A method and device for graphical management of coal piles in a coal yard of a thermal power plant

Abstract

This invention discloses a graphical management method and device for coal piles in a thermal power plant coal yard, belonging to the technical field of thermal power plant coal yard management. The method includes: acquiring three-dimensional laser point cloud data collected from a thermal power plant coal yard, and converting the three-dimensional laser point cloud data into an initial triangular mesh model; performing curvature-based adaptive simplification processing on the initial triangular mesh model to obtain a simplified triangular mesh model; performing normal-vector-based region growing segmentation on the simplified triangular mesh model to generate multiple coal pile sub-block meshes, each coal pile sub-block mesh corresponding to an independent coal pile; for each coal pile sub-block mesh, calculating the centroid coordinates and bounding box volume of the coal pile sub-block mesh, and generating a coal pile identification label corresponding to the coal pile sub-block mesh based on the centroid coordinates and bounding box volume, the coal pile identification label being used to uniquely identify the corresponding independent coal pile in the graphical interface.

Description

Technical Field

[0001] This invention relates to the field of coal yard management technology in thermal power plants, specifically to a graphical management method and equipment for coal piles in thermal power plant coal yards. Background Technology

[0002] In the fuel management stage of thermal power plants, accurate measurement and visual monitoring of coal piles in open or semi-enclosed coal yards are fundamental to improving coal inventory efficiency and optimizing coal blending strategies. Traditional coal pile management relies on manual measurement with tape measures or single-point laser rangefinders to estimate coal pile volume. This method provides extremely limited information on the spatial morphology of the coal pile, making it difficult to describe the irregular undulations and accumulation boundaries of the coal pile surface. With the widespread adoption of 3D laser scanning technology, acquiring dense point clouds of the entire coal yard surface using scanning equipment and then reconstructing a 3D model of the coal pile has become a trend. When reconstructing a coal pile model using 3D laser point clouds, the amount of data in the original scanned point cloud is enormous, and the directly generated high-resolution triangular mesh model contains millions or even tens of millions of faces. This full mesh consumes huge amounts of hardware resources during rendering, storage, and subsequent segmentation calculations, and there is a large amount of redundant geometric information in relatively flat areas of the coal pile surface. Conventional mesh simplification methods typically employ a globally uniform reduction ratio or iterative edge collapse strategy. These methods treat all regions of the model equally, failing to retain sufficient detail in geometrically complex areas with drastically varying curvature, such as coal pile edges and ridges, while simultaneously eliminating redundancy in vast, flat areas and gentle slopes. This results in either severe loss of local features in the simplified model, leading to inaccurate subsequent segmentation, or insufficient simplification rates to meet the lightweight requirements of interactive graphical management systems. Coal piles in coal yards are often adjacent and have blurred boundaries, making automatic identification and segmentation of individual coal piles difficult. Current practices often rely on operators manually drawing boundaries on point clouds or mesh models, or mechanically dividing them according to pre-defined rectangular meshes. Manual drawing is inefficient and highly subjective, while mechanical segmentation completely ignores the natural stacking morphology of coal piles, resulting in the abrupt fragmentation of the same coal pile or the incorrect merging of multiple coal piles. How to enable computers to automatically perceive the trend of normal changes on the surface of coal piles, group adjacent regions with consistent directions in continuous curved surfaces into independent coal pile units, and generate a unique and stable digital identity for each independent coal pile unit after segmentation is a problem that the industry urgently needs to solve. Summary of the Invention

[0003] This invention provides a graphical management method and equipment for coal piles in coal yards of thermal power plants, which solves the problems in the prior art where the grid simplification process is difficult to balance feature preservation and data simplification, as well as the difficulty in automatically dividing adjacent coal piles and the lack of stable and unique identifiers.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a graphical management method for coal piles in a thermal power plant coal yard, comprising the following steps: acquiring three-dimensional laser point cloud data collected from the coal yard of a thermal power plant, and converting the three-dimensional laser point cloud data into an initial triangular mesh model, using the triangular mesh to represent the surface morphology of the coal pile, providing a structured three-dimensional digital foundation for subsequent processing. An adaptive simplification process based on curvature is performed on the initial triangular mesh model to obtain a simplified triangular mesh model. This process treats different regions differently according to local curvature changes: in regions with low curvature and flat shapes, vertices are merged to reduce the amount of data; in regions with high curvature and rich features, new vertices are inserted to retain or even enhance details. This reduces model complexity while better preserving key geometric features such as coal pile edges and sharp corners, making the simplified model more suitable for segmentation and labeling. The simplified triangular mesh model is then segmented based on normal vectors to generate multiple coal pile sub-mesh blocks, each corresponding to an independent coal pile. This segmentation method determines whether triangular facets belong to the same coal pile surface based on the consistency of local normal vectors. It then uses plane fitting and angular constraints between adjacent regions to achieve intelligent fusion of neighboring areas. This accurately separates multiple coal piles that are close to each other but have varying slope directions, suppressing the problem of splitting the same coal pile or adhering different coal piles, and improving the consistency between the segmentation results and the actual coal pile deployment. For each coal pile sub-block grid, the centroid coordinates and bounding box volume of the sub-block grid are calculated. Based on these coordinates, a corresponding coal pile identification label is generated. This label uniquely identifies the corresponding independent coal pile in the graphical interface. Using centroid and volume to generate stable and unique digital labels for coal piles enables coal yard management to shift from manual numbering to automated identification. Even when local morphological changes occur in coal piles or new coal piles are added, reliable differentiation is still possible, providing a consistent identification basis for graphical scheduling and inventory statistics.

[0005] As a preferred embodiment of the present invention, the process of converting the three-dimensional laser point cloud data into an initial triangular mesh model includes: voxelizing and downsampling the three-dimensional laser point cloud data to obtain uniformly distributed point cloud data; using the moving least squares method to estimate the normal vector of the uniformly distributed point cloud data to obtain the normal vector of each data point; and using the Poisson surface reconstruction algorithm on the uniformly distributed point cloud data based on the normal vector of each data point to generate a closed initial triangular mesh model. This scheme, based on noise reduction and point cloud homogenization, recovers a closed mesh surface through Poisson reconstruction guided by normal vectors, reducing holes and distortion, and providing more accurate normal vectors and topological information for subsequent curvature calculation and region segmentation.

[0006] As a preferred embodiment of the present invention, the specific method for performing curvature-based adaptive simplification processing on the initial triangular mesh model to obtain a simplified triangular mesh model includes: traversing each triangular facet in the initial triangular mesh model, calculating the Gaussian curvature and average curvature of the triangular facet, and determining the curvature level of the triangular facet based on the Gaussian curvature and average curvature; for triangular facets with curvature levels lower than a preset low curvature threshold, merging the three vertices of the triangular facet into a new vertex and deleting the triangular facet; assigning curvature-based feature weights to the vertices in the initial triangular mesh model; performing global iterative simplification of the initial triangular mesh model using an edge collapse algorithm; when selecting edges to be collapsed, the edge collapse algorithm dynamically adjusts the collapse cost of each edge based on the vertex feature weights, so that vertices in high curvature regions are preferentially retained; when the number of vertices in the simplified triangular mesh reaches the target number of vertices, it is used as the simplified triangular mesh model. Selective simplification and subdivision based on curvature level in advance can enhance the expressive ability of coal pile edge and top transition areas without significantly increasing the total number of vertices. Global optimization of edge collapse makes the mesh density distribution more reasonable, and the running efficiency and accuracy of subsequent segmentation algorithms are improved.

[0007] Preferably, the number of target vertices is obtained by multiplying the total number of vertices of the initial triangular mesh model by a preset simplification ratio and then rounding down, thereby enabling models of different scanning batches and different coal yard sizes to automatically obtain an appropriate degree of simplification.

