Mineral powder volume measuring and calculating method based on three-dimensional shape data analysis

Abstract

The invention discloses a mineral powder volume measuring and calculating method based on three-dimensional shape data analysis, and relates to the technical field of mineral powder volume measurement. The method comprises the following steps: carrying out mineral powder pile scanning based on a mineral powder pile scanning module, and collecting three-dimensional point cloud data; preprocessing the collected point cloud data based on a point cloud preprocessing module; reconstructing the preprocessed point cloud data by using a three-dimensional reconstruction algorithm based on a mineral powder three-dimensional structure model construction module; based on the volume calculation module, performing mineral powder volume calculation by using a volume calculation algorithm based on voxel representation; and analyzing a calculation result. The method has the beneficial effects that the method has high automation and accuracy, the automatic processing and analysis process reduces manual intervention, the measurement accuracy and reliability are improved, the deep learning algorithm is applied, the capability of data processing and complex structure analysis is optimized, the method is suitable for mineral powder of different shapes and sizes, and the method is suitable for large-scale popularization and application. The method can be popularized to other industries requiring accurate volume measurement and shape analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of measuring the volume of ore powder, and particularly relates to a method for calculating the volume of ore powder based on three-dimensional shape data analysis. Background Art

[0002] Ore powder is a by-product in the production processes of ferrous metals such as steel, non-ferrous metals, and coal. The accurate measurement of its volume is crucial for the recycling, utilization, and environmental management of ore powder. However, the ore powder generated in the steel production process has highly irregular shapes and complex physical properties, which makes it difficult for traditional measurement techniques, such as manual measurement and those based on ordinary cameras, to accurately capture its true volume, and it is often limited by the experience of operators and the limitations of equipment. Although three-dimensional scanning technology provides a possible solution, it often lacks the ability to process large-scale data and analyze complex shapes when used alone. In addition, with the rapid development of artificial intelligence, especially deep learning technology, it has demonstrated powerful capabilities in fields such as image recognition, object reconstruction, and data analysis. Therefore, applying deep learning technology to three-dimensional data processing, especially in the automated and highly efficient measurement of ore powder volume, has significant technical and application prospects. The implementation schemes in the prior art are mostly: manually measuring the volume of the ore powder pile or extracting features and measuring the volume based on an ordinary camera; however, there are still the following deficiencies:

[0003] The technical scheme of manual measurement is difficult to accurately capture its true volume, and it is often limited by the experience of operators and the limitations of equipment. Although the ordinary camera scanning scheme provides a possible solution, it often lacks the ability to process large-scale data and analyze complex shapes when used alone. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for calculating the volume of ore powder based on three-dimensional shape data analysis in view of the deficiencies and defects in the prior art. This method has a high degree of automation and accuracy. The automatic processing and analysis process reduces manual intervention, improves the accuracy and reliability of measurement, and uses deep learning algorithms to optimize the ability to process data and analyze complex structures. It is applicable to ore powder of different shapes and sizes and can be extended to other industries that require accurate volume measurement and shape analysis.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A method for measuring the volume of ore powder based on three-dimensional shape data analysis, which includes the following steps: Scanning the ore powder pile based on the ore powder pile scanning module to collect three-dimensional point cloud data; Preprocessing the collected point cloud data based on the point cloud preprocessing module. The methods for preprocessing the point cloud data include denoising, filtering, and registration to ensure the quality and accuracy of the data. The specific methods for preprocessing the point cloud data are as follows: Using a graph-based sparse coding denoising algorithm for denoising, removing model noise and redundant details by maximizing the sparsity optimization function. At the same time, based on the octree data compression algorithm for data compression, applying the denoised three-dimensional model data to the octree data compression algorithm, representing the high-density data in the form of a hierarchical octree to reduce the storage and processing overhead of the data, and standardizing the compressed data based on the Gaussian correction method to ensure the consistency and comparability of the data, providing a unified data format and standard for subsequent analysis and calculation; Reconstructing the preprocessed point cloud data based on the ore powder three-dimensional structure model construction module and using a three-dimensional reconstruction algorithm. The method for reconstructing the point cloud data is as follows: Generating a three-dimensional model of the ore powder pile based on the Poisson three-dimensional plane reconstruction algorithm, and at the same time constructing a model based on the deep learning architecture of the Transformer, extracting data features through the Transformer architecture, and then using the Poisson surface reconstruction algorithm and the normal vector estimation and adjustment algorithm to generate a three-dimensional model of the ore powder pile; Based on the volume calculation module, using a volume calculation algorithm based on voxel representation to calculate the volume of the ore powder. The specific method for calculating the volume of the ore powder is as follows: Based on Delaunay triangulation and Laplace smoothing, converting the precisely constructed three-dimensional model into voxel representation to accurately calculate the volume of the model, and finally calculating the volume of the three-dimensional model to ensure the accuracy and reliability of the calculation result; Analyzing the calculation result, specifically: Generating the volume of the three-dimensional structure of the ore powder as a discriminator based on historical picture data, taking the output of using the volume calculation algorithm based on voxel representation to calculate the volume of the ore powder in the volume calculation module as the generating end of the generative adversarial network (GAN). If the output result of the discriminator does not conform to the actual situation, the above steps will be performed again.

