Abstract
High-precision 3D coal seam models are crucial for refined mining design and precise production in open-pit coal mining. However, due to limitations like sparse initial data and challenges in obtaining geological realism data, these models are often static with insufficient spatial resolution. We propose a UAV coal-rock recognition and 3D coal seam model dynamic correction technology, which generates a UAV color point cloud and applies an improved region-growing algorithm and Alpha-shape algorithm for coal-rock identification and boundary points extraction. During the ongoing mining process in open-pit mines, the latest geological realistic data is dynamically integrated, and the spatial interpolation correction technique is used to calculate the correction values of the spatial interpolation points. The model is then dynamically corrected through a TIN update and growth reconstruction algorithm, continuously improving the accuracy of the coal seam 3D model. Application results show that the coal-rock recognition and boundary points extraction are highly effective, with accuracies of 97.44% and 91.85%, respectively. The standard deviation of the 3D coal seam model before and after correction is 0.32 m. Field measurements reveal that the average elevation error of the corrected model is 0.34 m, representing a 78.76% reduction in error compared to the initial model.
Keywords: Open-pit coal mine, UAV color point cloud, Coal-rock recognition, 3D coal seam model, Dynamic precision correction
Subject terms: Civil engineering, Coal
Introduction
The accuracy of the 3D coal seam model in open-pit coal mining directly determines the reliability of production planning and execution and the efficiency of mining face operations. This, in turn, affects the mine’s ability to efficiently and reasonably complete the set production tasks1,2. In constructing 3D coal seam models for open-pit mining, the primary data sources are static information obtained during the geological exploration phase, such as borehole data and cross-sectional profiles. Typically, the 3D models are not dynamically updated, leading to low model accuracy and poor practicality3–7. As the overburden stripping progresses in open-pit mines, more geological information is exposed, and geological data continues to increase and improve. This development provides the conditions for the dynamic revision of the 3D coal seam model, transitioning it from a gray to a white state8.
The mining face is the most direct and effective source for obtaining geological realism data on coal and rock strata. Developing an accurate and transparent 3D coal seam geological model and achieving transparency of the mining face are key technologies that require further research9–14. Coal-rock recognition is critical to limiting the mining face’s transparency15. Currently, the acquisition of geological realism data from the mining face in open-pit mines primarily relies on GPS-RTK dynamic measurement technology. However, in areas with steep slopes within the mine, it is often difficult for surveyors to access, especially when collecting data from high and steep slopes, which poses safety risks. Moreover, given the extensive mining areas and numerous mining levels in open-pit mines, single-point measurements are labor-intensive and inefficient, and make the acquisition of geological realism data challenging.
Unmanned Aerial Vehicle (UAV) remote sensing technology, as an emerging measurement method, provides a broad field of view and precise positioning capabilities, allowing dynamic and flexible collection of large-scale point cloud and imagery data from the environment. It also offers temporal flexibility, significantly improving the efficiency of surveying operations in open-pit mining. Currently, UAV remote sensing technology has been extensively applied in various fields, including production planning, measurement acceptance, slope monitoring, and ecological restoration in open-pit mining16–21. Its high-precision 3D spatial point clouds and high-resolution imagery provide an ideal coal-rock recognition and geological realism data source.
Based on a thorough analysis of the 3D geological modeling process and the characteristics of UAV photogrammetry data, this study utilizes UAV-based coal-rock recognition and extraction methods to assist geological realism, enhancing the efficiency of traditionally labor-intensive geological realism tasks. A dynamic correction technique for the 3D coal seam model in open-pit mining is also proposed. This technique starts with an initial 3D coal seam model constructed from original data such as boreholes and cross-sections and iteratively refines the model using dynamic geological realism data from the coal seam roof and floor, acquired during the progression of the stripping and mining operations. This iterative process continually improves the modeling accuracy of the coal seam, providing a foundational model for refined mining design and precise production in open-pit mines.
UAV coal-rock recognition and boundary points extraction
Registration and fusion of UAV point cloud and imagery
This study employs a UAV-mounted LiDAR system to simultaneously capture 3D point clouds and imagery of open-pit coal mines during mobile surveys. The system integrates a global positioning system (GPS), a high-precision inertial navigation system (INS), and dynamic positioning technologies, with the LiDAR sensor and optical camera co-mounted to enable synchronized data acquisition. The LiDAR point cloud data accurately represents the 3D geometric information of the surface; however, due to hardware and operational limitations, the data is often visualized in monochrome, grayscale, or false colors, which fail to capture the actual color and texture features of the terrain and objects. Imagery provides high-resolution, texture-rich, and color-accurate two-dimensional Red-Green-Blue (RGB) images, which complement the point cloud data and effectively address the limitations of a single data source22. The fusion of 3D point clouds and optical imagery can significantly leverage the advantages of both data types, thereby enhancing the amount of information available. As shown in Fig. 1, for a point P in the 3D point cloud, let its coordinates in the point cloud coordinate system be (XP,YP,ZP) and its coordinates in the image space coordinate system be (XC,YC,ZC). The transformation relationship between these two coordinate systems is as follows:
 |
1 |
Fig. 1.
