateral overlap to meet the requirements, and obtain high-precision image data of the target rock mass.

[0079] First, the flight path of the UAV is planned based on the scope of the target open-pit mine slope and the required modeling accuracy. In this embodiment, to achieve centimeter-level or even millimeter-level ground resolution, the UAV flight altitude is set to 50-100 meters relative to the slope. A terrain-following oblique photography mode is adopted, meaning the UAV follows the terrain undulations in real time during flight, maintaining a relatively constant altitude relative to the slope to ensure image scale consistency. During flight path planning, the forward overlap is set to be no less than 80%, and the lateral overlap no less than 70%, to ensure sufficient feature matching points between adjacent images. The UAV is equipped with a five-lens oblique photography camera, simultaneously acquiring images from one vertical and four oblique angles, thereby comprehensively obtaining the texture and geometric information of the top, sides, and complex uneven areas of the slope rock mass. Flight operations should be conducted in well-lit conditions with no strong winds or precipitation to ensure image clarity and flight safety. After acquisition, a high-overlap, high-resolution real-scene image dataset covering the entire target area is obtained.

[0080] S102: Based on the acquired high-precision image data, a high-precision three-dimensional geological outcrop model is generated using photogrammetric modeling methods. This model has realistic spatial geometric relationships and texture information.

[0081] The high-resolution image data acquired in step S101 is imported into professional photogrammetric modeling software (such as ContextCapture, Photoscan, etc.). First, the software automatically calculates the precise exterior orientation elements (position and attitude) and sparse 3D point cloud for each image by extracting and matching feature points. This process is called aerial triangulation. Then, based on a dense matching algorithm, the sparse point cloud is encrypted to generate high-density 3D point cloud data. This point cloud accurately records the spatial coordinates of every sampling point on the slope surface. On this basis, a high-precision 3D geological outcrop model is finally generated by constructing a triangular mesh (TIN) and mapping image textures. This model not only possesses realistic 3D spatial geometry (i.e., any point in the model has precise X, Y, Z coordinates) but also incorporates color and texture information consistent with the actual ground conditions, enabling it to realistically reproduce the surface features of the slope rock mass and providing a reliable data foundation for subsequent structural surface identification.

[0082] S20: Perform two-dimensional projection on different regions of the three-dimensional geological outcrop model to generate corresponding two-dimensional projection images, annotate the structural surface contours on the two-dimensional projection images, construct a training dataset, and train a semantic segmentation model for intelligent recognition of structural surfaces based on the training dataset.

[0083] In this embodiment, the step surface on the high-precision three-dimensional geological outcrop model obtained in step S10 is projected to obtain several two-dimensional step images. The structural surfaces on them are annotated at the pixel level to form semantic segmentation training samples. A semantic segmentation model dataset D is established. Based on the dataset D, a deep learning semantic segmentation network is used for training. Through training, a semantic segmentation model for intelligent recognition of structural surfaces is obtained.

[0084] In some embodiments, such as Figure 4 As shown, the specific training process of this semantic segmentation model includes the following steps S201-S203.

[0085] S201: Obtain images and labels from the semantic segmentation model dataset, select a semantic segmentation model, and fix the random seed so that the same results can be obtained in each independent training.

[0086] For example, the two-dimensional projected images and their corresponding pixel-level labeled images required for training are obtained from the constructed dataset D. In the labeled images, the pixel values ​​of structural surface regions are set to 1 (foreground), and the pixel values ​​of non-structural surface regions are set to 0 (background), forming a one-to-one input-label pair. When loading data, an online data augmentation strategy is adopted to simultaneously perform random rotation, horizontal flipping, color jittering, and other transformations on the original images and labels to expand the diversity of training samples and improve the generalization ability of the model. In this embodiment, DeepLabV3+ is selected as the semantic segmentation model. This model uses ResNet101 as the backbone network and can effectively capture multi-scale contextual information. To ensure the reproducibility of the experiment, the random seed is fixed in the code (e.g., set to 42) to ensure that the initialization of model parameters, the shuffle order of the dataset, and the results of random data augmentation operations are completely consistent in each independent training, thereby obtaining reproducible training results.