[0006] Further, the preprocessing of the collected point cloud data by the point cloud preprocessing module is specifically as follows: Step 1, perform point cloud denoising. The specific method of point cloud denoising is: Based on the graph-based sparse coding denoising algorithm, construct a graph between data points to represent the local structure of the data, and then use sparse coding to recover and remove noise. Further, the graph-based sparse coding denoising algorithm regards data points as nodes in the graph, and the connections (edges) between nodes represent the similarity and distance relationships between data points, effectively removing the noise introduced by mechanical vibrations and environmental factors during the scanning process; Step 2, perform sparse coding. The specific method of sparse coding is: For the ore powder point cloud map after point cloud denoising, each node uses its neighbor nodes as a dictionary and performs sparse coding again; Sparse coding finds a sparse coefficient vector by solving an optimization problem. This vector can linearly combine neighbor nodes to approximately represent the target node, and the optimization goal is to maximize sparsity; Step 3, perform data compression and standardization. The octree data structure is used to compress the point cloud data. This method effectively retains key geometric information. Specifically: The octree data structure first divides the entire three-dimensional space into a cube, and then recursively divides each cube into eight sub-cubes. Each sub-cube corresponds to a different region of the space. This process continues until the preset resolution and maximum tree depth are reached. Further, the point cloud data is then stored in the corresponding tree nodes, effectively classifying the originally disordered point cloud data into the corresponding sub-spaces; In the tree nodes of each sub-cube, a certain number of point cloud data can be stored. When the data volume exceeds the preset threshold, further segmentation will be triggered. Among them, the tree nodes can store the indexes of points to reduce data redundancy and accelerate the subsequent query and operation speed. Further, in each sub-cube, duplicate and redundant points will be checked and removed to reduce data redundancy. According to actual needs, outliers beyond the predetermined range can also be deleted to further optimize the data. By adjusting the maximum depth of the octree, the resolution of the data can be controlled, and this resolution is related to the specific usage scenario; Preferably, in each node, a specific data standardization process is applied to ensure that the data is on the same scale and within the same range, facilitating subsequent calculations and reducing calculation errors caused by scale differences. Step 4, adopt a point cloud feature extraction technology based on deep learning to perform feature extraction. Use the improved version of the PointNet network, PointNet++, to capture the geometric and topological features of the ore powder pile to more effectively process non-uniformly sampled data. Specifically: PointNet++ learns the features of point cloud data through a hierarchical method, thereby improving the accuracy and processing ability of feature recognition. First, the point cloud is segmented into multiple local regions, and feature learning is independently performed within each region. This segmentation is based on the multi-scale grouping (MSG) strategy, allowing the model to capture features at different scales;In each of the segmented local regions, the model uses a small PointNet network to extract point features. This process involves multi-layer perceptron (MLP) and max pooling operations to extract features from each point and aggregate information within the region. Subsequently, through the error backpropagation algorithm, the PointNet++ network recursively applies these processing steps to the multi-level structure of the entire point cloud, gradually expanding from local to global, and finally forming a comprehensive point cloud feature description.

[0007] Furthermore, in the graph-based sparse coding denoising algorithm for point cloud denoising, each point in the ore powder point cloud map is first regarded as a node of the graph. Subsequently, based on the distance and similarity between points, edges are established to connect adjacent points. And the similarity between points is measured using the 2-norm based on Lagrangian penalty, and its expression is as follows: where x j are the three-dimensional coordinates of points i and j respectively, and λ is the penalty factor. This strategy effectively cancels out the noise caused by mechanical vibration. After similarity measurement, the system selectively sets thresholds and nearest neighbor methods to determine the specific number of neighbors connected to each node, thereby constructing a local structure diagram of the point cloud and using the K-nearest neighbor algorithm to determine the specific number of neighbors connected to each node.

