A method and system for underground mine surveying and resource reserve estimation

Abstract

The application discloses a kind of underground mine surveying and resource reserve estimation method and system, the method includes S1, obtains the point cloud data of underground space of mine in different period and the panoramic image of underground environment of mine;S2, the panoramic image and point cloud data of the same period are preprocessed, the mapping relationship between point cloud data and panoramic image is established, and the point cloud data set after coloring and lightweight three-dimensional model are obtained;S3, the point cloud data set after coloring is carried out ore block sample feature extraction, and the boundary of ore body is circled;S4, according to the boundary of the ore body circled to construct ore body block grid model;S5, repeat steps S2-S4, and obtain the ore body block grid model in different periods;S6, the ore body block grid model in different periods is superimposed, and the ore body change three-dimensional model is obtained;S7, according to the ore body change three-dimensional model to estimate the volume of ore body increase and decrease change, that is, the change result of underground mine resource reserve.This application can be more accurately estimated the volume of ore body increase and decrease change.

Description

Technical Field

[0001] This invention relates to prediction technology, specifically to a method and system for underground mine surveying and resource reserve estimation. Background Technology

[0002] In underground mining operations, accurately measuring resource distribution and estimating reserves is crucial for sustainable development, safe production, and improved economic efficiency. However, current measurement and reserve estimation methods have many shortcomings. Traditional measurement methods, such as total station surveying, are not only inefficient but also struggle to acquire comprehensive and accurate data in complex underground environments, such as narrow tunnels, irregular goafs, and ore bodies, easily leading to measurement errors. Single geological exploration techniques also cannot fully reveal the complex spatial morphology and occurrence patterns of ore bodies, resulting in inaccurate reserve calculations. As mining expands in depth and breadth, the requirements for measurement precision and reserve estimation accuracy become even higher. Existing research has applied laser scanners to underground mine surveying, but point cloud data acquired based on SLAM algorithms lacks color information, making it impossible to identify and delineate ore bodies. This limits its application to measuring the spatial extent of goafs and cannot be extended to areas such as mineral reserve calculation, resource monitoring, and digital mine construction. Under complex geological conditions, existing technologies are insufficient to meet the needs of intelligent mining and refined management. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for underground mine measurement and resource reserve estimation, so as to more accurately estimate the volume of ore body increase or decrease.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows:

[0005] In a first aspect, the present invention provides a method for underground mine surveying and resource reserve estimation, characterized in that it includes:

[0006] S1, acquire point cloud data of underground space in mines at different times and panoramic images of underground environment in mines;

[0007] S2, preprocess the panoramic image and point cloud data of the same period, establish the mapping relationship between the point cloud data and the panoramic image, assign the color information in the panoramic image to the corresponding point cloud data, and obtain the colored point cloud dataset and generate a lightweight real scene model.

[0008] S3, extract ore block sample features from the colored point cloud dataset, and perform color filtering and feature matching on the mine point cloud data based on the extracted ore block sample features to identify areas that may be ore bodies and delineate the boundaries of the ore bodies;

[0009] S4, Construct a grid model of the ore body based on the delineated ore body boundaries;

[0010] S5. Repeat steps S2-S4 to obtain ore body block grid models at different times.

[0011] S6, superimpose the ore body grid models from different periods to obtain a three-dimensional model of ore body variation;

[0012] S7. Estimate the volume of the ore body increase or decrease based on the three-dimensional model of the ore body change.

[0013] Optionally, step S2 includes:

[0014] S21: Denoise, correct, and enhance the panoramic image; filter, downsample, and transform the point cloud data; obtain the preprocessed panoramic image and point cloud data.

[0015] S22, perform key point detection on the preprocessed panoramic image and point cloud data respectively to select key points and calculate the feature descriptor of the selected key points;

[0016] S23, Initial matching is performed by calculating the distance metric of feature descriptors, and then the RANSAC algorithm is used to remove mismatched point pairs to obtain reliable matching point pairs;

[0017] S24, calculate the transformation matrix based on reliable matching point pairs and establish the mapping relationship between point cloud data and panoramic images;

[0018] S25, extract color information from the panoramic image according to the mapping relationship and assign it to the corresponding point cloud data to obtain the colored point cloud dataset;

[0019] S25', based on 3D Gaussian sputtering technology, takes points in point cloud data as the center, determines the Gaussian kernel according to the distribution of surrounding points and the color information of panoramic images, adjusts parameters to control the color diffusion range and intensity, and assigns the color information of panoramic photos to point cloud data, thereby realizing point cloud coloring and generating a lightweight real-world model.

[0020] Optionally, the step of using the SIFT algorithm to detect key points on the preprocessed panoramic image includes:

[0021] Panoramic images are convolved using Gaussian filters at different scales, and the Gaussian differences at different scales are obtained.

[0022] DoG is used to identify key points. The formula for calculating the Gaussian difference (DoG) is:

[0023] D(x,y,σ)=G(x,y,k i σ)-G(x,y,k j σ)

[0024] In the formula, G(x,y,σ) represents the Gaussian function, (x,y) is the coordinate position in the image, σ is the scale parameter, and k is the coordinate position in the image. i and k j These are different scale ratios;

[0025] The SIFT algorithm uses information about pixels near the keypoints, the size of the keypoints, and the principal curvature of the keypoints to locate each keypoint.

[0026] Optionally, the key point detection of the preprocessed point cloud data includes:

[0027] For each point in the preprocessed point cloud data, its normal vector is estimated by fitting a local plane. Based on the point's neighborhood information, the covariance matrix of the neighboring points is calculated, and then eigenvalue decomposition is performed on the covariance matrix. The eigenvector corresponding to the smallest eigenvalue is the estimated normal vector of that point: Let point p i The neighborhood set of N is i covariance matrix

[0028]

[0029] Where u is the centroid of the neighborhood point set, and the eigenvalue decomposition of C is C = U∧U. T The eigenvector u corresponding to the smallest eigenvalue min It's just point P. i The normal vector;

[0030] Let point p i Its neighboring point p j The normal vectors are n i and n j Then point p i The curvature estimate is represented by k. i Select points with curvature values ​​greater than a threshold as key points:

[0031]

[0032] Optionally, the step of using the RANSAC algorithm to remove mismatched point pairs includes:

[0033] Randomly select N non-collinear points from the initial matching point dataset as samples, and use these N sample points to calculate the homography matrix: N is a positive integer;

[0034] Calculate the distance from other points in the initial matching point dataset to the fitted model. If the distance is less than a set threshold, then these points are determined to be interior points.