Principle of registration and fusion of 3D point cloud and 2D image.
where: R is the rotation matrix; T is the translation matrix; α, β, and γ are the rotation angles around the x-axis, y-axis, and z-axis, respectively; tx, ty, and tz are the displacements along the x-axis, y-axis, and z-axis, respectively.
After the coordinate registration is completed, the proportional relationship in Eq. (2) can be derived using the principle of collinearity involving the image point, LiDAR point, and the camera center within the image space coordinate system.
 |
2 |
where: (x,y) is the coordinates of the image point corresponding to the point cloud (XP,YP,ZP) in image plane coordinates; (x0,y0) is the coordinates of the principal point of the image; f is the focal length; and λ is the scale factor.
In order to achieve precise registration and fusion, camera lens distortion must be considered in the model. By combining Eqs. (1) and (2), the collinearity equation between the image plane coordinates and the point cloud coordinates is obtained23:
 |
3 |
where
 |
4 |
where: (XS,YS,ZS) is the coordinates of the camera’s principal point under the point cloud coordinate system. Δx and Δy are the distortion correction parameters of the image point; k1, k2, and k3 are the camera’s radial distortion parameters. p1 and p2 are the camera’s tangential distortion parameters.
By applying Eq. (3), the RGB color value corresponding to the point (x,y) is assigned to the respective point cloud data, enabling the visualization of a colored point cloud that contains both 3D spatial coordinates and RGB color information.
Coal-rock recognition based on improved region-growing algorithm
Analysis of traditional region-growing-algorithm
After colorizing the 3D LiDAR point cloud, the data are segmented into clusters based on spatial, geometric, and texture features to identify and delineate coal-rock distribution accurately. Point cloud segmentation algorithms are generally categorized into feature clustering24, model fitting25,26, and region-growing27. Among them, the region-growing algorithm performs point cloud region segmentation using predefined growth criteria and similarity metrics. The algorithm is simple, easy to implement, and yields good segmentation results, making it widely applied. However, the traditional region-growing algorithm presents several challenges in practice: ① The raw point cloud data in open-pit mines is unordered and sparse, lacking a clear statistical distribution pattern. Direct segmentation of 3D point clouds is inefficient and computationally intensive. ② The selection of initial seed points is highly random. Due to the unique and complex environment of open-pit mining, the rich color features of the point cloud may result in inconsistent segmentation outcomes if the seed points are poorly chosen. ③ For coal and rock point cloud data, relying solely on normal vectors as a growth strategy can lead to both under-segmentation and over-segmentation, making it challenging to identify coal and rock regions accurately.
Improved region-growing algorithm
(1) Bidirectional KD-tree based search strategy.
Efficient region growth requires selecting highly correlated points for diffusion and rapidly identifying neighboring points to enhance search performance. The octree and KD-tree are commonly used data structures for point cloud organization. Both have a time complexity of O(NlogN) for K-nearest neighbor searches, significantly lower than the raw point cloud data O(N2) time complexity. While the octree focuses on saving storage space, the KD-tree provides more precise point localization during searches. However, its search process is unidirectional, which can lead to multi-to-one mismatches. Therefore, this study uses a bidirectional KD-tree to structure the point cloud during region growth to improve search efficiency and matching accuracy.
(2) Color space conversion.
When UAV captures color information in open-pit mining, the data is inevitably influenced by environmental conditions and lighting, leading to some degree of distortion in the reconstructed models. While RGB offers an intuitive representation of color, strong inter-channel correlations hinder consistent color difference measurements, limiting its suitability for direct point cloud segmentation. In contrast, the Hue-Saturation-Value (HSV) color space more closely aligns with human color perception and differentiation, making it better suited for extracting color attributes. Additionally, the HSV color space is less susceptible to variations in lighting intensity. The RGB color space can be represented as a cube in the Cartesian coordinate system based on its color attributes and value ranges, whereas the HSV color space is defined as an inverted cone in the polar coordinate system. The corresponding transformation formula between the values of these two coordinate systems is as follows:
 |
5 |
 |
6 |
 |
7 |
where: H∈[0,2π] is the hue value; S∈[0,1] is the saturation value; V∈[0,1] is the luminance value; r, g, b∈[0,1] are the values of the three primary colors of the LiDAR point R, G, and B after normalization; Cmax=max(r,g,b) is the maximum value of the three primary colors of this LiDAR point; and Cmin=min(r,g,b) is the minimum value of the three primary colors of this LiDAR point.
(3) Initial seed point selection and growth fusion strategy.
The selection of initial seed points and the determination of growth criteria directly influence the accuracy of the point cloud clustering results for coal-rock regions. Considering the significant color differences between coal and rock in the colored point cloud, this study selects the point with the minimum hue H value in the local coal seam point cloud near the coal-rock boundary as the initial seed point for growth. Additionally, spatial distance Dij, normal vector angle θij, and color similarity Cij are integrated as a growth fusion strategy for region growth. The corresponding formula is as follows:
 |
8 |
where: (xi,yi,zi),(xj,yj,zj) are the positions of points i and j, respectively; ni and nj are the normals of points i and j, respectively; (Hi,Si,Vi) and (Hj,Sj,Vj) are the HSV color values of points i and j, respectively.