[0008] Furthermore, the optimization objective in sparse coding can be achieved through L1-norm regularization, and the expression of this optimization objective is as follows: where is the set of neighbor points of point x i , a i is the coefficient vector of point x i , is the reconstruction error term, which ensures that the reconstructed x i is as similar as possible to the original x i , that is, the reconstruction error is as small as possible; the sparse regularization term ‖a i ‖1 promotes most elements to be zero through the l1 norm, thereby achieving sparsity.

[0009] Furthermore, in the data compression and standardization, the data standardization process can be executed according to the following expression: where y i,j is the j-th feature of octree node i, μj is the average value of the j-th feature of all octree nodes, and it is easy to obtain from the sampling theorem that the variance of the j-th feature of all octree nodes is Therefore is the j-th feature of the standardized octree node i.

[0010] Furthermore, the construction of the three-dimensional structure model based on mineral powder and the reconstruction of the preprocessed point cloud data using the three-dimensional reconstruction algorithm can be divided into two algorithms: normal vector estimation and adjustment, and Poisson surface reconstruction. The specific process of normal vector estimation and adjustment is as follows: Based on the feature information extracted by the Transformer architecture, the normal vector of the point cloud is automatically estimated to obtain more accurate results. The Transformer architecture learns the internal structure of the point cloud and selects a certain number of neighboring points around each point to estimate the normal vector. At the same time, based on the transformer architecture, the robustness of the algorithm can be guaranteed, that is, the influence of noise and outliers in the data can be fully filtered out to ensure the stability of the results. After normal vector estimation, a smoothing and adjustment algorithm is applied to ensure the consistency and correct direction of the normal vectors. By comparing the normal vectors of adjacent points, incorrect directions can be detected and corrected to ensure that the normal vector directions in the entire dataset are consistent, so as to obtain better surface quality in Poisson reconstruction. Subsequently, the Poisson surface reconstruction algorithm is used to convert the point cloud data after normal vector estimation and adjustment into a continuous three-dimensional surface. The specific process of Poisson surface reconstruction is as follows: Based on the position information and normal vectors of the points, a smooth and continuous three-dimensional surface is generated by solving the Poisson equation, and the octree obtained by preprocessing the collected point cloud data by the point cloud preprocessing module is analyzed. In the octree, each node represents a spatial region and contains the normal vector information of the points within this region. For each node of the octree, the divergence is calculated according to its normal vector. The divergence reflects the flow direction and intensity of the vector field in a specific region.

[0011] Furthermore, the expression for divergence calculation is as follows: where N(p) is the neighborhood of point p, which contains other points within a certain range from point p, and this distance is given as a hyperparameter by the actual scenario; is the normal vector of point p, and ∣q - p∣ is the Euclidean distance between point q and point p. In Poisson surface reconstruction, the core lies in solving the Poisson equation. Specifically, the Poisson equation defines a distributed vector field through the normal vector and position information of the points. The Poisson equation for calculating the continuous three-dimensional surface near point p is as follows: where f(p) is the three-dimensional surface function of point p to be generated, representing the position of the surface; is the Laplace operator, representing the second derivative in space. When solving the Poisson equation, the algorithm searches for a vector field consistent with the normal vectors of the points in the entire space and forms a smooth surface on this vector field. While generating a smooth surface, Poisson reconstruction can tolerate noise and holes in the data and is suitable for processing the irregular surface of the mineral powder pile.

[0012] Furthermore, the volume calculation module based on volume uses a volume calculation algorithm represented by voxels to calculate the volume of mineral powder, which is divided into three parts: grid generation, grid optimization, and volume calculation. Grid generation constructs a three-dimensional grid model from the three-dimensional structure model of mineral powder based on the triangulation method and the surface reconstructed from the preprocessed point cloud data using the three-dimensional reconstruction algorithm. The method of grid generation is as follows: Generate a complete three-dimensional structure through the triangulation method. Specifically, use the Delaunay Triangulation algorithm to divide the three-dimensional surface into a series of triangles. For any point P on the smooth surface generated by Poisson reconstruction, the circumcenter coordinates satisfying the Delaunay condition can be calculated according to the following expression:

[0013]