[0035] Repeat the above two steps multiple times, updating the interior point set M times in each iteration, and selecting the model with the most interior points as the optimal model.

[0036]

[0037] In the formula, z represents the probability that RANSAC obtains the correct model, p represents the proportion of inliers in the data, and k represents the number of points selected for fitting the model.

[0038] Optionally, step S3 includes:

[0039] S31. Extract features of mineral block samples from the colored point cloud dataset, including geometric features, reflection intensity features, and color features, and construct a feature database through learning and analysis of known mineral block samples;

[0040] S32, Analyze the characteristic distribution of ore block samples and surrounding rock samples to determine the classification threshold;

[0041] S33, based on the constructed feature database, uses machine learning algorithms to perform color filtering and shape matching on point cloud data, and preliminarily filters areas that may be ore bodies according to classification thresholds. Then, it uses feature matching algorithms to compare with the features of known ore block samples to identify areas that may be ore bodies and delineate the boundaries of ore bodies.

[0042] Optionally, in step S33, the boundary of the ore body is delineated in the following manner:

[0043] By comparing and analyzing point cloud data from different periods, the location and morphological changes of the ore body can be identified, and the boundaries of the ore body can be delineated.

[0044] Optionally, point cloud data of the underground space of the mine at different periods and panoramic images of the underground environment of the mine can be obtained through the following methods:

[0045] At different times, a mobile laser scanner was used to acquire point cloud data of the underground space of the mine, while a panoramic camera was used to capture panoramic images of the underground environment of the mine. A searchlight with supplementary light source was used for illumination, and the mobile laser scanner and the panoramic camera operated synchronously.

[0046] Optionally, step S6 includes:

[0047] S61, register the ore body grid models from different periods to make the ore body grid models from different periods spatially aligned.

[0048] S62, superimposes the geometric structure and attribute information of the ore body grid model at different times, and retains the change information of the ore body at different times to complete the model superposition.

[0049] S63, after completing the model superposition, performs a closed modeling operation to make the superimposed model form a complete and closed three-dimensional model of ore body variation.

[0050] Secondly, the present invention provides an underground mine surveying and resource reserve estimation system, comprising:

[0051] The data acquisition module is used to acquire point cloud data of the underground space of the mine and panoramic images of the underground environment of the mine at different times.

[0052] The point cloud color assignment and processing module is used to preprocess the panoramic image and point cloud data at the same time, establish the mapping relationship between the point cloud data and the panoramic image, assign the color information in the panoramic image to the corresponding point cloud data, and obtain the color-assigned point cloud dataset.

[0053] The ore body feature extraction and identification module extracts ore block sample features from the colored point cloud dataset. Based on the extracted ore block sample features, it performs color filtering and feature matching on the mine point cloud data to identify areas that may be ore bodies and delineate the boundaries of the ore bodies.

[0054] The reserve volume estimation module is used to construct a ore body grid model based on the delineated ore body boundary, and obtain ore body grid models for different periods based on panoramic images and point cloud data from different periods; the ore body grid models from different periods are superimposed to obtain a three-dimensional model of ore body changes; and the volume of ore body increase or decrease is estimated based on the three-dimensional model of ore body changes.

[0055] Optionally, the underground mine surveying and resource reserve estimation system also includes a human-machine interaction module, used to display point cloud data, panoramic images, ore body identification results and reserve estimation results through three-dimensional graphics and images.

[0056] Compared with the prior art, the advantages of this invention are as follows:

[0057] Compared with existing technologies, the present invention provides a method and system for estimating mineral resource reserves, which has the following advantages:

[0058] (1) Improve the visualization of measurement

[0059] By integrating mobile laser scanners and panoramic cameras with auxiliary lighting, high-quality colorization of point clouds in the dark environment of underground mines has been achieved. The generated true-color structured point clouds can intuitively display the underground environment of the mine, providing mine workers with clearer and more realistic information about underground space, which helps to quickly identify features and improve work efficiency.

[0060] (2) Improve the accuracy of ore body identification and delineation

[0061] In underground mine reserve monitoring, accurately identifying and distinguishing between ore bodies and surrounding rock is crucial, which differs from open-pit mines. In open-pit mines, distinguishing between ore bodies and surrounding rock is relatively easy, requiring only visual observation and simple geological analysis for preliminary judgment. However, in underground mines, due to the complex environment and insufficient light, advanced technical means are needed for identification. In this embodiment, through the aforementioned sub-steps, a mapping relationship is established to assign color information from the panoramic image to the point cloud data, resulting in a true-color structured point cloud. During the fusion process, key point detection is performed to calculate the feature descriptors of the key points. Initial matching is then performed based on the distance metric of the feature descriptors to obtain reliable matching point pairs. This accurately establishes the mapping relationship between the point cloud data and the panoramic image, ensuring accurate color assignment and providing support for more accurate delineation of the ore body's boundaries and extent, while also effectively improving work efficiency.

[0062] (3) Enhance the accuracy of reserves estimation

[0063] Changes in ore bodies in open-pit mines are relatively straightforward and easy to monitor. However, changes in ore bodies in underground mines are more complex due to mining activities and geological shifts, requiring more precise monitoring methods. By overlaying ore body block grid models from different periods, a three-dimensional model of ore body changes is obtained. This model is then used to estimate the volume of ore body increases and decreases, thus improving the accuracy of reserve estimation.