[0087] S202: Set the downsampling factor and the size of the input image, select the optimizer type, set the maximum and minimum learning rates of the model, and use weight decay to prevent overfitting.

[0088] Before model training, a series of hyperparameters need to be configured. In this embodiment, considering the memory capacity and the model's receptive field, the size of the input image is uniformly adjusted to 512×512 pixels, and the downsampling factor of the backbone network is set to 16 times to balance computational efficiency and semantic information of the feature maps. The optimizer AdamW is selected, with an initial learning rate of 0.0001 and a multinomial decay strategy adopted to gradually reduce the learning rate to the minimum learning rate of 0.000001 during training. To prevent overfitting, the weight decay coefficient is set to 0.0001 in the optimizer, and L2 regularization constraint is applied to the network weights. In addition, to monitor the training status, the model's mean intersection-over-union (mIoU) is evaluated on the validation set every 100 batches, and the optimal model weights are dynamically saved based on the validation set performance.

[0089] S203: Assign different loss weights to different types to prevent imbalance between positive and negative samples, and set the total number of training generations.

[0090] To address the potential severe imbalance between the number of structural surface pixels (positive samples) and background pixels (negative samples), a weighted cross-entropy loss function is employed for optimization. Specifically, based on the statistical ratio of foreground to background pixels in the training dataset, a loss weight of 1.0 is assigned to the background category, while the loss weight for the foreground (structural surface) category is set to the ratio of the total number of background pixels to the total number of foreground pixels (e.g., when background:foreground = 100:1, the foreground weight is set to 100), thereby increasing the model's focus on the sparse structural surface pixels. The total number of training generations (epochs) is set to 200, with each generation traversing all training samples once. After each training generation, the model performance is evaluated using a validation set. If the mIoU metric on the validation set does not improve for 20 consecutive generations, training is terminated early to avoid overtraining. After training, the optimal model weight file from the validation set is saved for subsequent slope projection image recognition.

[0091] It should be noted that the semantic segmentation model used for intelligent recognition of structural surfaces is not limited to a specific network structure. The semantic segmentation model can be a semantic segmentation model based on convolutional neural networks, a semantic segmentation model based on Transformer, or a combination of both, such as UNet, PSPNet, HRNet, DeepLab series networks, or improved models thereof.

[0092] This embodiment transforms the problem of identifying complex three-dimensional structural surfaces into a pixel-level classification problem of two-dimensional images by projecting a three-dimensional geological outcrop model into a two-dimensional image, thereby significantly reducing computational complexity and improving recognition efficiency. The specific implementation of the semantic segmentation model does not affect the overall technical solution of this application.

[0093] S30: Using the trained semantic segmentation model, identify the two-dimensional projection image of the target open-pit mine slope to obtain the two-dimensional structural surface recognition result.

[0094] In this embodiment, based on the trained semantic segmentation model, the projected image of the open-pit mine slope is identified to obtain structural surface recognition results. By setting a confidence threshold, low-confidence recognition results are eliminated to improve the accuracy of structural surface recognition, thus optimizing the structural surface recognition results. By magnifying or projecting the recognition area of ​​the open-pit mine slope steps from multiple angles, the number of structural surfaces identified is increased from different perspectives of the same area.

[0095] S40: Based on the projection parameters of the two-dimensional projected image, extract the structural surface trace contour from the two-dimensional structural surface recognition result, and map the structural surface trace contour to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface.

[0096] In this embodiment, based on the viewpoint parameters of the two-dimensional projection image, the trace contour of the structural surface identification area in the image is extracted and mapped to the three-dimensional geological outcrop model, so that the structural surface obtains three-dimensional spatial attributes.

[0097] S50: Clean and merge the obtained three-dimensional spatial structure surfaces, and calculate the normal vectors of the processed structure surfaces to obtain the three-dimensional spatial attribute information of the dip and tilt angle of the three-dimensional spatial structure surfaces.