Algorithmic process
The coal-rock recognition process based on the improved region-growing algorithm is as follows:
(1) Organize the colored point cloud data using a bidirectional KD-tree to establish spatial search relationships and perform the conversion from the RGB to the HSV color space.
(2) Set the spatial distance threshold D0, the normal vector angle threshold θ0, the color similarity threshold C0, and the coal-rock color threshold H0. Then, select the point with the minimum hue H value in the local coal seam point cloud near the coal-rock boundary as the initial seed point.
(3) Calculate the spatial distance Dij, normal vector angle θij, and color similarity Cij between the seed point and its neighboring points.
(4) Evaluate the spatial distance, normal vector angle, and color similarity between the seed point and neighboring points. If the conditions Dij<D0, θij<θ0, and Cij<C0 are satisfied, the neighboring point is considered close to the initial seed point, with similar texture and color, and is assigned to the same group as the seed point.
(5) Assess the hue value Hi of the newly added neighboring point and compare it with the predefined coal-rock color threshold. If Hi<H0, the point is set as a new seed point. Repeat steps 3 to 5 until all neighboring points have been processed, completing the coal seam clustering and growth.
Boundary points extraction in coal seam clustering point cloud
The Alpha-shape algorithm proposed by Edelsbrunner28 is a simple, efficient, and stable boundary points extraction method. Due to the randomness of LiDAR points and the inherent discreteness of point clouds, conventional fixed-radius algorithms struggle to balance the precision and completeness of contour extraction. Therefore, based on the Alpha-shape algorithm, this study uses a projection technique to map the 3D point cloud onto a 2D plane grid, thereby obtaining the boundary grid. The boundary grid is then weighted by Euclidean distance, and a rolling circle is adaptively generated to extract the point cloud boundary contour with varying radii. The expression for the adaptive radius a is given by:
 |
9 |
where: a is the adaptive radius; α is the adjustment coefficient; (xa, i,ya, i)is any point of the boundary grid; (xa, j,ya, j) is a point in the 3 × 3 neighborhood grid of the boundary grid; K is the number of proximity points searched in the neighborhood grid. Finally, the generated 2D boundary points are subjected to a 3D back-projection, yielding the 3D boundary points of the coal seam point cloud.
Framework of dynamic correction technology for 3D coal seam model
The overall framework of the dynamic correction technique for the 3D coal seam model in open-pit mining is shown in Fig. 2.
Fig. 2.
Framework of dynamic correction technology for 3D coal seam model.
(1) The initial 3D coal seam model is constructed using a ternary DEM envelope voxel modeling method29 developed through multi-source geological data preprocessing, stratigraphic division, and 3D spatial interpolation. Points, lines, and surfaces in the underlying layers are categorized and labeled to enable rapid retrieval of control and inferred data for the coal seam roof and floor.
(2) Assuming the open-pit mine is currently in the ith extraction phase, updated 3D coal seam data, including coal-rock boundaries, profile line data, and borehole point data, are acquired through coal-rock recognition, supplementary exploration, and geological surveys.
(3) The coal seam data from the ith extraction phase are interpreted and preprocessed to generate dynamic correction data for the roof and floor, which are then used to update the coal seam model during this phase.
(4) The dynamic correction data for the coal seam roof and floor during the ith extraction phase are evaluated to ensure compliance with geological principles, particularly the required micro and macro topological relationships for 3D modeling.
(5) The validated dynamic correction data is integrated with known control modeling data for the coal seam roof and floor from the ith extraction phase. The correction values for the spatial interpolation points of the coal seam roof and floor in the i + 1 extraction phase are output. Here, the control modeling data from the ith extraction phase refers to geological data obtained from exploration and measurements, not interpolated spatial estimates.
(6) The corrected values are propagated through the triangular mesh of the coal seam roof and floor, updating the 3D coal seam model and realizing the correction of the spatial interpolation points of the 3D model of the coal seam in the ith extraction phase.
(7) The dynamic correction data for the roof and floor of the coal seam from the ith extraction phase are embedded, and a dynamic growth algorithm for irregular triangular meshes is applied to reconstruct the local triangular mesh for the ith extraction phase. This generates the corrected 3D coal seam model for the (i + 1)th extraction phase.
(8) As the stripping and mining operations progress, geological data for the coal seam at the (i + 1)th extraction phase are re-acquired. This process is iteratively repeated, with continuous dynamic feedback and correction of the coal seam model for the subsequent mining stages, ultimately resulting in a progressively refined and optimized 3D geological model of the coal seam.