[0064] (4) Adaptable to complex underground environments

[0065] Open-pit mines utilize natural light for measurement, while underground mines, in their dark environments, employ supplementary and enhanced light sources like searchlights to ensure panoramic cameras can acquire clear images. This allows the method and system to adapt to complex underground environments, demonstrating strong practicality and versatility. Furthermore, the confined spaces of underground mines make it difficult to use large equipment, necessitating the use of handheld mobile laser scanners and panoramic cameras for measurement. These devices offer flexibility and facilitate comprehensive data acquisition. Attached Figure Description

[0066] Figure 1 The main flowchart of the underground mine surveying and resource reserve estimation method provided in the embodiments of this application;

[0067] Figure 2 The flowchart for the sub-steps of step S2 is shown below.

[0068] Figure 3 The flowchart for the sub-steps of step S3 is shown below.

[0069] Figure 4 The flowchart for the sub-steps of step S6 is shown below.

[0070] Figure 5A complete flowchart of the underground mine surveying and resource reserve estimation method provided in the embodiments of this application;

[0071] Figure 6 Photos of the ore body at the site;

[0072] Figure 7 This is a 3D point cloud classification result image;

[0073] Figure 8 Map showing the delineation of the ore body;

[0074] Figure 9 A map showing the distribution of areas of change in the ore body within the mine;

[0075] Figure 10 A three-dimensional model diagram of the ore body's movement;

[0076] Figure 11 A schematic diagram of the composition of the underground mine surveying and resource reserve estimation system provided in the embodiments of this application. Detailed Implementation

[0077] Example:

[0078] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0079] See Figure 1 As shown, the underground mine surveying and resource reserve estimation method provided in this embodiment mainly includes the following steps:

[0080] S1, Data Acquisition Stage: Acquire point cloud data of the underground space of the mine at different times and panoramic images of the underground environment of the mine.

[0081] S2, Point Cloud Color Assignment and Processing Stage: The panoramic image and point cloud data are preprocessed to establish a mapping relationship between the point cloud data and the panoramic image. The color information in the panoramic image is assigned to the corresponding point cloud data to obtain the colored point cloud dataset and generate a lightweight real-world model.

[0082] Thus, in this step, by fusing and colorizing panoramic images and point cloud data, high-quality colorization of point clouds in the dark environment of underground mines is achieved. The generated true-color structured point clouds can intuitively display the underground environment of the mine, providing mine workers with clearer and more realistic underground space information, which helps to quickly identify ground features and improve work efficiency.

[0083] S3, Ore Body Feature Extraction and Identification Stage: Extract ore block sample features from the colored point cloud dataset. Based on the extracted ore block sample features, perform color filtering and feature matching on the mine point cloud data to identify areas that may be ore bodies and delineate the boundaries of the ore bodies.

[0084] In this way, the boundaries and extent of the ore body can be accurately delineated, providing more accurate guidance for mining operations and reducing resource waste and mining risks.

[0085] S4. Construct a grid model of the ore body based on the delineated ore body boundaries.

[0086] S5. Repeat steps S2-S4 to obtain ore body grid models for different periods.

[0087] In other words, in this step, the point cloud data of the underground space of the mine and the panoramic image of the underground environment of the mine acquired in other periods are repeated to obtain the ore body grid model of different periods, identify the changes in the location and shape of the ore body, and thus monitor the ore body change area in real time, providing timely information for resource management and mining decisions, and ensuring the sustainable mining of the mine.

[0088] S6 superimposes the ore body grid models from different periods to obtain a three-dimensional model of ore body variation.

[0089] Changes in ore bodies in open-pit mines are relatively straightforward and easy to monitor. However, changes in ore bodies in underground mines are more complex due to mining activities and geological shifts, requiring more precise monitoring methods. Therefore, in this step, a three-dimensional model of ore body changes is obtained by overlaying ore body grid models from different periods. This model is then used to estimate the volume of ore body increases and decreases, thus enhancing the accuracy of reserve estimation.

[0090] S7. Estimate the volume of the ore body increase or decrease based on the three-dimensional model of the ore body change.

[0091] Therefore, this method can accurately estimate the volume of ore body increases or decreases.

[0092] In one specific embodiment, point cloud data of the underground space of the mine and panoramic images of the underground environment of the mine are acquired in the following manner:

[0093] A mobile laser scanner with SLAM (Simultaneous Localization and Mapping) capabilities works in conjunction with a panoramic camera. In the complex tunnels and mining areas of underground mines, the mobile laser scanner rapidly acquires high-precision point cloud data, recording the geometry of the underground space and the location of objects. Simultaneously, the panoramic camera captures panoramic images of the surrounding environment, providing rich color information for point cloud colorization. A high-intensity searchlight is also provided as a supplementary light source. Its illumination angle and brightness are flexibly adjusted according to the on-site lighting and measurement requirements, ensuring clear imaging even in the dark environment of underground mines. This allows the colorized point cloud to accurately reflect the color characteristics of the underground environment, improving visualization and facilitating subsequent feature identification and analysis. A synchronous control unit ensures that the mobile laser scanner and panoramic camera operate synchronously in time and space, guaranteeing an accurate correspondence between the point cloud data and the panoramic image.

[0094] In one specific embodiment, step S2 above includes the following sub-steps:

[0095] S21, Denoise, correct and enhance the panoramic image, and filter, downsample and transform the point cloud data to obtain the preprocessed panoramic image and point cloud data.

[0096] S22, perform key point detection on the preprocessed panoramic image and point cloud data respectively to select key points and calculate the feature descriptor of the selected key points;

[0097] S23, Initial matching is performed by calculating the distance metric between feature descriptors, and RANSAC algorithm is used to remove mismatched point pairs to obtain reliable matching point pairs;

[0098] S24, calculate the transformation matrix based on reliable matching point pairs and establish the mapping relationship between point cloud data and panoramic images;

[0099] S25, extract color information from the panoramic image according to the mapping relationship and assign it to the corresponding point cloud data to obtain the colored point cloud dataset;

[0100] S25', based on 3D Gaussian sputtering technology, takes points in point cloud data as the center, determines the Gaussian kernel according to the distribution of surrounding points and the color information of panoramic images, adjusts parameters to control the color diffusion range and intensity, and assigns the color information of panoramic photos to point cloud data, realizing point cloud coloring and generating a lightweight real scene model LCC.