[0098] In this embodiment, step S50 can be based on a self-developed structural surface cleaning algorithm to merge duplicate structural surfaces and remove non-diffuse structural surfaces. The normal vectors of the structural surfaces are calculated to obtain information such as the dip direction, dip angle, and spacing of the structural surfaces. The number of rock mass volume joints is calculated based on the spacing of the structural surfaces; the rock mass structural grade is calculated based on the number of rock mass volume joints; and the rock mass quality is analyzed based on the rock mass structural grade.

[0099] In some embodiments, such as Figure 5 As shown, the specific process of this structural surface cleaning algorithm includes the following steps S401-S405.

[0100] S401: For each set of structural surface points, use the Singular Value Decomposition (SVD) method to calculate its best-fit plane, and obtain the normal vector of the plane and the plane equation of the structural surface.

[0101] like Figure 6 As shown, the Singular Value Decomposition (SVD) method includes the following steps S4011-S4016.

[0102] S4011: Suppose a certain structural surface is composed of N points. Obtain the spatial coordinates of the N points, expressed as:

[0103] P i =(x i ,y i ,z i ), i=1,2,…,N (1)

[0104] Among them, P i Let x be the spatial coordinates of the i-th point. i ,y i ,z i , where are the X-axis, Y-axis, and Z-axis coordinates of the i-th point, respectively, and N is the total number of points contained in the structural surface.

[0105] S4012: Calculate the geometric center (centroid) of the point set on the structural surface. The calculation formula is as follows:

[0106] , , (2)

[0107] (3)

[0108] in, The geometric center of the set of points on the structural surface. , , These represent the arithmetic mean of the X, Y, and Z coordinates of all points, respectively.

[0109] S4013: Subtract the geometric center (centroid) from each point to construct the decentralized point set matrix A:

[0110] (4)

[0111] in, It is an N×3 matrix.

[0112] S4014: Perform singular value decomposition on matrix A to obtain:

[0113] (5)

[0114] in, C is an N×N orthogonal matrix; C is an N×3 diagonal matrix whose diagonal elements are singular values, namely the first singular value σ1, the second singular value σ2, and the third singular value σ3, where σ1≥σ2≥σ3≥0; V is a 3×3 orthogonal matrix; T is the matrix transpose operator.

[0115] S4015: The column vectors of matrix V correspond to the principal directions of the point set, where the right singular vector V3 corresponding to the minimum singular value σ3 is the normal vector of the best-fit plane of the structure surface point set:

[0116] (6)

[0117] in, , , These are the components of the normal vector along the X, Y, and Z axes, respectively.

[0118] S4016: Using the normal vector n and the centroid P, construct the plane equation of the structural surface:

[0119] (7)

[0120] in, , and Let be the coordinate variable of any point in the plane equation.

[0121] Or equivalently represented as:

[0122] (8)

[0123] (9)

[0124] in, is a constant term in the plane equation.

[0125] It should be noted that the singular value decomposition method used in this embodiment for plane fitting of the structural surface point set is only a preferred implementation. Without departing from the technical concept of this invention, the least squares plane fitting method, the Random Sample Consensus Algorithm (RANSAC), or other equivalent plane fitting algorithms can also be used to obtain the normal vector and plane parameters of the structural surface.

[0126] S402: Calculate the perpendicular distance from all points on each structural surface to the plane, determine whether it is an skewed or twisted surface, and delete it if it is.

[0127] In this embodiment, any point P i (x i ,y i ,z i Vertical distance from the structural plane The calculation formula is:

[0128] (10)

[0129] Through the All data are statistically analyzed to determine whether the structural surface is non-linear or twisted. The method for determining whether a surface is non-linear or twisted is as follows: if the distance from a point to the fitted plane is greater than a preset first distance threshold, the structural surface is determined to be a non-linear or twisted structural surface.

[0130] S403: By calculating the angle between the normal vectors of the two structural surfaces and the distance between the planes, determine whether the two structural surfaces are coplanar; if they are coplanar, construct a local coordinate system, project the two structural surfaces onto a two-dimensional plane, perform geometric operations in the two-dimensional plane to ensure that a single closed polygon is generated, and remap the merged two-dimensional structural surfaces back to three-dimensional space to form the merged three-dimensional structural surfaces.