[0008] As a preferred technical solution of the present invention, the specific method for generating multiple coal pile sub-block meshes by performing region growing segmentation based on normal vectors on the simplified triangular mesh model includes: selecting an unlabeled triangular facet from the simplified triangular mesh model as a seed facet, and obtaining the unit normal vector of the seed facet; starting from the seed facet, traversing the neighboring triangular facets adjacent to the seed facet, and calculating the cosine value of the angle between the unit normal vector of each neighboring triangular facet and the unit normal vector of the seed facet; when the cosine value of the angle is greater than a preset similarity threshold, merging the neighboring triangular facet and the seed facet into the same candidate region, and... The unit normal vector of the neighboring triangular facet is added to the regional normal vector set of the candidate region. When the cosine value of the included angle is less than or equal to the preset similarity threshold, the growth of the neighboring triangular facet stops. When the candidate region can no longer absorb new triangular facets, the average normal vector of the candidate region is calculated based on the regional normal vector set, and plane fitting is performed on all triangular facets in the candidate region based on the average normal vector to obtain the fitting plane of the candidate region. Based on the angle between the fitting plane of each candidate region and the fitting plane of adjacent candidate regions, adjacent candidate regions with an angle less than the preset fusion angle are merged into the same coal pile sub-block grid. This scheme distinguishes different slopes by controlling the cosine value threshold of the normal vector, and then combines candidate regions belonging to different slopes of the same coal pile into a complete coal pile block by means of plane fitting and angle fusion, effectively balancing the precision and integrity of the segmentation.

[0009] Furthermore, when the dot product of the average normal vector of the candidate region and the unit normal vector of the neighboring triangular facet is greater than a preset normal vector consistency threshold, the neighboring triangular facet is merged into the candidate region. Using the overall normal vector direction of the candidate region to guide subsequent growth can suppress growth interruptions caused by single normal noise, improving the continuity and fault tolerance of region growth.

[0010] Further, the specific steps for selecting an unlabeled triangular facet as a seed facet from the simplified triangular mesh model include: obtaining all unlabeled triangular facests in the simplified triangular mesh model; calculating the product of the area and curvature of each unlabeled triangular facet to obtain a weight value for each unlabeled triangular facet; selecting the unlabeled triangular facet with the largest weight value as a candidate seed facet; determining whether the maximum deviation between the normal vectors of the three vertices of the candidate seed facet is less than a preset deviation threshold; if it is less than the preset deviation threshold, then the candidate seed facet is determined as a seed facet; if it is not less than the preset deviation threshold, then the candidate seed facet is discarded, and the triangular facet with the largest weight value is reselected from the remaining unlabeled triangular facests as a new candidate seed facet. By selecting representative triangular facests in flat regions using a weighted method based on area and curvature to initiate growth, and by using vertex normal vector deviation to filter out edge distortion regions, the seed facets become more stable, reducing missegmentation caused by abnormal growth. Preferably, the curvature of each unmarked triangular facet is the average of the Gaussian curvatures of the three vertices of the triangular facet.

[0011] As a preferred embodiment of the present invention, the specific steps for calculating the centroid coordinates and bounding box volume of each coal pile sub-grid include: obtaining the three-dimensional coordinates of all vertices contained in the coal pile sub-grid; summing the three-dimensional coordinates of all vertices along the coordinate axes and dividing by the total number of vertices to obtain the centroid coordinates of the coal pile sub-grid; obtaining the maximum and minimum values ​​of the three-dimensional coordinates of all vertices contained in the coal pile sub-grid in the X-axis direction, and calculating the difference between the maximum and minimum values ​​in the X-axis direction as the X-axis side length; obtaining the maximum and minimum values ​​of the three-dimensional coordinates of all vertices contained in the coal pile sub-grid in the Y-axis direction, and calculating the difference between the maximum and minimum values ​​in the Y-axis direction as the Y-axis side length; obtaining the maximum and minimum values ​​of the three-dimensional coordinates of all vertices contained in the coal pile sub-grid in the Z-axis direction, and calculating the difference between the maximum and minimum values ​​in the Z-axis direction as the Z-axis side length; and multiplying the X-axis side length, the Y-axis side length, and the Z-axis side length as the bounding box volume of the coal pile sub-grid. The centroid and axial bounding box are calculated directly from all vertices of the mesh. The calculation is simple and efficient, and can stably reflect the approximate location and spatial occupancy of the coal pile in space.

[0012] As a preferred embodiment of the present invention, the specific steps for generating a coal pile identification tag corresponding to the coal pile sub-grid based on the centroid coordinates and bounding box volume of the coal pile sub-grid include: quantizing the centroid coordinates of the coal pile sub-grid by rounding and quantizing them according to a preset coordinate quantization step size to obtain quantized centroid coordinates; quantizing the bounding box volume of the coal pile sub-grid by rounding and quantizing it according to a preset volume quantization step size to obtain quantized volume values; concatenating the quantized centroid coordinates and the quantized volume values ​​into a binary code stream in the order of centroid first and volume second, and using the binary code stream as the original identification code of the coal pile sub-grid; checking whether there is an existing identification code that is the same as the original identification code among all the currently generated coal pile identification tags; if there is an existing identification code that is the same, adding a random number to the lower bits of the original identification code and re-encoding it until a target identification code that is different from all existing identification codes is obtained; converting the target identification code into a string form to obtain the coal pile identification tag corresponding to the coal pile sub-grid. By generating identification codes through quantified coordinates and volume information, the labels themselves carry information about the spatial location and scale of the coal pile, possessing a certain degree of semantic self-interpretation. The collision handling mechanism ensures that even if a very small number of conflicts occur after quantization, a globally unique label can still be generated, meeting the distinguishability requirements of large-scale coal piles in the coal yard.

[0013] As a preferred embodiment of the present invention, after generating multiple coal pile sub-grids, the method further includes: for each coal pile sub-grid, calculating the sum of the areas of all triangular faces in the sub-grid to obtain the coal pile surface area corresponding to the sub-grid; for each coal pile sub-grid, based on the coal pile identification label corresponding to the sub-grid, rendering the sub-grid as a three-dimensional entity with a first color in a graphical interface, and displaying the coal pile surface area corresponding to the sub-grid at the centroid of the three-dimensional entity. This automatically visualized surface area labeling allows operators to intuitively grasp the bulk material distribution of each coal pile without separate measurements, providing immediate reference for coal blending and conversion decisions.

[0014] As a preferred embodiment of the present invention, after obtaining the coal pile identification tag corresponding to the coal pile sub-block grid, the method further includes: acquiring the target coal pile identification tag input by the user through a graphical interface; searching for a matching identification tag that matches the target coal pile identification tag among all generated coal pile identification tags; highlighting the coal pile sub-block grid corresponding to the matching identification tag in a second color in the graphical interface; and displaying the centroid coordinates and bounding box volume of the coal pile sub-block grid in a text pop-up window. Managers only need to input the tag to quickly locate the target coal pile and obtain key parameters, significantly improving search and confirmation efficiency and adapting to the multi-point scheduling needs of large coal yards.

[0015] As a preferred embodiment of the present invention, after obtaining the simplified triangular mesh model, the method further includes: calculating the Laplace coordinates of each vertex in the simplified triangular mesh model, and identifying whether the vertex is a boundary vertex based on the Laplace coordinates of each vertex; connecting all boundary vertices into at least one boundary closed loop, each boundary closed loop corresponding to the outline boundary line of a coal pile area; projecting each boundary closed loop onto a horizontal plane to obtain the horizontal projection polygon corresponding to the boundary closed loop, and calculating the area of ​​the horizontal projection polygon as the land area of ​​the corresponding coal pile area. By extracting the coal pile boundary using Laplace coordinates and projecting it to calculate the land area, the actual land area of ​​the coal pile can be automatically obtained from the three-dimensional model, facilitating the management of coal yard layout and land utilization rate calculation.

[0016] This invention also provides a graphical management device for coal piles in a thermal power plant coal yard, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the graphical management method for coal piles. This device solidifies the above method into a one-click automatic processing flow, performing model reconstruction, adaptive simplification, region segmentation, identifier generation, and visualization on the input 3D laser point cloud. While ensuring the integrity of the coal pile features, it significantly reduces manual intervention, providing a highly efficient and reliable digital management tool for thermal power plant coal yards.