Key technologies for dynamic correction of 3D coal seam model
Spatial interpolation correction technique
When performing 3D spatial interpolation using existing modeling sample point data, the attribute values of the interpolation points are related to those of the surrounding sample points and adhere to the first law of geography. Based on these conditions, this study utilizes the semi-variogram function to describe the spatial distribution of regional variability30, thereby enabling the correction of spatial interpolation points in the 3D coal seam model. The detailed calculation process is as follows:
(1) Establishment of the semi-variational function
 |
10 |
where: γ*(ξ) is the value of the experimental variance function for the deviation distance ξ; M(ξ) is the number of sample point pairs with a deviation distance of ξ; k is the sequence number of the sample point pairs; and Z(xi) and Z(xi + ξ) are the spatial area variable elevation values corresponding to the spatial locations at xi and xi + ξ, respectively, where i∈[1,n].
(2) Theoretical fitting modeling of the variational function.
Standard theoretical models for fitting experimental semi-variograms include the spherical, exponential, and Gaussian models. This study employs a spherical function model to fit the semi-variogram by conducting a comprehensive analysis and comparison. The semi-variogram is plotted with ξ on the horizontal axis and γ(ξ) on the vertical axis, and the expression for the spherical model is given by:
 |
11 |
where: γ(ξ) is the value of the theoretical variation function of the deviation distance ξ; C0 is the nugget constant (Nugget), reflecting the degree of stochastic variability of the regional variable; (C0 + C) is the abutment value (Sill); C is the partial abutment value, reflecting the degree of structural variation within the regional variable; a is the range, indicating the spatial autocorrelation scale of the regional change volume. When 0 < ξ ≤ a, to achieve better goodness of fit, the weighted least squares method can be applied:
 |
12 |
where: k is the kth deviation interval to be calculated; n is the number of intervals; and wk is the weight coefficient of the kth term, which can be expressed as:
 |
13 |
(3) Calculation of correction weighting factors.
The corrected weight coefficient λi at the interpolation position o is recalculated based on Eq. (14). The corrected weight coefficient can be obtained using the following formula:
 |
14 |
where: µ is the Lagrange multiplier; n is the number of sample points involved in the valuation.
(4) Calculation and updating of correction values for spatial interpolation points.
The corrected value zo for the interpolated position o is:
 |
15 |
The elevation value at the interpolation position o is updated by traversing the triangular mesh using the winged-edge data structure31.
Dynamic growth technique of TIN
The purpose of 3D geological modeling is to accurately reflect the layers’ accurate stratigraphic positions and geometries. During the modeling process, known geological data must be transformed into rigid constraints for geological modeling, ensuring that geological interfaces strictly pass through the corresponding stratigraphic sampling points. This guarantees the validity of the geological constraints32. Therefore, after completing the spatial interpolation correction, it is necessary to embed the collected dynamic correction data of the coal seam roof and floor and reconstruct the local triangular mesh.
Delaunay triangulation is currently the most commonly used method for generating triangulation irregular network (TIN) due to its uniqueness and favorable triangular properties, which make it particularly well-suited for surface approximation. The network construction algorithms can be categorized into three types: divide-and-conquer algorithms, triangle mesh growth algorithms, and point-by-point insertion algorithms. Among these, the point-by-point insertion algorithm is a dynamic triangulation method that is straightforward to implement and effectively addresses issues related to data selection for modeling. However, it cannot accurately construct the mesh around constraint boundaries. Therefore, based on the point-by-point insertion method for constructing the TIN, we employ a two-step approach33 that incorporates a multi-diagonal exchange algorithm34 to forcibly embed constraint edges into the pre-constructed triangulation, thus generating a constrained delaunay triangulation (CDT). The dynamic growth process of the TIN construction algorithm is shown in Fig. 3.
Fig. 3.
Flow of TIN dynamic growth construction algorithm.
The algorithm facilitates the generation of unrestrained and constrained TIN while enabling dynamic updates of the existing TIN within local regions. When local modeling data need to be refined, densified, or updated to obtain a more detailed 3D geological model of the coal seam, the top, and bottom TINs can be dynamically grown and reconstructed by embedding the refined or corrected data, thereby generating an updated and refined 3D coal seam model within the localized area.
Model iterative step-by-step dynamic correction technique
As the open-pit mining stripping operation progresses, coal seam data is continuously acquired. The reconstruction of the coal seam geological model should follow an iterative process of continuous refinement and correction. A stepwise iterative approach is used to dynamically and precisely correct the 3D coal seam model, with the detailed correction procedure illustrated in Fig. 4. Specifically, Fig. 4(a) represents the current state of the open-pit mining site during the ith extraction phase; Fig. 4(b) shows the mining site at the i + 1 extraction phase, where the working face has advanced a certain distance compared to the ith phase. Figure 4(c) depicts the mining site at the i + 2 extraction phase, with the working face further advanced compared to the i + 1 phase. This process continues iteratively as new coal seam roof and floor data exposed by the advancing open-pit operation are continuously obtained. The newly exposed coal seam information from the previous extraction phase is merged with the known coal seam control modeling data, enabling dynamic and fine-tuned updates of the 3D model in localized areas. A dynamically updated 3D coal seam model is created through repeated iteration and continuous improvement, progressively improving the modeling accuracy and ensuring transparency regarding the coal seam’s depositional state throughout the extraction process.
Fig. 4.
Step-by-step iteration dynamic correction of 3D coal seam model.