[0131] S404: Perform a non-plane recheck on the merged structural surfaces. Repeat the point-to-plane distance calculation on the merged structural surfaces. If non-plane or twisted situations still exist, delete them. Finally, retain the structural surfaces that meet geometric consistency.

[0132] S405: For each cleaned 3D structural surface, spatial parameters of the structural surface are calculated based on its normal vector, including the surface dip, dip angle, average spacing, and attitude parameters, thereby obtaining a large amount of highly accurate 3D structural surface information. The specific calculation formula is as follows:

[0133] Dip calculation: The dip direction of a structural surface is the projection direction of its maximum tilt direction onto the horizontal plane. Based on the normal vector components of the fitted plane, its horizontal projection direction angle can be calculated using the arctangent function.

[0134] (11)

[0135] in: The structural plane dips (°). , These are the components of the normal vector in the x and y directions.

[0136] To ensure the dip angle is within the range of 0°–360°, the result needs to be corrected based on the quadrant relationship of the normal vector components. Finally, the true dip angle of the structural surface on the horizontal plane is obtained.

[0137] Inclination angle calculation: The inclination angle (Dip) of a structural surface is the angle between the structural surface and the horizontal plane, which can be calculated from the vertical component of the normal vector.

[0138] (12)

[0139] in: The slope is the inclination angle of the structural surface (°). This represents the vertical component of the normal vector.

[0140] Average spacing: The spacing between structural surfaces is an important parameter describing the spatial distribution characteristics of structural surfaces. For the same set of nearly parallel structural surfaces, the normal distance between the structural surfaces can be calculated based on their plane equations.

[0141] Let the plane equations of two adjacent structural surfaces be:

[0142] (13)

[0143] The spatial distance between the two parallel structural planes is:

[0144] (14)

[0145] Where: d is the spacing between structural surfaces. These are the components of the plane normal vector. These are the constant terms of the two plane equations.

[0146] For the same set of structural surfaces, the average spacing of the set of structural surfaces can be obtained by calculating the normal distance between adjacent structural surfaces and taking their average value.

[0147] Through steps S10-S50 above, this application enables automated, batch, and high-precision extraction of structural surfaces in complex open-pit mine slopes and steep rock slopes. Specifically, the combination of multi-view 2D projection and semantic segmentation improves the recognition coverage of incomplete exposed areas of the structural surfaces; the structural surface cleaning and merging algorithm effectively eliminates non-linear and noisy structural surfaces, improving the spatial consistency and geometric reliability of the final structural surfaces. Based on the obtained 3D structural surface spatial parameters, indicators such as the number of rock mass joints and structural grade can be further calculated, providing reliable data support for slope stability analysis and engineering decisions.

[0148] Figure 7 The image shows the effect of the intelligent identification and extraction method for three-dimensional structural surfaces of open-pit mine slope rock mass provided in the embodiments of this application. Figure 7 As shown, after acquiring a two-dimensional projection image of the open-pit mine slope, a trained semantic segmentation model is used to identify the structural surfaces in the image, resulting in an initial identification result consisting of several red areas. Based on this, multiple-view projections and identifications are performed on the same location to obtain structural surface identification results from different perspectives. The identification results for this area are then mapped and superimposed according to the projection perspective to obtain the overall identification result for all perspectives. After performing three optimization operations—non-linear surface removal, duplicate structural surface merging, and non-linear surface removal—on this result, the final structural surface identification result for the area is obtained.

[0149] This application also provides an intelligent identification and extraction device for rock mass structure surfaces on open-pit mine slopes, such as... Figure 8 As shown, the intelligent identification and extraction device for the rock mass structure surface of the open-pit mine slope includes:

[0150] Data acquisition module 801 is configured to acquire a three-dimensional geological outcrop model of the open-pit mine slope rock mass;

[0151] The model building module 802 is configured to perform two-dimensional projection on different regions of the three-dimensional geological outcrop model, generate corresponding two-dimensional projection images, annotate the structural surface contours on the two-dimensional projection images, build a training dataset, and train a semantic segmentation model for intelligent recognition of structural surfaces based on the training dataset.