[0017] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By employing curvature-based adaptive simplification, differentiated simplification strategies are adopted for regions with different geometric features in the initial triangular mesh model. After calculating the Gaussian curvature and average curvature of each triangular facet, the facets are divided into different levels according to their curvature. For flat facets with extremely low curvature, vertices are directly merged and facets are deleted, significantly reducing redundant geometric elements. For feature edge regions with extremely high curvature, facet subdivision is performed and new vertices are inserted to enhance the expression of local geometric features. This bidirectional operation allows the simplified mesh model to maintain a clear geometric structure in key areas such as the ridges and transition boundaries of the coal piles while significantly reducing the overall number of vertices. Compared with a globally uniform simplification ratio, this avoids the dual contradiction of insufficient simplification in flat regions and loss of detail in feature regions. The output simplified model is lighter and more realistically preserves the natural segmentation clues between coal piles. In the region growth and segmentation stage, normal vector consistency is used to guide the aggregation of neighboring facets. Starting with seed patches that have stable curvature characteristics and large area weights, the system gradually absorbs patches with consistent spatial orientation into the same candidate region by comparing the cosine of the angle between the normal vectors of neighboring patches. When a region can no longer grow, a plane fitting is performed. Furthermore, by evaluating the angle between the fitting planes of adjacent candidate regions, adjacent clusters with gentle slope transitions are merged into complete coal pile sub-blocks. This process automatically conforms to the geometric flow of the coal pile surface, eliminating the need for manual boundary delineation and naturally grouping continuous slopes formed by mechanical sweeping. After segmentation, a unique identifier is generated based on the centroid coordinates and bounding box volume of each coal pile sub-block. Random bits are introduced for re-encoding in case of occasional conflicts, ensuring that each independent coal pile maintains a traceable and stable identity across multiple scans and model updates, enabling automated and graphical independent management of batch coal piles. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart illustrating the graphical management method for coal piles in coal yards of thermal power plants. Figure 2 This is a flowchart of the 3D laser point cloud data reconstruction process; Figure 3 This is a flowchart of coal pile mesh segmentation based on curvature adaptive simplification and normal vector region growth; Figure 4 This is a flowchart of seed patch selection and region growth based on geometric saliency and normal vector consistency; Figure 5 This is a flowchart of the process for generating grid identification labels for coal pile blocks; Figure 6 This is a flowchart illustrating a graphical management method for coal piles in a thermal power plant's coal yard. Figure 7 This is a schematic diagram of statistical filtering and voxelization downsampling of 3D laser point cloud data from a coal yard in a thermal power plant; Figure 8 This is a schematic diagram of the curvature level distribution of triangular facets in an initial triangular mesh model based on curvature. Figure 9 It simplifies the distribution of Laplace coordinate magnitudes and the determination of boundary vertices in triangular mesh models. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] See Figure 1 This invention provides a graphical management method for coal piles in a thermal power plant's coal yard, comprising: acquiring three-dimensional laser point cloud data collected from the coal yard of a thermal power plant, and converting the three-dimensional laser point cloud data into an initial triangular mesh model; performing curvature-based adaptive simplification processing on the initial triangular mesh model to obtain a simplified triangular mesh model; performing normal vector-based region growing segmentation on the simplified triangular mesh model to generate multiple coal pile sub-block meshes, each coal pile sub-block mesh corresponding to an independent coal pile; for each coal pile sub-block mesh, calculating the centroid coordinates and bounding box volume of the coal pile sub-block mesh, and generating a coal pile identification label corresponding to the coal pile sub-block mesh based on the centroid coordinates and bounding box volume of the coal pile sub-block mesh, the coal pile identification label being used to uniquely identify the corresponding independent coal pile in the graphical interface.

[0022] Example 1: In a specific implementation, voxelization downsampling is performed on the 3D laser point cloud data to obtain uniformly distributed point cloud data. The voxelization downsampling process includes: obtaining the coordinate range of the 3D laser point cloud data in 3D space, determining the 3D point cloud bounding box along the X-axis, Y-axis, and Z-axis; setting the voxel grid side length, dividing the 3D point cloud bounding box into multiple cubic voxel grids; for each cubic voxel grid, calculating the average coordinate of all 3D laser points falling within the cubic voxel grid, and using the point corresponding to the average coordinate as the retained point of the cubic voxel grid. The retained points of all cubic voxel grids constitute uniformly distributed point cloud data. In some embodiments, the voxel grid side length is determined based on the average point spacing of the 3D laser point cloud data, and the voxel grid side length is set to an integer multiple of the average point spacing, so that the uniformly distributed point cloud data reduces the data volume while maintaining geometric features.

[0023] See Figure 2 Before voxelization downsampling, statistical filtering is performed on the 3D laser point cloud data to remove outlier noise points. The statistical filtering process includes: for each laser point in the 3D laser point cloud data, searching for the K nearest neighbor points and calculating the average distance from the laser point to the K neighbor points; calculating the mean and standard deviation of the average distances of all laser points; and removing laser points whose average distance exceeds the mean plus or minus a preset multiple of the standard deviation as outlier noise points.

[0024] ; in, This represents the three-dimensional coordinate vector of the data point currently being estimated using normal vector estimation. Represents the first local neighborhood point set A three-dimensional coordinate vector of neighborhood data points This represents the number of neighboring data points in the local neighborhood point set. Indicates the relationship between the current data point and the first data point. Euclidean distance between neighboring data points This represents the weight function related to the Euclidean distance. This represents the unit normal vector of the local plane to be solved. The weighting function is a Gaussian function, and the smoothing bandwidth parameter in the weighting function is adaptively determined based on the point spacing of the local neighborhood point set. Solving the problem of minimizing this weighted sum of squared errors is equivalent to performing eigenvalue decomposition on the covariance matrix of the local neighborhood point set; the eigenvector corresponding to the smallest eigenvalue is the normal vector of the current data point. This operation is performed on each data point to obtain its normal vector. After obtaining the normal vector of each data point, the orientation of all normal vectors is adjusted to ensure consistency, so that the normal vectors of adjacent data points point to the same side of the point cloud surface.

[0025] Based on the normal vector of each data point, the Poisson surface reconstruction algorithm is applied to the uniformly distributed point cloud data to generate a preliminary closed mesh model. Subsequently, the RANSAC algorithm is used to identify and remove ground point clouds, or Boolean operations are used to remove the portion of the mesh below ground level, resulting in a non-closed triangular mesh model that only represents the surface of the coal pile. The execution process of the Poisson surface reconstruction algorithm includes: constructing the gradient field of a 3D indicator function using the uniformly distributed point cloud data and the normal vector of each data point (the 3D indicator function takes a value of 1 inside the point cloud and a value of 0 outside the point cloud); constructing the Poisson equation by approximating the gradient field of the 3D indicator function using the normal vector field; discretizing and solving the Poisson equation on an octree structure to obtain the value of the 3D indicator function at each node of the octree; setting an isosurface extraction threshold, and using the moving cube algorithm to extract isosurfaces from the 3D indicator function to generate a closed initial triangular mesh model. The maximum depth of the octree is set according to the density of the uniformly distributed point cloud data and the reconstruction accuracy requirements. The greater the maximum depth of the octree, the richer the details of the reconstructed initial triangular mesh model, but the greater the computational cost.

[0026] See Figure 7 The figure illustrates the voxel downsampling effect of 3D laser point cloud data collected from a coal yard in a thermal power plant, as described in Example 1. The original laser point cloud data (hollow circles) is represented by scattered points, showing point cloud clusters in two coal pile areas. The coordinate axes cover the X-axis (0-55 meters) and Z-axis (0-30 meters), reflecting the actual spatial dimensions of the coal yard. Black crosses mark outlier noise points removed by statistical filtering; these are relatively dispersed and clearly distinct from the point cloud clusters in the main coal pile area, verifying the effectiveness of statistical filtering in removing outliers. Gray solid triangles represent the retained points after voxel downsampling, evenly distributed in the two coal pile areas, with a significantly lower density than the original point cloud, reflecting the downsampling mechanism of voxel mesh division and coordinate average calculation. The voxel mesh side length setting ensures that the retained points maintain the spatial geometry of the coal pile while effectively reducing data redundancy. Overall, the distribution trend and label classification of the data points in the figure intuitively demonstrate the processing flow of voxelization downsampling combined with statistical filtering in Example 1, providing a high-quality and uniform point cloud foundation for subsequent normal vector estimation and Poisson surface reconstruction.