[0101] In underground mine reserve monitoring, accurately identifying and distinguishing between ore bodies and surrounding rock is crucial, which differs from open-pit mines. In open-pit mines, distinguishing between ore bodies and surrounding rock is relatively easy, requiring only visual observation and simple geological analysis for preliminary judgment. However, in underground mines, due to the complex environment and insufficient light, advanced technical means are needed for identification. In this embodiment, through the aforementioned sub-steps, a mapping relationship is established to assign color information from the panoramic image to the point cloud data, resulting in a true-color structured point cloud. During the fusion process, key point detection is performed to calculate the feature descriptors of the key points. Initial matching is then performed based on the distance metric of the feature descriptors to obtain reliable matching point pairs. This accurately establishes the mapping relationship between the point cloud data and the panoramic image, ensuring accurate color assignment and providing support for more accurate delineation of the ore body's boundaries and extent, while also effectively improving work efficiency.

[0102] Specifically, the key point detection process for the preprocessed panoramic image and point cloud data includes:

[0103] The SIFT algorithm is used to detect key points in the preprocessed panoramic image, including:

[0104] Panoramic images are convolved using Gaussian filters at different scales, and the Gaussian differences at different scales are obtained.

[0105] DoG is used to identify key points. The formula for calculating the Gaussian difference (DoG) is:

[0106] D(x,y,σ)=G(x,y,k i σ)-G(x,y,k j σ)

[0107] In the formula, G(x,y,σ) represents the Gaussian function, (x,y) is the coordinate position in the image, σ is the scale parameter, and k is the coordinate position in the image. i and k j These are different scale ratios;

[0108] To obtain the maxima of DoG images at different scales, the original images are first convolved with Gaussian blurs at different scales. These Gaussian blurred images are grouped in units of 2 times their scale, and k... i σ is usually a selected fixed value. The DoG image is obtained by subtracting each pair of adjacent Gaussian blurred images in the same group. The SIFT algorithm uses information about the pixels near the keypoint, the size of the keypoint, and the principal curvature of the keypoint to locate each keypoint, thereby eliminating keypoints located on the edge or susceptible to noise.

[0109] In this way, the key points of the panoramic image can be accurately selected through the above operations.

[0110] Keypoint detection is performed on the preprocessed point cloud data, including:

[0111] Normal vector estimation: For each point in the point cloud data, its normal vector is estimated by fitting a local plane. Based on the point's neighborhood information, the covariance matrix of the neighboring points is calculated, and then eigenvalue decomposition is performed on the covariance matrix. The eigenvector corresponding to the smallest eigenvalue is the estimated normal vector of that point. Let point p... i The neighborhood set of N is i covariance matrix

[0112]

[0113] Where u is the centroid of the neighborhood point set, and the eigenvalue decomposition of C is C = U∧U. T The eigenvector u corresponding to the smallest eigenvalue min It's just point P. i The normal vector;

[0114] Key point selection: Let point p i Its neighboring point p j The normal vectors are n i and n j Then point p i The curvature estimate is represented by k. i Select points with curvature values ​​greater than a threshold as key points;

[0115]

[0116] In this way, the key points of the point cloud data can be accurately selected through the above operations.

[0117] For keypoints in selected panoramic images and point cloud data, their feature descriptors are calculated. Common point cloud feature descriptors include PFH (Point Feature Histogram) or FPFH (Fast Point Feature Histogram). Taking PFH as an example, for a keypoint p, considering the normal vector information of its neighboring points, the angle and distance relationships between each neighboring point and the keypoint are calculated, and a histogram of these relationships within a certain range is formed to create the PFH feature descriptor. Specifically, for a neighboring point q, the distance d = ||pq|| between it and the keypoint p is calculated, and the angle θ = arccos(n) between their normal vectors is calculated. p ·n q Then, the value of (d,θ) is quantized into the corresponding histogram interval to obtain the PFH feature vector.

[0118] Initial matching is performed between key points in panoramic images and point cloud data by calculating distance metrics of feature descriptors (such as Euclidean distance or distance calculation based on point cloud feature descriptors), and then the RANSAC (Random Sample Consensus) algorithm is used to remove mismatched point pairs.

[0119] When applying the RANSAC algorithm, parameters such as the model, the number of randomly sampled points, and the number of iterations need to be specified. Taking the homography matrix as an example, the specific steps of the RANSAC algorithm are as follows:

[0120] (1) Random sampling: Randomly select 4 non-collinear points from the matching point dataset as samples.

[0121] (2) Model fitting: The homography matrix is ​​calculated using these 4 sample points.

[0122] (3) Inner point determination: Calculate the distance from other points in the dataset to the fitted model. If the distance is less than a certain threshold, these points are determined to be inner points.

[0123] (4) Iterative update: Repeat the above steps multiple times (number of iterations M), and update the interior point set after each iteration. Select the model with the most interior points as the optimal model.

[0124] In the RANSAC algorithm, it is usually necessary to estimate the number of iterations M to determine the number of times the algorithm needs to run. The formula for calculating this number is as follows:

[0125]

[0126] In the formula, z represents the probability that RANSAC obtains the correct model (generally required to be greater than 95%), p represents the proportion of inliers in the data (probability of inliers), and k represents the number of points selected for fitting the model.

[0127] Thus, through the above operations, the RANSAC algorithm is used to eliminate mismatched point pairs, thereby obtaining reliable matching point pairs. Based on these reliable matching point pairs, the least squares method is used to calculate the transformation matrix, achieving accurate spatial registration between the point cloud and the panoramic photograph. Then, based on 3D Gaussian sputtering technology, using points in the point cloud data as the center, a Gaussian kernel is determined according to the distribution of surrounding points and the color information of the panoramic photograph. Parameters are adjusted to control the color diffusion range and intensity, imbuing the point cloud data with the color information of the panoramic photograph, achieving point cloud colorization and generating a lightweight real-world model (LCC), improving data processing efficiency and model visualization effects.