[0152] The structural surface recognition module 803 is configured to use a trained semantic segmentation model to recognize the two-dimensional projection image of the target open-pit mine slope and obtain the two-dimensional structural surface recognition result.

[0153] The trace contour mapping module 804 is configured to extract the structural surface trace contour from the two-dimensional structural surface recognition result based on the projection parameters of the two-dimensional projection image, and map the structural surface trace contour to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface.

[0154] The structural surface information calculation module 805 is configured to clean and merge the obtained three-dimensional spatial structural surfaces, and to calculate the normal vectors of the processed structural surfaces to obtain the three-dimensional spatial attribute information of the three-dimensional spatial structural surfaces, including their dip and tilt angles.

[0155] This application provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus.

[0156] The processor executes computer execution instructions stored in memory, causing the processor to perform the scheme in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0157] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0158] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.

[0159] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes described in the above embodiment.

[0160] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the intelligent identification and extraction method for rock mass structure surface of open-pit mine slope in the above embodiments.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0162] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0163] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0164] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0165] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0166] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0167] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Architecture (EISA) buses, etc. Buses can be categorized into address buses, data buses, control buses, etc.

[0168] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0169] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.

[0170] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent identification and extraction of rock mass structural surfaces on open-pit mine slopes, characterized in that, The method includes: Obtain a three-dimensional geological outcrop model of the rock mass on the open-pit mine slope; Two-dimensional projections are performed on different regions of the three-dimensional geological outcrop model to generate corresponding two-dimensional projection images. The structural surface contours on the two-dimensional projection images are labeled to construct a training dataset. A semantic segmentation model for intelligent recognition of structural surfaces is trained based on the training dataset. Using the trained semantic segmentation model, the two-dimensional projection image of the target open-pit mine slope is identified to obtain the two-dimensional structural surface recognition result; Based on the projection parameters of the two-dimensional projected image, the structural surface trace contours in the two-dimensional structural surface recognition results are extracted, and the structural surface trace contours are mapped to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface. The obtained three-dimensional spatial structure surfaces are cleaned and merged, and the normal vectors of the processed structure surfaces are calculated to obtain the three-dimensional spatial attribute information of the dip and tilt angle of the three-dimensional spatial structure surfaces. The three-dimensional structural surfaces are cleaned and merged, specifically including: For each set of three-dimensional structural surface points, perform plane fitting, calculate the structural surface normal vector, and calculate the distance from each point in the point set to the fitting plane. If there is a point whose distance to the fitting plane is greater than a preset first distance threshold, then the structural surface is determined to be an irregular structural surface or a twisted structural surface and is deleted. The coplanarity of two structural surfaces is determined based on the angle between their normal vectors and the distance between their planes. When the angle between their normal vectors is less than a preset angle threshold and the distance between their planes is less than a preset second distance threshold, the two structural surfaces are determined to be coplanar. The coplanar structural surfaces are then merged. The merging process includes: constructing a local two-dimensional coordinate system, projecting the multiple structural surfaces to be merged onto a two-dimensional plane for geometric merging, generating a single closed polygon, and then converting it back to three-dimensional space. The merged structural surfaces are fitted with planes and distances are checked again. If there is a point whose distance to the fitted plane is greater than a preset first distance threshold, the merged structural surface is deleted. For each set of points on the 3D structural surface, perform plane fitting, calculate the normal vector of the structural surface, and calculate the distance from each point in the set to the fitted plane. Specifically, this includes: Obtain the structural surface point set, which consists of all points on the structural surface, and the spatial coordinates of the points are represented as follows: P i =( x i , y i , z i ), i =1,2,…, N （1） in, P i For the first i The spatial coordinates of the points x i , y i , z i The first i The X, Y, and Z coordinates of each point N This represents the total number of points contained within the structural surface. The geometric center of the point set of the structural surface is calculated using the following formula: (2) (3) in, The geometric center of the set of points on the structural surface. Let X, Y, and Z represent the arithmetic mean of the X, Y, and Z coordinates of all points, respectively. Subtract the geometric center from each point to construct the decentralized point set matrix A, which is represented as: （4） in, It is an N×3 matrix; Performing singular value decomposition on the point set matrix A yields: （5） in, C is an N×N orthogonal matrix; C is an N×3 diagonal matrix whose diagonal elements are singular values, namely the first singular value σ1, the second singular value σ2, and the third singular value σ3, where σ1≥σ2≥σ3≥0; V is a 3×3 orthogonal matrix. T This is the matrix transpose operator; The right singular vector corresponding to the third singular value σ3 The normal vector of the plane that best fits the set of structural points , is represented as: （6） in, These are the components of the normal vector along the X, Y, and Z axes, respectively. Using the normal vector And the geometric center P, construct the plane equation of the structural surface, expressed as: （7） in, Let be the coordinate variable of any point in the plane equation; Alternatively, the planar equation of the structural surface can be equivalently expressed as: in, This is the constant term in the plane equation; Calculate any point P i ( x i , y i , z i Vertical distance from the structural plane The calculation formula is: （10）。 2. The method according to claim 1, characterized in that, A three-dimensional geological outcrop model of the open-pit mine slope rock mass was obtained using the following method: Using oblique photography by drones, real-world geological image data of the target open-pit mine slope rock mass are collected by flying in a terrain-following manner, and the drone flight path, heading overlap and lateral overlap are set according to the target area range; Based on the real-scene geological image data, the three-dimensional geological outcrop model is constructed.