[0027] Example 2: In specific implementation, refer to Figure 3The process of performing curvature-based adaptive simplification on the initial triangular mesh model is implemented through the following steps: Traverse each triangular facet in the initial triangular mesh model, calculating the Gaussian curvature and mean curvature of each facet. Specifically, first calculate the Gaussian curvature and mean curvature of each vertex in the initial triangular mesh model. The Gaussian curvature of a vertex is calculated using the angular deficit in its neighborhood, and the mean curvature of the vertex is calculated using the discrete Laplace-Beltrami operator. Take the arithmetic mean of the Gaussian curvatures of the three vertices of the facet as the Gaussian curvature of that facet, and take the arithmetic mean of the mean curvatures of the three vertices as the mean curvature of that facet. For the inherent adaptation to the processing objective of the mesh simplification stage, the arithmetic mean of the vertex curvatures is used instead of the facet area-weighted average to calculate the curvature of the triangular facet. The purpose of calculating curvature in this stage is to provide a region flatness criterion for subsequent vertex merging, rather than pursuing analytical curvature accuracy on the continuous surface of the triangular facet. The discrete curvature calculated based on vertex curvature is directly bound to the spatial position of the mesh vertices. Since the subsequent vertex merging operation is performed on vertices, using a direct arithmetic mean of vertex curvature simplifies the transfer of curvature information from facets to vertices. This eliminates the need for additional area weight redistribution calculations during vertex merging, thus ensuring the overall execution efficiency of the adaptive simplification algorithm. In typical working conditions where the coal pile surface is largely flat, the area difference between adjacent triangular facets is small. The arithmetic mean and area-weighted mean have limited impact on the curvature level classification results and are insufficient to change the basic determination of flat and characteristic regions. A comprehensive curvature index is calculated for each triangular facet based on its Gaussian curvature and average curvature to quantify the significance of its geometric features. The comprehensive curvature index is defined. for: ; in, and These are the maximum values ​​of the Gaussian curvature and the average curvature of all triangular faces in the initial mesh, respectively, used to normalize the two terms to dimensionless quantities. and These are the corresponding weight coefficients. This represents the maximum absolute value of the Gaussian curvature of all triangular faces in the initial triangular mesh model. This represents the maximum absolute value of the average curvature of all triangular faces in the initial triangular mesh model. Based on the characteristic that the surface of coal piles in thermal power plants is typically relatively flat, and that the average curvature better reflects the pile slope, this is set... , Calculated A larger value indicates that the region containing the triangular facet has more pronounced geometric features. A low curvature threshold is preset. and high curvature threshold Directly acting on Values ​​are used to divide regions.

[0028] An optimized design was developed for the unique stacking morphology and measurement environment of coal piles in thermal power plant coal yards. The values ​​were not arbitrarily set but based on a deep understanding of the physical properties of coal piles. Regardless of the varied surface morphology of coal piles in thermal power plant coal yards—such as steep slopes, steps, or deep pits—they are formed by the natural accumulation of loose particles under gravity or by coal-pushing machinery. Unlike man-made mechanical parts containing sharp edges and free-form surfaces, the local abrupt changes on the surface of coal piles are essentially formed by particle slippage and collapse, and their macroscopic characteristics are determined by the internal friction angle and angle of repose between particles. Gaussian curvature describes the degree of curvature of a local shape at a point on a surface relative to a plane and is extremely sensitive to local noise and sharp abrupt changes. In point cloud data, the edges of deep pits or steps on the surface of coal piles often appear as abrupt isolated points or broken lines. Their Gaussian curvature can produce extremely large outliers that do not conform to the continuous stacking characteristics of coal piles. If a large weight is given to Gaussian curvature, these scanning noises or irregularities at the particle scale will be misjudged as key geometric features that need to be retained, significantly increasing mesh complexity and disrupting the macroscopic segmentation criteria of the main slope of the coal pile. In contrast, the mean curvature, as the arithmetic mean of the two principal curvatures, can more stably reflect the overall tilt trend and slope transition of the coal pile surface. Even at the boundary between steep slopes and flat land, the mean curvature can smoothly transition and is less susceptible to interference from extreme curvature values ​​at the edges of local small deep pits.

[0029] The preset low curvature threshold and preset high curvature threshold are determined based on the curvature value distribution of all triangular faces on the entire initial triangular mesh model. All triangular facet curvature values ​​are arranged in ascending order. The preset low curvature threshold is the curvature value at the first percentile after sorting, with the first percentile set to 10%. The preset high curvature threshold is the curvature value at the second percentile after sorting, with the second percentile set to 90%. Directly using the global 10% and 90% quantiles as the thresholds for dividing high and low curvature regions is a core design feature for achieving adaptive simplification and avoiding the use of preset absolute empirical thresholds. In multi-batch, multi-coal-type, and multi-condition scans in industrial sites, the overall shape of coal piles varies greatly, making a single absolute curvature threshold unsuitable. Using quantile thresholds essentially maps the absolute value of curvature to its statistical distribution position within the entire mesh model, a positively correlated adaptive normalization strategy with the overall geometric complexity of the model.

[0030] When a coal yard has a highly flat grid due to the storage of a single type of coal (such as long-flame coal with good flowability) and leveling operations, it means that the entire coal pile surface lacks significant undulations such as protrusions, steep slopes, or deep pits. Under this condition, curvature calculation and subsequent processing can accurately reflect physical reality: an extremely flat grid has uniform and homogeneous surface geometry everywhere, lacking sharp geometric features that traditionally require special preservation. Even if the relative partitioning rules classify most areas as "high curvature," their absolute curvature values ​​are extremely small. In the subsequent edge collapse iteration process dominated by quadratic error metrics, the collapse cost of these areas remains essentially very low, and they will still be prioritized for simplification, avoiding erroneous over-subdivision. Conversely, for scenarios with an extremely flat grid, forcibly introducing a set of correction rules to stretch its internal distribution to create "high curvature" regions artificially creates feature levels in areas without geometric features, resulting in wasted computational resources on regions contrary to physical reality. For a generally flat coal pile, the primary goal is to accurately and holistically identify it, rather than forcibly creating dividing boundaries within it. Quantile rules are not lacking in adaptive corrections, ensuring that no spurious features are generated under flat conditions.

[0031] After the above vertex merging process, each vertex in the mesh... Assign a feature weight ,set up Take the combined curvature index of all triangular faces within a neighborhood of the vertex. The average value. When constructing the linear multiplicative form of the weighted collapse cost function, the larger of the sums of the feature weights of the two vertices connected by the edge is taken as the feature weight of that edge for cost correction, denoted as . This maximization strategy ensures that as long as any vertex connected by an edge is located in a high-curvature feature region, the collapse cost of the entire edge will be sufficiently amplified. This effectively suppresses the preferential collapse of edges crossing feature regions during global iteration, thus conservatism in protecting key geometric structures such as coal pile edges and ridges. The use of a linear multiplicative form instead of other complex nonlinear corrections is based on maintaining the numerical stability of the original quadratic error metric matrix in flat regions. In the linear form, when the feature weights approach zero, the collapse cost degenerates into the original quadratic error cost, ensuring that the standard behavior of the edge collapse algorithm remains undisturbed in flat regions. The edge collapse algorithm is used for global iterative simplification of the mesh. In each iteration, when selecting edges to be collapsed, the quadratic error cost of the edge is calculated. and introduce feature weights To correct this, construct a weighted collapse cost function: ; This makes edges located in high curvature regions less likely to be selected and collapsed due to their high cost, thus automatically preserving the edges and transition details of the coal pile during the global simplification process. The above iterative process is repeated until the number of vertices in the simplified triangular mesh reaches the target number of vertices (i.e., 20% of the initial total number of vertices), at which point the mesh is used as the simplified triangular mesh model.