[0014] where, (x P , y P ) are the circumcenter coordinates satisfying the Delaunay condition; (x1, y1), (x2, y2), (x3, y3) are the coordinates of the three vertices of the Delaunay triangle;

[0015] Use the divide-and-conquer algorithm to implement the Delaunay triangulation algorithm. The steps are as follows: First, divide the point set into two parts based on the binary tree until it cannot be divided; then perform triangulation on each node of the binary tree; then merge the two parts based on the binary tree merge algorithm, and adjust the edges and triangle structures to ensure that the Delaunay condition is satisfied; Optimize the grid, delete redundant vertices and edges, and improve the stability and efficiency of the model. The method of grid optimization is as follows: Analyze the connection relationship of each vertex and delete duplicate and redundant vertices; Specifically, for the edges, ensure that each edge is only connected to two triangles to maintain the coherence of the grid; Then use Laplacian Smoothing to eliminate sharp corners and abnormal protrusions in the grid and improve the smoothness and stability of the grid. For any vertex vi on the smooth surface, the basic expression of its Laplacian smoothing is: where, N(ν i ) is the set of vertices adjacent to vertex ν i , |N(ν i )| represents the number of adjacent vertices, and ν′ i is the position of the vertex after smoothing. Based on the basic expression of Laplacian smoothing, the application process of Laplacian smoothing can be obtained: First, for each vertex ν i on the smooth surface, calculate the average position of its adjacent vertices based on the basic expression of Laplacian smoothing; Then move vertex ν iAdjust the position to the calculated average position until the given convergence value δ; The specific method for volume calculation is as follows: For the finally generated three-dimensional grid model, use the voxelization method to convert it into a voxel representation for accurate volume calculation; Preferably, apply a mathematical integration method to calculate the volume of the entire model, that is, the volume of the ore powder pile. The specific process is as follows: First, divide the three-dimensional grid after grid optimization into a series of tetrahedrons. Each tetrahedron is composed of a triangle and a reference point; By calculating the volumes of these tetrahedrons and adding them up, the volume of the entire model is obtained; The calculation of the tetrahedron volume is based on the coordinates of the points in the three-dimensional grid model, and the calculation formula is: Among them, the coordinates of the four points involved in this formula are (x1, y1, z1), (x2, y2, z2), (x3, y3, z3), (x4, y4, z4) respectively.

[0016] Furthermore, the analysis of the calculation results is specifically as follows: Analyze the calculation results of the ore powder based on the generative adversarial network (GAN); Specifically, this GAN network verifies and improves the quality of the ore powder volume calculation results through the adversarial training of the generative model and the discriminative model. The specific steps are as follows: First, train the GAN network based on the data in the ore powder database. The data includes the real pictures and volume values of the ore powder pile; The generator scans the ore powder pile based on the ore powder pile scanning module to collect three-dimensional point cloud data; Preprocess the collected point cloud data based on the point cloud preprocessing module; Reconstruct the preprocessed point cloud data based on the ore powder three-dimensional structure model construction module and using a three-dimensional reconstruction algorithm; Based on the volume calculation module, use a volume calculation algorithm based on voxel representation to calculate the volume of the ore powder to generate the volume of a certain ore powder pile. The discriminator tries to distinguish between the generated data and the real data; If the error between the volume generated by calculating the volume of a certain ore powder pile using a volume calculation algorithm based on voxel representation and the volume output by the discriminator is less than ∈ when scanning the ore powder pile based on the ore powder pile scanning module to collect three-dimensional point cloud data, preprocess the collected point cloud data based on the point cloud preprocessing module, reconstruct the preprocessed point cloud data based on the ore powder three-dimensional structure model construction module and using a three-dimensional reconstruction algorithm, and based on the volume calculation module, it means that the real calculation result is relatively accurate. At this time, output the result, transmit this group of data to the database for storage for subsequent offline model training to improve the accuracy of subsequent analysis; Otherwise, it indicates that there is a large error in the calculation result; The error ∈ between the estimated volume of the ore powder pile and the volume output by the discriminator is related to specific requirements; At the same time, the calculation result analysis module transmits a recalculation instruction to the point cloud processing system to re-perform the point cloud reconstruction work.