Experimental results and analysis
Engineering area overview and UAV data collection
To validate the feasibility of the proposed method, a case study was conducted at an open-pit coal mine, focusing on the 12 primary coal seam as the target for verification. The experimental prototype system was developed in C + + using the Visual Studio 2017 platform and executed on a 64-bit Windows 11 operating system.
Data acquisition in the study area was conducted using a PH-20 multirotor UAV equipped with a JoLiDAR-1500 LiDAR sensor (range accuracy: 5 mm) and a 61 MP camera (resolution: 9540 × 6336 pixels; focal length: 21 mm). Data were collected under clear weather and Beaufort scale three wind conditions, with the UAV flying at an altitude of 100 m and a speed of 10 m/s, resulting in a ground sampling density of 1.8 cm/pixel. The flight range extended 100 m beyond the working face to capture point cloud and imagery data of the coal-rock boundary area during the back-mining process of the 12 coal seam roof. After preprocessing the data, the 3D point cloud was colorized based on the corresponding imagery. To reduce noise and simplify computation, the colored point clouds were further cropped to define the research area, resulting in a 3D colored point cloud of the coal-rock strata, as shown in Fig. 5(a). The elevation information of the colored point cloud is presented in Fig. 5(b).
Fig. 5.
Coal-rock recognition and coal seam point cloud boundary points extraction results.
Analysis of coal-rock recognition and boundary points extraction results
To validate the feasibility of the proposed coal-rock recognition and boundary points extraction algorithm, the improved region-growing algorithm was applied to segment and identify the coal-rock colorized point cloud data. The recognition results are shown in Fig. 5(c), where the points belonging to the coal seam are rendered in cyan. The outer boundary points of the coal seam cluster point cloud were extracted using both traditional and improved Alpha-shape algorithms, as shown in Fig. 5(d) and 5(e). Since the traditional region-growing algorithm uses only the normal vector as the growth criterion, it is unsuitable for coal-rock point cloud recognition in open-pit mining and cannot accurately distinguish coal-rock regions. Therefore, no comparative experiments with the traditional region-growing algorithm were conducted in this study.
To assess the accuracy of the coal-rock recognition and boundary points extraction results, this paper calibrates the color point cloud using CloudCompare software, counts the number of real coal-rock point clouds, and extracts the coal seam clustering point cloud boundary points. Overall accuracy (OA) is the key index to quantitatively analyze the effect of coal rock segmentation and boundary points extraction, as shown in Eq. (16). This index intuitively reflects the overall correctness of the segmentation and extraction results.
 |
16 |
where: TP is the number of points correctly identified as coal seams (the number of points where the boundary points of coal seams are correctly extracted); TN is the number of points correctly identified as rock seams; FP is the number of points of rock seams incorrectly identified as coal seams (the number of points where the boundary points are not extracted); and FN is the number of points of coal seams incorrectly identified as rock seams.
The statistical analysis results of the improved region-growing algorithm, the traditional Alpha-shape algorithm, and the improved Alpha-shape algorithm are shown in Table 1. As seen in Table 1, the coal-rock recognition accuracy of the improved region-growing algorithm is 97.44%. This is because the improved algorithm integrates color information, spatial distance, and normal vector features, optimizing the selection of initial seed points and the growth and merging strategy. As a result, it better meets the requirements of coal-rock recognition in open-pit mining using UAV-collected color point clouds. Comparative analysis reveals that the improved Alpha-shape algorithm provides a more accurate extraction of coal seam point cloud boundary points, with better completeness, making it more suitable for extracting coal seam point cloud boundaries.
Table 1.
Coal-rock recognition and coal seam point cloud boundary points extraction results.
| Applied algorithm | Evaluation indicators | OA/% | |||
|---|---|---|---|---|---|
| TP | TN | FP | FN | ||
| Improved region-growing algorithm | 2,545,691 | 5,107,754 | 43,955 | 156,728 | 97.44 |
| Traditional Alpha-shape algorithm | 13,261 | 0 | 2770 | 0 | 82.72 |
| Improved Alpha-shape algorithm | 14,724 | 0 | 1307 | 0 | 91.85 |
Dynamic fine correction of 3d coal seam model
Initial 3D coal seam modeling
Based on borehole data from geological exploration of the open-pit coal mine (Fig. 6(a)), a 3D geological modeling method was applied to model the 12 coal seam at the current mining face. The LCM software platform35 was used as the modeling tool, with a spatial interpolation distance of 25 m. DEMs for the top, bottom, and side surfaces of the 12 coal seam were generated, and a 3D geological model of the 12 coal seam was constructed using the multi-face mesh solidification technique, as shown in Fig. 6(b).
Fig. 6.
Modeling data and 3D coal seam modeling update of 12 coal seam.