[0032] After the vertex merging and triangular facet subdivision processes described above, an intermediate triangular mesh with adjusted geometric features is obtained. An edge collapse algorithm is then used for global iterative simplification of the intermediate triangular mesh. The core of the edge collapse algorithm is to reduce the number of vertices by iteratively selecting and collapsing edges in the mesh. Each time an edge is collapsed, its two endpoints are merged into a new vertex, and the two triangular faces connected to that edge are deleted. In each iteration, the collapse cost of each edge in the current intermediate triangular mesh is calculated. The collapse cost is calculated using a quadratic error metric matrix, where the quadratic error matrix of an edge is the sum of the quadratic error matrices of the two vertices at its two ends. The position of the new vertex is determined by minimizing the quadratic error function. The edge with the minimum collapse cost is selected for the collapse operation, updating the vertex, edge, and face information of the mesh. This iterative process is repeated until the number of vertices in the simplified triangular mesh reaches the target number of vertices. The target number of vertices is obtained by multiplying the total number of vertices in the initial triangular mesh model by a preset simplification ratio and then rounding down. The preset simplification ratio is set to 0.2, meaning 20% ​​of the initial total number of vertices is retained, and the rounding method is floor-rounding. The basis for setting the preset simplification ratio to 0.2 is that, in numerous coal yard point cloud reconstruction experiments, retaining 20% ​​of the vertices allows the mesh to significantly reduce storage and computational overhead while maintaining the macroscopic outline and volumetric characteristics of the coal pile. The simplified triangular mesh model is defined as the triangular mesh whose vertex count meets the target vertex count after the above iterative simplification.

[0033] When performing region growing segmentation based on normal vectors on a simplified triangular mesh model, initial candidate regions are generated. An unlabeled triangular facet is selected from the simplified triangular mesh model as a seed facet; unlabeled indicates that the facet has not been assigned to any existing candidate region. The unit normal vector of the seed facet is obtained. The unit normal vector is calculated by normalizing the normal vector of the plane containing the seed facet. In some implementations, the seed facet selection strategy is to traverse the indices of all triangular faces in the simplified triangular mesh model and select the first unlabeled facet encountered as the seed facet.

[0034] Starting from the seed facet, traverse all neighboring triangular faces adjacent to the seed facet. A neighboring triangular facet is a triangular facet that shares at least one edge with the seed facet. For each neighboring triangular facet, calculate the cosine of the angle between the unit normal vector of the neighboring triangular facet and the unit normal vector of the seed facet. This cosine is equal to the dot product of the two unit normal vectors. When the cosine is greater than a preset similarity threshold, it indicates that the neighboring triangular facet and the seed facet have similar orientations locally. The neighboring triangular facet is then merged into the current candidate region, and its unit normal vector is added to the region normal vector set of the current candidate region. The preset similarity threshold is set to 0.8. The reason for setting it to 0.8 is that the corresponding angle is approximately 36.9 degrees, which can encompass the local undulations on the coal pile surface caused by particle accumulation, while effectively distinguishing coal pile boundaries with obvious directional changes. When the cosine is less than or equal to the preset similarity threshold, region growth along the direction of that neighboring triangular facet stops, and the neighboring triangular facet is no longer included in the current candidate region.

[0035] The process of neighborhood traversal and angle judgment is continuously performed on newly formed triangular faces on the boundary of the candidate region, absorbing those that meet the conditions, until the cosine value of the included angle of all neighboring triangular faces of the boundary triangular face of the candidate region no longer meets the condition of being greater than the preset similarity threshold. At this point, the candidate region can no longer absorb new triangular faces. The average normal vector of the candidate region is calculated based on the set of normal vectors of the current candidate region. The average normal vector is the normalized result of the vector sum of all unit normal vectors in the set of normal vectors of the region. Plane fitting is performed on all triangular faces in the candidate region based on the average normal vector. The normal vector of the fitted plane is taken as the average normal vector, and the fitted plane passes through the centroid of the coordinates of all vertices in the candidate region. After obtaining the fitted plane of the candidate region, all triangular faces contained in the candidate region are marked as marked in the simplified triangular mesh model. New seed faces are repeatedly selected and grown until all triangular faces in the simplified triangular mesh model are marked, forming multiple initial candidate regions.

[0036] For all initial candidate regions, each pair of candidate regions that are adjacent in the mesh topology is examined. The angle between the fitting planes of the two adjacent candidate regions is calculated. The angle between the two fitting planes is obtained by calculating the magnitude of the cross product of the normal vectors of the two planes using the arcsine function. When the angle is less than the preset fusion angle, the two adjacent candidate regions are merged into the same coal pile sub-block mesh. The preset fusion angle is set to 15 degrees. The reason for setting it to 15 degrees is that, under natural slip accumulation, the turning angle between different accumulation surfaces of a coal pile is usually greater than 15 degrees. Adjacent regions with an angle less than 15 degrees are more likely to belong to the same coal pile due to the deviation of the fitting plane caused by surface unevenness, rather than truly independent coal pile bodies. The merging process merges all triangular faces of the two candidate regions into a new coal pile sub-block mesh object and updates the mesh topology connectivity. Finally, each coal pile sub-block mesh corresponds to an independent coal pile.

[0037] See Figure 8 In the figure, the horizontal axis represents the Gaussian curvature (dimensionless) of the triangular facets, and the vertical axis represents the average curvature (dimensionless). The legend distinguishes three levels of curvature for the triangular facets: low-curvature triangular facets (represented by hollow cubes, processed by vertex merging), medium-curvature triangular facets (represented by solid circles, preserved in their original state), and high-curvature triangular facets (represented by solid rhombuses, with high weights retained). Low-curvature triangular facets are mainly clustered in regions where both Gaussian curvature and average curvature are close to zero, indicating that their corresponding mesh regions are geometrically flat and suitable for reducing mesh complexity through vertex merging. Medium-curvature triangular facets are distributed in the middle range of Gaussian curvature (approximately 0.01 to 0.25) and average curvature (approximately 0.05 to 0.30), indicating that these facets have moderate geometric detail, and preserving their original state effectively retains the macroscopic features of the coal pile. High-curvature triangular patches are distributed in the higher range of Gaussian curvature (approximately 0.20 to 0.65) and average curvature (approximately 0.2 to 0.75), representing areas with rich surface details in the coal pile. These patches require mesh refinement through subdivision to ensure the accuracy of the reconstructed details. Overall, the curvature level division strictly corresponds to the preset low-curvature and high-curvature threshold ranges in the figure, reflecting the technical solution of the curvature-based adaptive simplification method in Example 2 for the reasonable classification and differentiated processing of the initial triangular mesh model of the coal pile. This figure intuitively illustrates the basis of the triangular patch simplification strategy, which helps maintain the key geometric features of the coal pile mesh model while effectively reducing model complexity and improving subsequent processing efficiency. Regarding attribute recalculation, due to the changes in the topological connectivity of the mesh caused by edge collapse and vertex merging operations, the original vertex normals, Laplacian coordinates, and curvature attributes are no longer valid. After obtaining the simplified triangular mesh model, all vertices in the model need to be retraced. Based on the new neighborhood topology, the normal vector of each vertex (which can be obtained by moving least squares method or PCA principal component analysis) and Laplace coordinates are recalculated and stored to ensure that the attribute data on which the subsequent coal pile segmentation and land area calculation depend are consistent with the current mesh structure.

[0038] Example 3: In specific implementation, the process of selecting seed patches from the simplified triangular mesh model adopts a screening strategy based on geometric saliency. See [reference needed]. Figure 4 The process involves obtaining all unlabeled triangular faces in the simplified triangular mesh model. An unlabeled face indicates that it does not currently belong to any formed candidate region. For each unlabeled face, its area and curvature are obtained. The product of the area and curvature is calculated and used as the weight value of the face. The area of ​​the face is obtained by calculating the Euclidean area of ​​the region enclosed by the three vertices of the face in 3D space, specifically by calculating half the magnitude of the cross product of the two edge vectors of the face. The curvature of the face is the average of the Gaussian curvatures of the three vertices. The Gaussian curvatures of the three vertices have already been calculated from the initial triangular mesh model and stored in the attribute data of each vertex during the adaptive simplification stage; the Gaussian curvature value of each vertex can be directly read. After obtaining the weight values ​​of all unlabeled faces, they are sorted from largest to smallest, and the unlabeled face with the largest weight value is selected as the candidate seed face. The weight values ​​reflect the importance of the triangular facets in terms of geometric features. Large areas with high curvature often correspond to locations with significant geometric contours on the coal pile surface. Using the facet with the largest weight value as a candidate seed facet helps to initiate region growth from the most representative location in the model.