[0128] In one specific embodiment, step S3 includes the following sub-steps:

[0129] S31 Feature Extraction and Filtering: Features of mineral block samples are extracted from the colored point cloud dataset, including geometric features (such as shape factor, volume, surface area, texture roughness, etc.), reflection intensity features, and color features (statistical features such as hue, saturation, and brightness). A feature database is constructed through learning and analysis of a large number of known mineral block samples.

[0130] S32 Classification Threshold Calculation: A large number of known ore block samples and surrounding rock samples are collected, their feature distribution is analyzed, and classification thresholds are determined using algorithms based on statistical hypothesis testing or machine learning (such as decision trees and support vector machines). In the mine point cloud data, color filtering and geometric shape matching are performed based on these features to preliminarily identify areas where ore bodies may exist.

[0131] S33 Classification and Recognition: Using machine learning algorithms, such as support vector machine (SVM) and random forest, the point cloud data of the mine is subjected to color screening and shape matching. Based on the classification threshold determined in step S32, areas that may be ore bodies are initially screened. Then, the feature matching algorithm is used to compare with the features of known ore block samples to determine the probability of ore bodies and delineate the boundaries of ore bodies.

[0132] Specifically, the boundaries of ore bodies can be accurately delineated through comparative analysis of data from two or more phases. The classification results are then validated and optimized by comparing them with actual geological exploration data, adjusting feature extraction methods, classification thresholds, or algorithm parameters to improve identification accuracy.

[0133] In one specific embodiment, step S4 includes:

[0134] Based on the defined orebody boundaries, a grid model of the orebody blocks is constructed. Using the Poisson modeling algorithm or other suitable construction methods, the orebody point cloud is converted into a model to accurately describe the surface morphology of the orebody. During the construction process, the grid density and accuracy are adjusted appropriately according to the complexity of the orebody and the required measurement precision to ensure that the model accurately reflects the shape of the orebody. Based on the orebody point cloud over a time range, considering factors such as orebody mining, growth, or geological changes, a three-dimensional model of orebody variation is created to monitor reserve changes in real time. Specifically, the orebody block grid model includes:

[0135] a) Goaf orebody model: Using the Poisson reconstruction algorithm, first, normal vector estimation is performed, and the normal vector is determined by fitting a local plane to neighboring points; then, the Poisson equation is constructed and solved to obtain the function f; finally, isosurfaces are extracted using algorithms such as marching cubes to generate a 3D model. The Poisson modeling algorithm for converting discrete point clouds into grid models includes the following steps:

[0136] Normal vector estimation: For a given point cloud dataset, for each point, search for its neighboring points within a certain radius. Fit a local plane based on these neighboring points, and use the normal vector of this plane as the normal vector of the current point.

[0137] Poisson equation establishment: The calculated normal vector field is used as input to construct the Poisson equation Δf = g, where Δ is the Laplace operator, f is the function to be solved, the normal vector field is transformed into the source function g, and the function f is finally used to define the reconstructed surface.

[0138] Equation Solving and Surface Generation: The Poisson equation is solved numerically to obtain the function f. Using algorithms such as marchingcubes, isosurfaces are extracted based on the value of function f. These isosurfaces are the final Poisson reconstructed surface, completing the conversion from point cloud data to a 3D surface model. OBJ format 3D model files can be generated. The generated mesh model undergoes detail optimization, adjusting mesh resolution and accuracy, and mapping panoramic photo texture information onto the mesh model to enhance its realism.

[0139] b) Dynamic modeling of ore bodies using the non-parallel arbitrary cross section method

[0140] Delineating boundaries by selecting different middle sections of ore bodies: The non-parallel arbitrary section method is an important method for constructing a vertical distribution morphology model of ore bodies.

[0141] The division of intermediate sections is based on the mining plan and geological structure characteristics of the underground mine, which are then vertically divided into multiple intermediate sections. The division of each intermediate section typically considers factors such as the depth and thickness variations of the ore body, as well as the requirements of the mining process. When dividing the intermediate sections, it is essential to ensure that each section possesses relatively independent geological characteristics and mining value, facilitating subsequent ore body delineation and reserve estimation.

[0142] Boundary delineation method: Within each intermediate section, based on the orebody identification method described above, combined with geological exploration data and point cloud data, the boundaries of the orebody are accurately delineated. Multi-period data comparison and analysis are used to determine the location and morphological changes of the orebody at different times, further optimizing the delineation of the orebody boundaries. This method accurately obtains the boundary information of each intermediate section of the orebody, providing fundamental data for constructing a non-parallel arbitrary cross-section model.

[0143] Connecting different mid-section boundaries: After delineating the boundaries of different mid-section ore bodies, these boundaries need to be connected to construct a continuous ore body model.

[0144] Connection Strategies and Algorithms: Appropriate connection strategies and algorithms are employed to connect the boundaries of ore bodies in different sections. Based on the geological strike and spatial distribution of the ore bodies, triangulation algorithms or spline interpolation algorithms can be used to establish connections between boundary points in different sections. During the connection process, the continuity and integrity of the ore bodies are fully considered to ensure that the connected model accurately reflects the true vertical morphology of the ore bodies.

[0145] Geological rationality verification: The geological rationality of the connected boundaries is verified. Combining geological knowledge and actual exploration data, the model is checked to ensure it conforms to the geological characteristics and occurrence patterns of the ore body. If any unreasonable aspects are found in the connected model, such as the presence of faults, folds, or other geologically inconsistent features, the connection strategy and algorithm are adjusted, and the boundaries are reconnected to ensure the geological rationality of the model.

[0146] Connected model construction: After completing the boundary connection, the three-dimensional model of the ore body is constructed by connecting the connected areas.