3D coal seam model correction and result analysis
Based on the boundary points of the coal seam extracted from the UAV point cloud, the boundary points were first screened to identify the demarcation points at the coal-rock interface. After data verification, sampling was conducted at 20 m intervals. The sampled points were then sequentially connected to generate the coal-rock boundary lines for the 3D model update, as shown in Fig. 6(a). Following the methodology outlined in Sect. 3, the 3D coal seam model was corrected. The updated 3D coal seam model is shown in Fig. 6(c), and the cross-sections before and after the two phases of model updates are shown in Fig. 6(d). The DEM of difference (DoD) method was used to evaluate the coal seam update’s effectiveness. The pre-update DEM of the coal seam roof was set as the reference DEM, while the post-update DEM was set as the test DEM. Both DEMs were analyzed through overlay comparison within the same coordinate system36. The results of the overlay analysis are shown in Fig. 7, and the corresponding distribution of correction deviations is presented in Fig. 8. The correction method has a noticeable localized impact on the coal seam model. The corrected area is primarily concentrated around the coal-rock boundary line used for correction, increasing progressively from left to right along the boundary line. The correction effect is minor in areas closer to the borehole control points. This distribution pattern is consistent with the borehole density pattern of higher density on the left and lower density on the right, which aligns with the geological statistical laws discussed in this study. The standard deviation of the model before and after correction is 0.32 m, with the maximum deviations of positive and negative terms 4.27 m and 4.80 m, respectively. The cumulative corrected area exceeding 0.5 m accounts for 4.72% of the total area, indicating a good localized correction effect.
Fig. 7.
Cloud image of DEM superposition analysis.
Fig. 8.
Deviation distribution of correction.
Based on the current mining state, the accuracy of model updates is verified using the newly exposed coal seam roof data from the first three sections (corresponding to three mining belt widths, each 20 m wide) in the following mining phase. GPS-RTK measurements were taken for the elevations of multiple sampling points, and the distribution of these sampling points, along with the measured elevation information, is shown in Fig. 9. Using bilinear interpolation, the elevation at the sampling points’ planar locations was extracted from the constructed pre- and post-update DEM models. Specifically, the XY coordinates remain unchanged, while the Z-coordinate is obtained through interpolation methods to retrieve the corresponding elevation. Table 2 presents the 3D location information of the sampling points and the extracted corresponding Z values.
Fig. 9.
Sampling point distribution and measured elevation information.
Table 2.
Sampling point coordinates and extracted elevation values.
| Sampling point number | Sampling point coordinates (m) |
Measured elevation (m) | Elevation in the initial model (m) | Elevation in the updated model (m) | |
|---|---|---|---|---|---|
| X | Y | Z | Z | Z | |
| 1 | 1623.74 | 1155.47 | 593.39 | 593.63 | 593.56 |
| 2 | 1903.38 | 1150.74 | 588.24 | 589.53 | 588.45 |
| 3 | 2198.97 | 1156.69 | 586.23 | 588.27 | 586.13 |
| 4 | 2469.07 | 1197.37 | 587.05 | 585.44 | 586.88 |
| 5 | 2622.22 | 1192.22 | 580.19 | 583.74 | 580.58 |
| 6 | 2903.42 | 1235.17 | 584.62 | 584.95 | 584.85 |
| 7 | 3079.77 | 1237.68 | 588.92 | 585.79 | 589.26 |
| 8 | 3278.22 | 1240.25 | 589.57 | 590.29 | 590.03 |
| 9 | 1673.24 | 1171.72 | 595.39 | 594.85 | 595.13 |
| 10 | 1869.73 | 1179.08 | 589.77 | 591.56 | 590.12 |
| 11 | 2146.00 | 1175.33 | 587.72 | 588.41 | 587.48 |
| 12 | 2443.32 | 1215.76 | 587.53 | 586.36 | 587.89 |
| 13 | 2629.30 | 1216.50 | 581.24 | 584.27 | 581.57 |
| 14 | 2838.74 | 1254.82 | 584.92 | 584.59 | 585.09 |
| 15 | 3033.04 | 1251.02 | 588.15 | 585.99 | 588.64 |
| 16 | 3305.47 | 1255.21 | 590.01 | 591.60 | 590.44 |
| 17 | 1644.14 | 1196.92 | 595.76 | 594.95 | 595.43 |
| 18 | 1885.43 | 1191.24 | 589.80 | 591.42 | 590.39 |
| 19 | 2208.13 | 1194.63 | 586.42 | 588.93 | 586.89 |
| 20 | 2458.77 | 1233.61 | 588.32 | 586.44 | 587.90 |
| 21 | 2670.13 | 1237.11 | 581.62 | 584.36 | 582.12 |
| 22 | 2898.18 | 1274.47 | 585.95 | 585.78 | 585.81 |
| 23 | 3104.14 | 1273.68 | 588.92 | 587.02 | 589.18 |
| 24 | 3308.22 | 1278.37 | 590.13 | 592.24 | 590.78 |
The calculated elevation error for the Z-values is shown in Fig. 10. The errors before and after correction are distributed between − 3.13 m and 3.55 m and between − 0.42 m and 0.65 m, respectively. Compared to the actual measured elevations, the average absolute errors for the corrected models in the first three sampling regions are 0.26 m, 0.33 m, and 0.42 m. The average absolute error of the coal seam elevation in the first three working faces after correction is 0.34 m, representing a 78.76% reduction compared to the initial 3D coal seam model. The dynamically corrected 3D model of the coal seam more precisely represents the actual coal seam morphology at the working face, enhancing the model’s transparency. This refined model provides essential foundational data for detailed mining design and precise production in open-pit mining operations.