[0039] After selecting candidate seed faces, their local flatness is verified. The normal vectors of the three vertices of each candidate seed face are obtained. These normal vectors were calculated and stored during the normal vector estimation stage and are directly read from the vertex attribute data. The deviation angles between each pair of normal vectors of the three vertices of the candidate seed face are calculated. The inverse cosine of the dot product of the normal vectors is used to obtain the angle between two normal vectors, and the maximum value among all angles is taken as the maximum deviation. The maximum deviation is compared with a preset deviation threshold, which is set to 10 degrees. The reason for setting the preset deviation threshold to 10 degrees is that if the angles between the normal vectors of the three vertices do not exceed 10 degrees, it indicates that the triangular face is located on a locally approximately flat surface or a surface with only slight curvature. Using such a face as a growth seed can ensure the representativeness of the initial normal vectors and the consistency of region growth.

[0040] If the maximum deviation between the normal vectors of the three vertices of a candidate seed facet is less than a preset deviation threshold, the candidate seed facet is confirmed as a formal seed facet and enters the region growth process. If the maximum deviation between the normal vectors of the three vertices of a candidate seed facet is not less than the preset deviation threshold, it indicates that the candidate seed facet is located in a local region with drastic curvature changes or significant geometric noise, and is not suitable as a seed facet. This candidate seed facet is removed from the selectable set, and the triangle with the largest weight value is selected from the remaining unlabeled triangle faces as a new candidate seed facet. This deviation judgment process is repeated until a seed facet that meets the deviation condition is found. This mechanism avoids initiating segmentation at sharp edges or concentrated point cloud noise on the coal pile surface, improving the stability of coal pile sub-block mesh generation.

[0041] During the candidate region growth process, in addition to merging based on the cosine of the included angle and a preset similarity threshold, a merging condition based on normal vector consistency is added. When a candidate region has formed a set of regional normal vectors and continues to attempt to absorb neighboring triangular facets, the dot product between the current average normal vector of the candidate region and the unit normal vector of the neighboring triangular facet to be merged is calculated. If the dot product between the average normal vector of the candidate region and the unit normal vector of the neighboring triangular facet is greater than the preset normal vector consistency threshold, then the neighboring triangular facet is merged into the candidate region. The preset normal vector consistency threshold is set to 0.9, corresponding to an included angle of approximately 25.8 degrees between the two normal vectors. The basis for setting the preset normal vector consistency threshold to 0.9 is that, in the coal yard coal pile scenario, the fluctuation of the normal vector of the natural slope of the same coal pile in the overall direction usually does not exceed 30 degrees. The threshold of 0.9 can accommodate the local roughness of the coal pile surface while ensuring that the normal vector direction of the same sub-block is consistent.

[0042] Optionally, the aforementioned normal vector consistency threshold and the preset similarity threshold corresponding to the cosine of the angle work synergistically. When traversing neighboring triangular faces, the cosine of the angle between the unit normal vector of the neighboring triangular face and the unit normal vector of the seed face is calculated simultaneously, as well as the dot product between the unit normal vector of the neighboring triangular face and the average normal vector of the candidate region. Only when both conditions are met simultaneously are the neighboring triangular faces merged into the candidate region. If either condition is not met, growth along the direction of that neighboring triangular face is stopped. Through this dual constraint, the candidate region maintains both the similarity of its local growth direction to the seed face and the consistency of the overall normal vector of the region, thus adapting to the continuous curved shape of the coal pile surface and avoiding over-segmentation.

[0043] Example 4: In specific implementation, the process of calculating the centroid coordinates and bounding box volume for each coal pile sub-block grid is as follows. (See reference...) Figure 5The process involves obtaining the 3D coordinates of all vertices within the coal pile mesh, summing the X-axis coordinates of all vertices, and dividing the sum by the total number of vertices in the coal pile mesh to obtain the X-axis component of the centroid coordinates. Similarly, summing the Y-axis coordinates and dividing the sum by the total number of vertices yields the Y-axis component of the centroid coordinates. Finally, summing the Z-axis coordinates and dividing the sum by the total number of vertices yields the Z-axis component of the centroid coordinates. The mathematical expression for this process is as follows: ; in, This represents the total number of vertices in the coal pile sub-block grid. Represents the first element in the coal pile sub-block grid. The X-axis coordinate components of each vertex Represents the first element in the coal pile sub-block grid. The Y-axis coordinate components of each vertex Represents the first element in the coal pile sub-block grid. Z-axis coordinate components of each vertex The X-axis component represents the coordinates of the centroid of the coal pile grid. The Y-axis component represents the coordinates of the centroid of the coal pile block grid. The Z-axis component represents the coordinates of the centroid of the coal pile block grid.

[0044] Obtain the X-axis coordinate components of the 3D coordinates of all vertices contained in the coal pile mesh. Iterate through all vertices, recording the maximum and minimum values ​​of the X-axis coordinate components. Calculate the difference between the maximum and minimum values ​​of the X-axis coordinate components and use this difference as the X-axis side length. Obtain the Y-axis coordinate components of the 3D coordinates of all vertices contained in the coal pile mesh. Iterate through all vertices, recording the maximum and minimum values ​​of the Y-axis coordinate components. Calculate the difference between the maximum and minimum values ​​of the Y-axis coordinate components and use this difference as the Y-axis side length. Obtain the Z-axis coordinate components of the 3D coordinates of all vertices contained in the coal pile mesh. Iterate through all vertices, recording the maximum and minimum values ​​of the Z-axis coordinate components. Calculate the difference between the maximum and minimum values ​​of the Z-axis coordinate components and use this difference as the Z-axis side length. Multiply the X-axis, Y-axis, and Z-axis side lengths together to obtain the bounding box volume of the coal pile block mesh. This bounding box volume is an axially aligned bounding box volume, reflecting the outer dimensions of the coal pile block mesh in three-dimensional space.

[0045] In some implementations, when the vertex distribution of the coal pile block mesh is sparse or the mesh only covers a local surface, a small expansion amount can be introduced into the calculation of the bounding box side length to avoid zero side length. The expansion amount is set to one-tenth of the point cloud spatial resolution.

[0046] During the coal pile identification label generation stage, the centroid coordinates of the coal pile sub-block grid are rounded and quantized according to a preset coordinate quantization step size. The preset coordinate quantization step size is set to 0.1 meters. The reason for setting 0.1 meters as the preset coordinate quantization step size is that the overall scale of a coal yard in a thermal power plant is usually in the range of tens to hundreds of meters. The quantization accuracy of 0.1 meters can distinguish the spatial location of different coal piles, and converts the floating-point values ​​of the centroid coordinates into integer values ​​with finite precision, which is convenient for subsequent coding. The specific method of rounding and quantization is as follows: divide the X-axis component of the centroid coordinate by 0.1 meters to obtain the quotient value, and round the quotient value to the nearest integer to obtain the integer value of the quantized centroid coordinate X-axis; perform the same operation on the Y-axis and Z-axis components to obtain the integer values ​​of the quantized centroid coordinate Y-axis and Z-axis, respectively. The three integer values ​​together constitute the quantized centroid coordinates.

[0047] The bounding box volume of the coal pile sub-block grid is rounded and quantized according to a preset volume quantization step size. The preset volume quantization step size is set to 1 cubic meter. The reason for setting 1 cubic meter as the preset volume quantization step size is that the volume of coal piles is usually tens to thousands of cubic meters. Using 1 cubic meter as the step size can distinguish coal piles of different volume sizes in the identification label, while avoiding different identification codes due to small volume fluctuations. The specific method of rounding and quantization is as follows: divide the floating-point value of the bounding box volume by 1 cubic meter, and round the quotient to the nearest integer to obtain the quantized volume value.