[0147] Model building algorithms: 3D modeling algorithms, such as mesh-based or voxel-based models, are used to construct a complete 3D model from the connected boundary data. Mesh-based models convert boundary data into triangular meshes to build the surface model of the ore body; voxel-based models divide space into voxel units, filling them with voxels based on the location and properties of the ore body boundaries to build the solid model of the ore body. Appropriate modeling algorithms are selected based on actual needs and data characteristics to ensure that the constructed model accurately reflects the 3D morphology and spatial distribution of the ore body.

[0148] Model Optimization and Display: The constructed model is optimized to improve its quality and visualization. Adjustments such as mesh resolution and surface smoothing make the model smoother and more accurate. Using 3D visualization software, the constructed ore body model is displayed, providing intuitive model support for mine planning, reserve estimation, and geological analysis. During visualization, the model can be rotated, scaled, and sectioned to observe the morphology and structure of the ore body from different angles, providing more comprehensive information for related work.

[0149] In one specific embodiment, step S6 includes:

[0150] S61, register the ore body grid models from different periods to ensure that the ore body grid models from different periods are accurately aligned in space.

[0151] S62 superimposes the geometric structure and attribute information of the ore body grid model from different periods, preserving the change information of the ore body in different periods to complete the model superposition; during the superposition process, the overlapping part of the model is processed according to the actual changes of the ore body to ensure that the superimposed model can accurately reflect the real shape changes of the ore body.

[0152] S63, after model stacking, performs a closure modeling operation to form a complete and closed 3D model of the ore body's deformation. This closure modeling operation includes filling voids in the model and repairing model boundaries to create a complete and closed 3D model. During the closure modeling process, triangulation algorithms and mesh repair algorithms are used to ensure the correct topological structure and reasonable geometry of the model. Through closure modeling, the 3D morphology of the ore body after deformation can be accurately described, providing an accurate model basis for subsequent volume calculations and analysis.

[0153] Finally, after constructing the three-dimensional model of the ore body's deformation, the deformation volume of the ore body was calculated, and the calculation results were verified.

[0154] a) Volume Calculation Method: Based on the characteristics of the 3D model of the ore body deformation, a suitable volume calculation method is selected. The voxel method can be used, dividing the model into several voxel units and calculating the volume by counting the number of voxels containing the ore body; alternatively, the integral method can be used, calculating the volume through integration operations based on the model's geometry and boundary conditions. During the calculation process, the accuracy and precision of the calculation method must be ensured, taking into account the complexity of the model and the accuracy requirements of the data, and the calculation parameters should be selected appropriately.

[0155] b) Result Verification and Analysis: The calculated change volume is verified. This can be done by comparing the results with actual mining volumes, geological estimates, and other data to check the accuracy of the calculations. If a discrepancy is found between the calculated volume and the actual data, the cause should be analyzed in depth. This could be due to inaccurate model construction, inappropriate selection of the volume calculation method, or data errors. Based on the analysis results, the model should be adjusted, a new calculation method should be selected, or the data should be corrected to ensure the accuracy and reliability of the change volume calculation results, providing accurate data support for mine resource management and mining decisions. For the spatial model of reserve changes, a dynamic spatial model of reserve changes is constructed based on the ore body point cloud within the time range, considering factors such as ore body mining, growth, or geological changes. This allows for real-time monitoring of reserve changes, providing timely and accurate data support for mine resource management.

[0156] In addition, to monitor the changes in the ore body in real time, the elevation differences of point clouds from different periods at the same geographical location can be identified, the change values ​​can be quantified, and a rendering distribution model can be constructed. The identified elevation difference values ​​are quantified and a rendering distribution model is constructed to visually display the distribution of the ore body's changing areas, specifically including:

[0157] Point cloud registration and comparison: First, point cloud data acquired at different times are registered to unify them into the same coordinate system, ensuring data comparability. Point cloud registration algorithms, such as ICP (Iterative Closest Point) algorithm, are used to find the correspondence between point clouds from different periods, ensuring precise spatial alignment of points at the same geographical location. Then, the registered point cloud data is compared point by point to calculate the elevation changes of points at the same geographical location at different times.

[0158] Data Processing and Analysis: The calculated elevation change data is processed and analyzed. Filtering algorithms are used to remove noise and outliers to ensure data accuracy. Statistical analysis methods are employed to calculate the mean, standard deviation, and other statistical measures of elevation changes to understand the overall trend and dispersion of ore body movement. Analysis of the elevation change data allows for the determination of ore body movement in different regions, such as subsidence caused by mining or uplift caused by geological changes.

[0159] Quantification methods: Choose an appropriate quantification method based on actual needs and data characteristics. The elevation difference changes can be divided into intervals, such as dividing the elevation difference range into several intervals, each corresponding to a quantified value. Alternatively, normalization methods can be used to map the elevation difference changes to a specific numerical range, facilitating subsequent processing and analysis.

[0160] Rendering Distribution Model Construction: A rendering distribution model is constructed based on quantified elevation changes using rendering technology. Different colors or transparency levels are used in the model to represent varying degrees of elevation change; for example, red indicates areas with significant elevation changes, while blue indicates areas with smaller changes. This method visually displays the distribution of orebody movement areas, providing crucial information for mine management and resource assessment. Furthermore, topographic data and geological information can be incorporated into the rendering distribution model to more realistically and accurately reflect the relationship between orebody movement and the surrounding environment.

[0161] The following application scenario example will be used to further illustrate this method:

[0162] Data collection phase:

[0163] In an underground mine in a certain region, data collection was conducted first. Due to the dim lighting underground, operators carried a mobile laser scanner, a panoramic camera, and a searchlight into the mine. The mobile laser scanner was selected based on its high precision and rapid scanning capabilities, while the panoramic camera featured high resolution and a wide field of view. The brightness and illumination range of the searchlight were chosen according to the actual conditions of the mine.

[0164] Before entering the measurement area, the route was planned based on the tunnel layout and measurement range, and control points were set at key locations. During the measurement, the operators moved at a constant speed, and the mobile laser scanner continuously emitted laser beams to quickly acquire point cloud data of the tunnels, goaf, and ore body, recording spatial geometric information; the panoramic camera simultaneously captured the surrounding environment, color information for coloring the point cloud.