Fig. 10.
Elevation error before and after correction of 3D coal seam model.
Discussion
Unlike traditional geological survey or drilling methods for coal seam data acquisition in open-pit mining, this study innovatively applies UAV-derived data for geological realism, significantly improving the efficiency and timeliness of geological surveys while reducing data acquisition costs. This approach provides a novel technological solution to obtaining large-scale coal seam data in open-pit mining. By conducting a detailed analysis of the elements and principles in constructing 3D coal seam geological models, we dynamically refine the model by continuously integrating the latest mining data. This enhances the automation of model updates and improves local accuracy, ultimately achieving greater transparency in the actual distribution of coal seams. The refined 3D coal seam models provide a solid foundation for refined mining design and precise production in open-pit coal mines.
Although the improved region-growing algorithm has successfully achieved coal-rock point cloud segmentation and clustering, the complex operating environment of open-pit mining, with numerous influencing factors, can lead to over-segmentation when the coal-rock boundary is poorly defined. Therefore, we will further investigate methods for acquiring coal seam geological realism data under complex conditions in mining faces to enhance the applicability of UAV-based geological realism work.
Conclusions
The following conclusions and comments can be made from the present study:
(1) UAV-based mobile measurement provides a novel technical approach to the geological realism of open-pit coal mining faces. By integrating UAV point cloud and image data to generate 3D colored point clouds, this study designed a coal-rock recognition method based on an improved region-growing algorithm and a boundary points extraction method based on an enhanced Alpha-shape algorithm. The feasibility and practicality of these methods were validated through a case study.
(2) The dynamic correction process and technical framework for the 3D coal seam model in open-pit mining were proposed, along with exploring the associated core algorithms. Spatial interpolation correction values were derived by continuously utilizing the geological information obtained from the previous mining phase. These values were then applied to dynamically update and reconstruct the 3D coal seam model for the following mining phase using TIN, thus improving the local accuracy of the coal seam model.
(3) The application results confirm that the correction technique proposed in this study can effectively improve the modeling accuracy of the 3D coal seam model, providing essential foundational model data for refined mining design and precise production in open-pit mining.
Acknowledgements
This research was supported by the National Natural Science Foundation of China (52374123), the Youth Fund of the National Natural Science Foundation of China (52204158), and the Basic Research Project of Liaoning Provincial Education Department (LJ212410147019).
Author contributions
Jie Yuan: Methodology, Software, Validation, Investigation, Writing–original draft, Writing-review & editing, Visualization. Guangwei Liu: Conceptualization, Writing–review & editing, Supervision, Project administration, Funding acquisition. Senlin Chai: Validation, Supervision, Software, Project administration, Visualization. Weiqiang Guo: Validation, Data curation. Yunlong Huang: Supervision, Resources.
Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Hu, L. Q., Wang, L. G. & Bi, L. 3D visualization system based production plan scheduling of underground mine. J. China Coal Soc.32 (9), 930–933 (2007). [Google Scholar]
- 2.Cai, Q. X. & Chen, Y. L. Review of 70 years’ achievements and high-quality development architecture system of surface coal mining in China. J. China Coal Soc.49 (1), 235–260 (2024). [Google Scholar]
- 3.Qu, H. G. et al. An efficient method for high-precision 3D geological modeling from intersected folded cross-sections. Acta Scientiarum Naturalium Universitatis Pekinensis. 44 (6), 915–920 (2008). [Google Scholar]
- 4.Liu, G. W. et al. Establishment and application of surface mine 3D geological model based on multi DEM. Coal Eng.9, 73–75 (2010). [Google Scholar]
- 5.Liu, G. W. et al. Study on integrated construction technology of noncontinuous coal seam solid mode. Coal Sci. Technol.46 (5), 205–211 (2018). [Google Scholar]
- 6.Zhang, H. P., Zhou, X. X. & Dai, W. A preliminary on applicability analysis of Spatial interpolation method. Geogr. Geo-Information Sci.33 (6), 14–18 (2017). [Google Scholar]
- 7.Wang, T. L. & Liu, G. W. Study on dynamic upgraded technology of surface mine topography and geologic integrated entity model. Coal Eng.46 (10), 229–231 (2014). [Google Scholar]