3. The method according to claim 1, characterized in that, The method of generating corresponding two-dimensional projection images by performing two-dimensional projection on different regions of the three-dimensional geological outcrop model includes: dividing the three-dimensional geological outcrop model into regions according to the step height, slope normal, or spatial position, and performing two-dimensional projection on the step surfaces of each region obtained by the region division; wherein, the two-dimensional projection includes one or more of orthographic projection, oblique projection, or perspective projection, and at least two different perspectives are used to generate two-dimensional projection images for the same region.

4. The method according to claim 1, characterized in that, The semantic segmentation model is a pixel-level classification model based on convolutional neural networks, Transformer networks, or a combination of both. During the training process, the semantic segmentation model uses a weighted cross-entropy loss function, a Dice loss function, or a combination thereof, and sets different loss weights for different categories to alleviate the sample imbalance problem.

5. The method according to claim 1, characterized in that, After using the trained semantic segmentation model to identify the two-dimensional projection image of the target open-pit mine slope and obtain the two-dimensional structural surface recognition result, the method further includes: setting a confidence threshold for the two-dimensional structural surface recognition result, and the recognition result below the confidence threshold is not included in the subsequent three-dimensional mapping processing.

6. An intelligent identification and extraction device for rock mass structure surfaces on open-pit mine slopes, used to implement the method as described in any one of claims 1-5, characterized in that, The device includes: The data acquisition module is configured to acquire a three-dimensional geological outcrop model of the open-pit mine slope rock mass; The model building module is configured to perform two-dimensional projection on different regions of the three-dimensional geological outcrop model, generate corresponding two-dimensional projection images, annotate the structural surface contours on the two-dimensional projection images, build a training dataset, and train a semantic segmentation model for intelligent recognition of structural surfaces based on the training dataset. The structural surface recognition module is configured to use a trained semantic segmentation model to recognize the two-dimensional projection image of the target open-pit mine slope and obtain the two-dimensional structural surface recognition result. The trace contour mapping module is configured to extract the structural surface trace contour from the two-dimensional structural surface recognition result based on the projection parameters of the two-dimensional projected image, and map the structural surface trace contour to the three-dimensional geological outcrop model to obtain the three-dimensional spatial structural surface. The structural surface information calculation module is configured to clean and merge the obtained three-dimensional spatial structural surfaces, and calculate the normal vectors of the processed structural surfaces to obtain the three-dimensional spatial attribute information of the three-dimensional spatial structural surfaces, including their dip and tilt angles.

7. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the intelligent identification and extraction method for rock mass structure surfaces of open-pit mine slopes as described in any one of claims 1-5.

Images