[0165] During the scanning process, the synchronous control unit ensures that the mobile laser scanner and panoramic camera start and stop simultaneously, maintaining the same speed and trajectory. Operators move along a pre-planned route through the mine tunnels and stopes. The mobile laser scanner continuously emits laser beams to acquire the three-dimensional coordinates of surrounding objects, forming point cloud data. The panoramic camera, on the other hand, captures real-time panoramic images of the surrounding environment. By adjusting the angle of the searchlight, the panoramic camera ensures that it clearly captures the color and texture details of the ore body surface, guaranteeing accurate subsequent colorization. The searchlight provides ample light for the panoramic camera, ensuring image clarity.

[0166] Point cloud coloring and processing stage:

[0167] The acquired panoramic images are denoised to remove noise points and interference information; then image enhancement is performed to improve the contrast and clarity of the images.

[0168] A mapping relationship between point cloud data and panoramic images is established using methods such as feature matching and coordinate transformation. Based on this mapping relationship, color information from the panoramic image is accurately assigned to the corresponding point cloud data, achieving colorization of the point cloud and obtaining a true-color structured point cloud.

[0169] Ore body feature extraction and identification stage:

[0170] From the processed point cloud dataset, technicians extracted geometric, reflectance intensity, and color features of ore block samples to construct a feature database. Using machine learning algorithms, color filtering and shape matching were performed on the mine point cloud data to initially identify ore body areas. By comparing point cloud data from different periods, the location changes of the ore body were precisely determined, thus more accurately delineating its boundaries. In one area, through comparison of multiple data periods, a new ore body was discovered. Further analysis of its spatial information provided crucial information for mining planning.

[0171] Mineral block sample features are extracted from the colored point cloud dataset. Geometric features are described by calculating parameters such as the shape factor and volume of the point cloud; color features are obtained by analyzing the color distribution and statistical characteristics of the point cloud; and reflection intensity features are extracted by the reflection signal intensity of the lidar.

[0172] Based on the extracted features, color filtering and feature matching are performed on the mine point cloud data. For example, based on the color characteristics of the ore body, point cloud regions that may be ore bodies are selected. Then, using a feature matching algorithm, the selected regions are compared with the features of known ore block samples to further determine the likelihood of an ore body. After processing, the generated true-color structured point cloud can clearly distinguish between the ore body and the surrounding rock. For example, the copper yellow color unique to copper mines is clearly visible in the point cloud model, greatly improving the visualization effect and facilitating subsequent analysis.

[0173] Point cloud data is acquired at different times through regular data measurements. Comparative analysis of the point cloud data from different periods identifies changes in the location and shape of the ore body. For example, by calculating the displacement and deformation of the point cloud, the mining progress and the emergence of new ore bodies can be determined. Based on the comparative analysis results, the boundaries of the ore body are precisely delineated.

[0174] Reserve volume estimation stage:

[0175] Based on the delineated orebody boundaries, a point cloud Poisson modeling algorithm is used to construct a orebody grid model. This model can accurately describe the surface shape and spatial structure of the orebody. By overlaying orebody grid models from different periods, a three-dimensional model of orebody variation is obtained. Thus, the resulting three-dimensional model of orebody variation considers factors such as the mining process and geological changes, and can reflect the real-time changes in orebody reserves.

[0176] The volume of the ore body changes is calculated using the constructed 3D model of ore body variation. If the calculated volume does not match the ore output, adjustments are made promptly.

[0177] Finally, the calculated ore body volume was compared and verified with the mine's ore output data. The specific calculation results are as follows:

[0178] Model estimated volume: 99.9 cubic meters; the volume of the ventilation opening, approximately 10 cubic meters, needs to be deducted; total: 89.9 cubic meters.

[0179] Ore output: 124 * 0.7 = 86.8 cubic meters.

[0180] The error is less than 5 cubic meters.

[0181] In summary, the underground mine surveying and resource reserve estimation method provided in this embodiment can accurately estimate the ore body volume.

[0182] Accordingly, this embodiment also provides an underground mine surveying and resource reserve estimation system 100, including:

[0183] The data acquisition module 110 is used to acquire point cloud data of underground space in the mine at different times and panoramic images of the underground environment of the mine.

[0184] The point cloud color assignment and processing module 120 is used to preprocess the panoramic image and point cloud data at the same time, establish the mapping relationship between the point cloud data and the panoramic image, assign the color information in the panoramic image to the corresponding point cloud data, and obtain the color-assigned point cloud dataset.

[0185] The ore body feature extraction and recognition module 130 extracts ore block sample features from the colored point cloud dataset, performs color filtering and feature matching on the mine point cloud data based on the extracted ore block sample features, in order to identify areas that may be ore bodies and delineate the boundaries of the ore bodies.

[0186] The reserve volume estimation module 140 is used to construct a ore body grid model based on the delineated ore body boundary, and obtain ore body grid models of different periods based on panoramic images and point cloud data of different periods; superimpose the ore body grid models of different periods to obtain a three-dimensional model of ore body change; and estimate the volume of ore body increase or decrease based on the ore body change three-dimensional model.

[0187] As a preferred embodiment of the underground mine surveying and resource reserve estimation system, it also includes a human-machine interface module 150, used to display point cloud data, panoramic images, ore body identification results, and reserve estimation results through 3D graphics and images. It also supports multiple data visualization methods, such as 3D model display and data report generation, enabling users to quickly understand the measurement and estimation results of mine resources, providing strong support for decision-making, and facilitating operators in equipment control, parameter setting, data viewing, and analysis result display.

[0188] As a preferred embodiment of the underground mine surveying and resource reserve estimation system, it also includes a data storage and management module for storing collected point cloud data, panoramic images, processed results, and model data in a database. The database employs hierarchical storage and index management to improve data storage efficiency and query speed. This module provides data query, management, and backup functions. Staff can input query conditions through the user interface module to quickly obtain the required data; simultaneously, the system regularly backs up the data to prevent data loss.