[0048] The quantized centroid coordinates and quantized volume values ​​are concatenated into a binary code stream in the order of centroid first, then volume. The specific concatenation method is as follows: a fixed-width binary buffer is allocated in memory, with a total width of 128 bits. The X-axis integer value of the quantized centroid coordinates occupies 32 bits, the Y-axis integer value occupies 32 bits, the Z-axis integer value occupies 32 bits, and the quantized volume value occupies 32 bits. These are written to the buffer sequentially in the order of X-axis integer value, Y-axis integer value, Z-axis integer value, and quantized volume value, forming a 128-bit binary code stream. In some implementations, 136 bits or other bit width schemes may also be used, with the bit width chosen to ensure that the quantized component values ​​do not overflow. The resulting binary code stream is used as the original identifier code for the coal pile block grid.

[0049] Retrieve all currently generated coal pile identification tags to form an existing identification code set. Check if there is an existing identification code in the existing identification code set that is exactly the same as the original identification code. The checking method is to compare each bit of the original identification code with each bit of the existing identification code in the existing identification code set. If no identical existing identification code is found, the original identification code is directly used as the target identification code. If an identical existing identification code exists, add a random number to the lower bits of the original identification code. The random number is an integer between 0 and 15, generated by a pseudo-random number generator. Specifically, shift the original identification code left by 4 bits, fill the lower 4 bits with the random number, and obtain a new code. Check again whether the new code is the same as the existing identification codes in the existing identification code set. If they are still the same, add another random number and repeat the above process until a target identification code that is different from all existing identification codes is obtained. After obtaining the target identification code, convert the binary sequence of the target identification code into corresponding hexadecimal characters in groups of 8 bits, and concatenate all hexadecimal characters into a string. This string is the coal pile identification tag corresponding to the coal pile sub-block grid. For example, a 128-bit target identifier code is converted into a 32-hexadecimal character string. Coal pile identification tags are displayed, retrieved, and associated as unique identifiers for individual coal piles in the graphical interface.

[0050] Example 5: In specific implementation, refer to Figure 6 After generating multiple coal pile sub-block meshes, surface area calculation and display operations are performed for each coal pile sub-block mesh. All triangular faces contained in the coal pile sub-block mesh are obtained, and the area of ​​each triangular face is calculated by traversing each triangular face. The area calculation method for each triangular face is as follows: obtain the 3D coordinates of the three vertices of the triangular face, calculate the first and second edge vectors of the triangular face, calculate the magnitude of the cross product of the first and second edge vectors, and multiply this magnitude by 0.5 to obtain the area of ​​the triangular face. The surface areas of all triangular faces in the coal pile sub-block mesh are summed to obtain the coal pile surface area corresponding to the coal pile sub-block mesh. In the graphical interface, based on the coal pile identifier label corresponding to the coal pile sub-block mesh, the coal pile sub-block mesh is rendered as a 3D entity with a first color. The first color uses a preset coal pile entity color value, such as tan or brown, to simulate the appearance of the coal pile. At the centroid of the rendered 3D coal pile mesh, create a floating text label. The text label contains the surface area value of the coal pile corresponding to the coal pile mesh. The display position of the text label is set to the centroid coordinate of the coal pile mesh superimposed with a preset vertical offset, so that the surface area value floats above the 3D entity for easy observation by the operator.

[0051] Optionally, the coal pile surface area value is updated in real time in the graphical interface. When the coal pile sub-block grid changes due to subsequent processing, the coal pile surface area is recalculated and the floating text label is refreshed.

[0052] After obtaining the coal pile identification tags corresponding to the coal pile sub-block grids, an interactive query function based on the coal pile identification tags is also provided. The graphical interface provides a tag input control, allowing users to input the target coal pile identification tag, which is a string. After obtaining the user-input target coal pile identification tag, the system performs a precise string matching search within the tag set consisting of all generated coal pile identification tags. Each generated coal pile identification tag in the tag set is compared character by character with the target coal pile identification tag. If a coal pile identification tag is found to be completely identical to the target coal pile identification tag, that tag is identified as the matching tag. The coal pile sub-block grid corresponding to the matching tag is highlighted in the graphical interface with a second color. The second color is chosen to have a significant visual difference from the first color, such as bright orange or bright yellow, while maintaining the edge outline of the first color to enhance recognizability. Simultaneously, the centroid coordinates and bounding box volume of the coal pile block mesh corresponding to the matching identifier label are retrieved from the stored coal pile sub-block mesh attribute data. The three components of the centroid coordinates and the value of the bounding box volume are generated into a text string, which is displayed as a pop-up window in the graphical interface. The pop-up window is positioned near the center of the bounding box of the 3D entity of the coal pile block mesh corresponding to the matching identifier label, in order to guide the user's attention to the corresponding coal pile.

[0053] After obtaining the simplified triangular mesh model, the area occupied by the coal pile is calculated. First, the bottom region of the coal pile mesh is identified by RANSAC plane fitting or manually setting the reference plane height. Boolean operations are used to remove the mesh portion below the reference plane, creating an opening at the bottom of the coal pile mesh, resulting in an open mesh model. Then, the Laplacian coordinates of each vertex in the open mesh model are calculated. For each vertex in the simplified triangular mesh model, the 3D coordinates of all neighboring vertices directly connected to that vertex via edges are obtained. The arithmetic mean of the 3D coordinates of all neighboring vertices is calculated to obtain the coordinates of the neighborhood center. Subtracting the neighborhood center coordinates from the vertex's 3D coordinates yields a 3D vector, which is the vertex's Laplacian coordinate. The Laplacian coordinates of each vertex are used to identify whether it is a boundary vertex. The identification method is as follows: traverse each edge in the simplified triangular mesh model and count the number of times each edge is referenced by a triangular facet. If an edge is referenced by only one triangular facet, then that edge is a boundary edge, and both endpoints of the boundary edge are marked as boundary vertices. The preset boundary determination threshold is set to twice the average Laplacian coordinate magnitude of all vertices in the simplified triangular mesh model. Since the bottom surface of the coal pile mesh model has an opening at the ground contact point after being cut, boundary edges (edges used by only one triangular facet) exist in the mesh. All vertices marked as boundary vertices are connected to form at least one closed boundary loop. The connection process is as follows: starting from any boundary vertex, trace along the boundary edge in the simplified triangular mesh model. Since the bottom surface of the coal pile mesh model has an opening at the ground contact point, boundary edges (edges used by only one triangular facet) exist in the mesh. Starting from any boundary vertex, trace along the boundary edge until returning to the starting point, forming a bottom contour closed loop. Each closed boundary loop corresponds to the contour boundary line of a coal pile region. Project the 3D coordinates of the boundary vertices contained in each closed boundary loop onto the horizontal plane. The projection method is to set the Z-axis coordinate component of each boundary vertex to zero, retaining the X-axis and Y-axis coordinate components, to obtain the horizontal projection polygon corresponding to the closed boundary loop. Calculate the area of ​​the horizontal projection polygon as the area of ​​the corresponding coal pile region. The area of ​​a horizontally projected polygon is obtained by arranging the vertices of the horizontally projected polygon in the order surrounding the polygon's boundary, calculating the directed area using the shoelace formula, and taking the absolute value.

[0054] A graphical management device for coal piles in a thermal power plant's coal yard includes a memory, a processor, and a computer program stored in the memory and running on the processor. The memory stores 3D laser point cloud data, triangular mesh model data, coal pile sub-block mesh data, coal pile identification tag data, and computer program instructions. When executing the computer program, the processor implements all the steps of the graphical management method for coal piles described in all the foregoing embodiments. In specific implementations, this device can be integrated into the host computer of a thermal power plant's coal yard monitoring system or deployed on a standalone graphical workstation. It acquires scanning data transmitted from a 3D laser point cloud acquisition device via a data interface, completes the fully automated processing from point cloud to coal pile identification tags, and outputs visual results to a graphical interface.

[0055] See Figure 9 In the figure, the horizontal axis represents the Laplace coordinate modulus of the vertices, in meters, and the vertical axis represents the probability density distribution of the corresponding modulus. The solid curve represents the probability density distribution of the Laplace coordinate modulus of all vertices in the simplified triangular mesh model of the coal yard of the thermal power plant, the dashed curve represents the probability density distribution of the Laplace coordinate modulus of only the boundary vertices, and the dotted line represents the preset boundary judgment threshold, which is twice the average value of the Laplace coordinate modulus of all vertices.