- 8.Mao, S. J. Gray geographical information system - the theory and technology of correct geological Spatial data dynamically. Acta Scientiarum Naturalium Universitatis Pekinensis. 38 (4), 556–562 (2002). [Google Scholar]
- 9.Mowrey, G. L. A new approach to coal interface detection: the in-seam seismic technique. IEEE Trans. Ind. Appl.24 (4), 660–665 (1988). [Google Scholar]
- 10.Yuan, L. Scientific conception of precision coal mining. J. China Coal Soc.42 (1), 1–7 (2017). [Google Scholar]
- 11.Mao, S. J. et al. Design and key technology research of coal mine distributed cooperative one map system. J. China Coal Soc.43 (1), 280–286 (2018). [Google Scholar]
- 12.LI, S. B. Progress and development trend of intelligent mining technology. Coal Sci. Technol.47 (10), 102–110 (2019). [Google Scholar]
- 13.Jiang, B. C. et al. Using BHR to detect coal seam interface in coal mine. J. Appl. Geophys.209, 104894 (2023). [Google Scholar]
- 14.Zhang, L. et al. Definition and realization of unmanned mining in fully-mechanized coal mining face. Coal Sci. Technol.42 (9), 26–29 (2014). [Google Scholar]
- 15.Zhang, Q. et al. Review on coal and rock identification technology for intelligent mining in coal mines. Coal Sci. Technol.50 (2), 1–26 (2022). [Google Scholar]
- 16.Fu, E. S. et al. Short-term production planning of open-pit coal mine based on UAV data. Coal Sci. Technol.52 (7), 222–234 (2024). [Google Scholar]
- 17.Liu, G. W. et al. Study on applicability of unmanned aerial vehicle mobile survey of open-pit mine in alpine region. Coal Sci. Technol.47 (10), 43–50 (2019). [Google Scholar]
- 18.Zhao, Y. Q. et al. Intelligent identification and information calculation of slope crack in open-pit mine. Chin. J. Eng.46 (6), 1041–1053 (2022). [Google Scholar]
- 19.Du, W. J. et al. Application of unmanned aerial vehicle remote sensing on dangerous rock mass identification and deformation analysis: case study of a high-steep slope in an open pit mine. J. Earth Sci.36 (2), 750–763 (2025). [Google Scholar]
- 20.Padró, J. C. et al. Monitoring opencast mine restorations using unmanned aerial system (UAS) imagery. Sci. Total Environ.657, 1602–1614 (2019). [DOI] [PubMed] [Google Scholar]
- 21.Wen, C. H. & Wang, W. D. Study on ecological restoration of open pit mines based on UAV oblique photogrammetry technology. Coal Sci. Technol.48 (10), 212–217 (2020). [Google Scholar]
- 22.Tong, P. F. Monitoring surface deformation research in mining area by integrating UAV LiDAR and orthophoto. Changsha: Central South University 10–18 (2022).
- 23.Li, C. L. et al. Camera calibration method with multi-condition constraints based on laser points. Acta Photonica Sinica. 50 (4), 180–189 (2021). [Google Scholar]
- 24.Rabbani, T., Van Den Heuvel, F. & Vosselmann, G. Segmentation of point clouds using smoothness constraint. Int. Archives Photogrammetry Remote Sens. Spat. Inform. Sci.36 (5), 248–253 (2006). [Google Scholar]
- 25.Fischler, M. A. & Bolles, R. C. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Commun. ACM. 24 (6), 381–395 (1981). [Google Scholar]
- 26.Boulaassal, H., Landes, T. & Grussenmeyer, P. Automatic extraction of planar clusters and their contours on Building façades recorded by terrestrial laser scanner. Int. J. Architectural Comput.7 (1), 1–20 (2009). [Google Scholar]
- 27.Adams, R. & Bischof, L. Seeded region growing. IEEE Trans. Pattern Anal. Mach. Intell.16 (6), 641–647 (1994). [Google Scholar]
- 28.Edelsbrunner, H., Kirkpatrick, D. & Seidel, R. On the shape of a set of points in the plane. IEEE Trans. Inf. Theory. 29 (4), 551–559 (1983). [Google Scholar]
- 29.Jin, L. et al. Research on closed shell envelope voxel modeling of open-pit coal mine based on ternary DEM. Coal Eng.56 (2), 31–38 (2024). [Google Scholar]
- 30.Pan, D. & Li, X. D. Development of a 3D geological model of the mine based on SURPAC. Min. Technol.6 (3), 499–501 (2006). [Google Scholar]
- 31.Raghothama, S. & Shapiro, V. Boundary representation deformation in parametric solid modeling. ACM Trans. Graphics (TOG). 17 (4), 259–286 (1998). [Google Scholar]
- 32.Li, Z. L., Pan, M. & Han, D. K. Three-dimensional structural modeling technique. Earth Sci.41 (12), 2136–2146 (2016). [Google Scholar]
- 33.Bernal, J. & Bernal, J. On Constructing Delaunay Triangulations for Sets Constrained by Line Segments (US Department of Commerce, National Institute of Standards and Technology, 1988).
- 34.Li, L. X. & Tan, J. R. Multiple diagonal exchanging algorithm for inserting constrained boundary constrained Delaunay triangulation. Chin. J. Comput.22 (10), 1114–1118 (1999). [Google Scholar]
- 35.Liu, G. W. et al. Development and application of open pit coal mine mining full life cycle planning and decision-making platform. J. Intell. Mine. 5 (6), 73–81 (2024). [Google Scholar]
- 36.Liu, G. W. et al. Study on extraction method of slope surface deformation in open-pit mine by UAV remote sensing. J. Saf. Environ.24 (9), 3449–3457 (2024). [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.