[0189] It should be noted that the underground mine surveying and resource reserve estimation device provided in this application embodiment can execute the underground mine surveying and resource reserve estimation method provided in any embodiment of this application, and has the corresponding functions and beneficial effects of the execution method. Therefore, it will not be described again in this embodiment.

[0190] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for underground mine surveying and resource reserve estimation, characterized in that, include: S1, acquire point cloud data of underground space in mines at different times and panoramic images of underground environment in mines; S2, preprocess the panoramic image and point cloud data of the same period, establish the mapping relationship between the point cloud data and the panoramic image, assign the color information in the panoramic image to the corresponding point cloud data, and obtain the colored point cloud dataset and generate a lightweight real scene model. S3, extract ore block sample features from the colored point cloud dataset, and perform color filtering and feature matching on the mine point cloud data based on the extracted ore block sample features to identify areas that may be ore bodies and delineate the boundaries of the ore bodies; S4, Construct a grid model of the ore body based on the delineated ore body boundaries; S5. Repeat steps S2-S4 to obtain ore body block grid models at different times. S6, superimpose the ore body grid models from different periods to obtain a three-dimensional model of ore body variation; S7. Estimate the volume of the ore body increase or decrease based on the three-dimensional model of the ore body change. Step S2 includes: S21: Denoise, correct, and enhance the panoramic image; filter, downsample, and transform the point cloud data; obtain the preprocessed panoramic image and point cloud data. S22, perform key point detection on the preprocessed panoramic image and point cloud data respectively to select key points and calculate the feature descriptor of the selected key points; S23, Initial matching is performed by calculating the distance metric of feature descriptors, and then the RANSAC algorithm is used to remove mismatched point pairs to obtain reliable matching point pairs; S24, calculate the transformation matrix based on reliable matching point pairs and establish the mapping relationship between point cloud data and panoramic images; S25, extract color information from the panoramic image according to the mapping relationship and assign it to the corresponding point cloud data to obtain the colored point cloud dataset; The lightweight real-world model is generated as follows: Based on 3D Gaussian sputtering technology, the Gaussian kernel is determined according to the distribution of surrounding points and the color information of the panoramic image, the parameters are adjusted to control the color diffusion range and intensity, and the color information of the panoramic photo is assigned to the point cloud data to achieve point cloud coloring and generate a lightweight real-world model. Step S3 includes: S31. Extract features of mineral block samples from the colored point cloud dataset, including geometric features, reflection intensity features, and color features, and construct a feature database through learning and analysis of known mineral block samples; S32, Analyze the characteristic distribution of ore block samples and surrounding rock samples to determine the classification threshold; S33, based on the constructed feature database, uses machine learning algorithms to perform color filtering and shape matching on point cloud data, and preliminarily filters areas that may be ore bodies according to classification thresholds. Then, it uses feature matching algorithms to compare with the features of known ore block samples to identify areas that may be ore bodies and delineate the boundaries of ore bodies.

2. The method for underground mine surveying and resource reserve estimation as described in claim 1, characterized in that, The step of using the SIFT algorithm to detect key points in the preprocessed panoramic image includes: Panoramic images are convolved using Gaussian filters at different scales. The Gaussian difference (DoG) at different scales is used to find key points. The formula for calculating the Gaussian difference (DoG) is: In the formula, Let (x, y) represent the Gaussian function, and (x, y) be the coordinates in the image. It is a scale parameter. and These are different scale ratios; The SIFT algorithm uses information about pixels near keypoints, the size of the keypoints, and the principal curvature of the keypoints to locate each keypoint. The key point detection of the preprocessed point cloud data includes: For each point in the preprocessed point cloud data, its normal vector is estimated by fitting a local plane. Based on the point's neighborhood information, the covariance matrix of the neighboring points is calculated, and then eigenvalue decomposition is performed on the covariance matrix. The eigenvector corresponding to the smallest eigenvalue is the estimated normal vector of that point. Let point... The neighborhood point set is Covariance matrix: in It is the centroid of the neighborhood point set, and eigenvalue decomposition is performed on C. The eigenvector corresponding to the smallest eigenvalue Just a point The normal vector; Setting points Its neighboring points The normal vectors are respectively and Then point The curvature estimate is expressed as Select points with curvature values ​​greater than a threshold as key points: 。 3. The method for underground mine surveying and resource reserve estimation as described in claim 1, characterized in that, The process of using the RANSAC algorithm to remove mismatched point pairs includes: S231, randomly select N non-collinear points from the initial matching point dataset as samples, and use these N sample points to calculate the homography matrix: N is a positive integer; S232, calculate the distance from other points in the initial matching point dataset to the fitted model. If the distance is less than the set threshold, then these points are determined to be interior points. Repeat S231-S232 multiple times, updating the interior point set M times in each iteration, and selecting the model with the most interior points as the optimal model. In the formula, z represents the probability that RANSAC obtains the correct model, p represents the proportion of inliers in the data, and k represents the number of points selected for fitting the model.

4. The method for underground mine surveying and resource reserve estimation as described in claim 1, characterized in that, Point cloud data of underground space in mines at different times and panoramic images of the underground environment of mines are obtained through the following methods: At different times, a mobile laser scanner was used to acquire point cloud data of the underground space of the mine, while a panoramic camera was used to capture panoramic images of the underground environment of the mine. A searchlight with supplementary light source was used for illumination, and the mobile laser scanner and the panoramic camera operated synchronously.

5. The method for underground mine surveying and resource reserve estimation as described in claim 1, characterized in that, Step S6 includes: S61, register the ore body grid models from different periods to make the ore body grid models from different periods spatially aligned. S62, superimposes the geometric structure and attribute information of the ore body grid model at different times, and retains the change information of the ore body at different times to complete the model superposition. S63, after completing the model superposition, performs a closed modeling operation to make the superimposed model form a complete and closed three-dimensional model of ore body variation.

Images