[0056] As seen from the solid curve, the Laplacian coordinate magnitudes of most vertices are concentrated in the range of 0 to 8 meters, peaking at approximately 0.5 meters, indicating that most vertices are located in relatively smooth or regular grid regions. The dashed curve shows that the Laplacian coordinate magnitudes of boundary vertices are significantly shifted to a larger range, peaking around 2 meters, and the distribution width is wider than that of the solid curve, reflecting a larger local coordinate deviation at the boundary vertices. The preset boundary determination threshold is approximately 1.2 meters, located near the intersection of the two curves, effectively distinguishing boundary vertices from non-boundary vertices.

[0057] The threshold setting is based on twice the average value of the Laplacian coordinate modulus of all vertices in the simplified triangular mesh model, ensuring that only vertices with a modulus significantly greater than the average value are marked as boundary vertices, thereby accurately extracting the contour boundary of the coal pile area. This figure vividly illustrates the rationality and effectiveness of boundary vertex identification using the Laplacian coordinate modulus in Example 5, providing data support for the subsequent formation of boundary closure loops and the calculation of the coal pile's footprint.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A graphical management method for coal piles in a thermal power plant's coal yard, characterized in that, include: Acquire three-dimensional laser point cloud data collected from the coal yard of a thermal power plant, and convert the three-dimensional laser point cloud data into an initial triangular mesh model; The initial triangular mesh model is subjected to curvature-based adaptive simplification to obtain a simplified triangular mesh model; The simplified triangular mesh model is divided into regions based on normal vectors to generate multiple coal pile sub-block meshes, each coal pile sub-block mesh corresponding to an independent coal pile; For each coal pile sub-block grid, calculate the centroid coordinates and bounding box volume of the coal pile sub-block grid, and generate a coal pile identification label corresponding to the coal pile sub-block grid based on the centroid coordinates and bounding box volume of the coal pile sub-block grid. The coal pile identification label is used to uniquely identify the corresponding independent coal pile in the graphical interface.

2. The graphical management method for coal piles in a thermal power plant coal yard according to claim 1, characterized in that, The specific steps for converting the three-dimensional laser point cloud data into an initial triangular mesh model include: The three-dimensional laser point cloud data is voxelized and downsampled to obtain uniformly distributed point cloud data; The normal vector of each data point is estimated by using the moving least squares method on uniformly distributed point cloud data. Based on the normal vector of each data point, the Poisson surface reconstruction algorithm is used to generate a closed initial triangular mesh model from the uniformly distributed point cloud data.

3. The graphical management method for coal piles in a thermal power plant coal yard according to claim 1, characterized in that, The specific steps for performing curvature-based adaptive simplification on the initial triangular mesh model to obtain a simplified triangular mesh model include: Traverse each triangular facet in the initial triangular mesh model, calculate the Gaussian curvature and mean curvature of the triangular facet, and determine the curvature level of the triangular facet based on the Gaussian curvature and mean curvature of the triangular facet. For triangular faces with curvature levels lower than the preset low curvature threshold, merge the three vertices of the triangular face into a new vertex and delete the triangular face. For triangular patches with curvature levels higher than the preset high curvature threshold, the midpoint of each edge of the high curvature triangular patch is inserted as a new vertex, and the original triangular patch is re-divided into 4 smaller triangular patches to form a finer-grained triangular patch division; after the above vertex merging and triangular patch subdivision processing, an intermediate triangular mesh with adjusted geometric features is obtained. Assign curvature-based feature weights to the vertices in the intermediate triangular mesh; perform global iterative simplification of the intermediate triangular mesh using an edge collapse algorithm; When selecting edges to be collapsed, the edge collapse algorithm dynamically adjusts the collapse cost of each edge based on the vertex feature weights, so that vertices in high curvature regions are preferentially preserved. When the number of vertices in the simplified triangular mesh reaches the target number of vertices, it is used as the simplified triangular mesh model.

4. The graphical management method for coal piles in a thermal power plant coal yard according to claim 3, characterized in that, The target number of vertices is obtained by multiplying the total number of vertices in the initial triangular mesh model by a preset simplification ratio and then rounding down.

5. A graphical management method for coal piles in a thermal power plant coal yard according to claim 3, characterized in that, The specific steps for performing region growing and segmentation based on normal vectors on the simplified triangular mesh model to generate multiple coal pile sub-block meshes include: Select an unlabeled triangular facet from the simplified triangular mesh model as a seed facet, and obtain the unit normal vector of the seed facet. Starting from the seed facet, traverse the neighboring triangular faces adjacent to the seed facet and calculate the cosine of the angle between the unit normal vector of each neighboring triangular facet and the unit normal vector of the seed facet. When the cosine value of the included angle is greater than the preset similarity threshold, the neighboring triangular facet and the seed facet are merged into the same candidate region, and the unit normal vector of the neighboring triangular facet is added to the region normal vector set of the candidate region. When the cosine value of the included angle is less than or equal to the preset similarity threshold, the growth of the neighborhood triangular facet is stopped; When the candidate region can no longer absorb new triangular facets, the average normal vector of the candidate region is calculated based on the set of normal vectors of the region, and a plane fitting is performed on all triangular facets in the candidate region based on the average normal vector to obtain the fitting plane of the candidate region. Based on the angle between the fitting plane of each candidate region and the fitting plane of the adjacent candidate regions, adjacent candidate regions with an angle smaller than the preset fusion angle are merged into the same coal pile block grid. The process of merging neighboring triangular faces into the candidate region specifically involves: simultaneously calculating the cosine of the angle between the unit normal vector of the neighboring triangular face and the unit normal vector of the seed face, as well as the dot product between the unit normal vector of the neighboring triangular face and the average normal vector of the candidate region. Only when both conditions are met simultaneously will the neighboring triangular faces be merged into the candidate region.

6. A graphical management method for coal piles in a thermal power plant coal yard according to claim 5, characterized in that, The specific steps for selecting an unlabeled triangular facet as a seed facet from the simplified triangular mesh model include: Obtain all unlabeled triangular faces in the simplified triangular mesh model, calculate the product of the area and curvature of each unlabeled triangular face, and obtain the weight value of each unlabeled triangular face. Select the unlabeled triangular facet with the largest weight value as the candidate seed facet; Determine whether the maximum deviation between the normal vectors of the three vertices of the candidate seed face is less than a preset deviation threshold. If it is less than the preset deviation threshold, then the candidate seed face is determined as a seed face. If the deviation is not less than the preset deviation threshold, the candidate seed facet is removed, and the triangle facet with the largest weight value is selected from the remaining unmarked triangle facets as the new candidate seed facet.

7. A graphical management method for coal piles in a thermal power plant coal yard according to claim 6, characterized in that, The curvature of each unlabeled triangular facet is the average of the Gaussian curvatures of the three vertices of that triangular facet.

8. A graphical management method for coal piles in a thermal power plant coal yard according to claim 5, characterized in that, For each coal pile sub-grid, the specific steps for calculating the centroid coordinates and bounding box volume of that sub-grid include: Obtain the three-dimensional coordinates of all vertices contained in the coal pile block mesh, sum the three-dimensional coordinates of all vertices according to the coordinate axis direction, and divide by the total number of vertices to obtain the centroid coordinates of the coal pile block mesh; Obtain the maximum and minimum values ​​of the three-dimensional coordinates of all vertices contained in the mesh of the coal pile in the X-axis direction, and calculate the difference between the maximum and minimum values ​​in the X-axis direction as the X-axis side length; Obtain the maximum and minimum values ​​of the 3D coordinates of all vertices contained in the grid of the coal pile in the Y-axis direction, and calculate the difference between the maximum and minimum values ​​in the Y-axis direction as the Y-axis side length; Obtain the maximum and minimum values ​​of the three-dimensional coordinates of all vertices contained in the coal pile sub-block mesh in the Z-axis direction, and calculate the difference between the maximum and minimum values ​​in the Z-axis direction as the Z-axis side length; take the product of the X-axis side length, the Y-axis side length and the Z-axis side length as the bounding box volume of the coal pile sub-block mesh.

9. A graphical management device for coal piles in a thermal power plant's coal yard, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the graphical management method for coal piles in a thermal power plant coal yard as described in any one of claims 1 to 8.

Images