[0017] Furthermore, a method for measuring the volume of ore powder based on the analysis of three-dimensional shape data also includes a three-dimensional shape data measurement system for measuring the volume of ore powder, specifically:

[0018] a. A scanning module configured to collect point cloud data of the ore powder pile using three - dimensional scanning technology;

[0019] b. A data processing module configured to denoise, filter, and register the collected point cloud data. Further, the data processing module is configured to apply a graph - based sparse coding denoising algorithm and an octree data compression algorithm;

[0020] c. A feature extraction and model construction module configured to process the pre - processed point cloud data using a deep - learning - based algorithm and construct a three - dimensional structure model. Further, the feature extraction and model construction module is configured to use a Poisson surface reconstruction algorithm and a deep - learning architecture based on Transformer;

[0021] d. A volume calculation module configured to calculate the volume of the ore powder pile based on the constructed three - dimensional model. Further, the volume calculation module is configured to use a voxel representation method based on Delaunay triangulation and Laplacian smoothing;

[0022] e. An analysis module configured to analyze and optimize the calculation results using a generative adversarial network (GAN). The generative adversarial network (GAN) includes a generative model to generate the ore powder volume calculation results and a discriminative model to distinguish the generated calculation results from the actual calculation results.

[0023] After adopting the above - mentioned technical solutions, the beneficial effects of the present invention are as follows: This method has a high degree of automation and accuracy. The automatic processing and analysis process reduces manual intervention, improves the accuracy and reliability of measurement, and uses deep - learning algorithms to optimize the ability to process data and analyze complex structures. It is applicable to ore powders of different shapes and sizes and can be extended to other industries that require precise volume measurement and shape analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following - described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0025] Figure 1 It is a flow module diagram for analyzing the ore powder volume measurement by three - dimensional shape data in the present invention;

[0026] Figure 2 It is a detailed flow chart of step S2 in the analysis of the ore powder volume measurement by three - dimensional shape data in the present invention;

[0027] Figure 3This is the architecture diagram of the system for calculating the volume of mineral powder by analyzing three-dimensional shape data in the present invention. Detailed implementation mode

[0028] Refer to Figure 1 As shown, the technical solution adopted in this detailed implementation mode is as follows: It includes the following steps:

[0029] S1. Based on the mineral powder pile scanning module, scan the mineral powder pile to collect three-dimensional point cloud data; the ways of scanning the mineral powder pile include: laser scanning, depth camera, binocular camera and other high-precision scanning technologies; among them, when using a depth camera and a binocular camera to collect three-dimensional point cloud data, the specific models of the depth camera and the binocular camera are not limited.

[0030] S2. Based on the point cloud preprocessing module, preprocess the collected point cloud data. The ways of preprocessing the point cloud data include denoising, filtering, and registration to ensure the quality and accuracy of the data. And the preprocessing of the point cloud data also includes applying a graph-based sparse coding denoising algorithm and an octree data compression algorithm to optimize the data storage and processing efficiency. The specific ways of preprocessing the point cloud data are as follows:

[0031] Use the graph-based sparse coding denoising algorithm for denoising, remove model noise and redundant details by maximizing the sparsity optimization function. At the same time, based on the octree data compression algorithm, perform data compression. Apply the denoised three-dimensional model data to the octree data compression algorithm, represent the high-density data in the form of a hierarchical octree to reduce the storage and processing overhead of the data, and perform standardization processing on the compressed data based on the Gaussian correction method to ensure the consistency and comparability of the data, providing a unified data format and standard for subsequent analysis and calculation.

[0032] It can be referred to Figure 2 As shown, more specifically:

[0033] Step 1. Perform point cloud denoising. The specific way of point cloud denoising is as follows: Based on the graph-based sparse coding (Graph-Based Sparse Coding) denoising algorithm, construct a graph between data points to represent the local structure of the data, and then use sparse coding to recover and remove noise. Further, the graph-based sparse coding denoising algorithm regards data points as nodes in the graph, and the connections (edges) between nodes represent the similarity and distance relationships between data points, effectively removing the noise introduced by mechanical vibration and environmental factors during the scanning process. Among them, the graph-based sparse coding denoising algorithm first regards each point in the mineral powder point cloud map as a node of the graph, and then establishes edge connections between adjacent points according to the distance and similarity between points; and uses the 2-norm based on Lagrangian penalty to measure the similarity between points. Its expression is as follows:

[0034]

[0035] where x j are the three-dimensional coordinates of points i and j respectively, and λ is the penalty factor. This strategy effectively cancels out the noise caused by mechanical vibration;

[0036] After performing similarity measurement, the system selectively sets the threshold and the nearest neighbor method to determine the specific number of neighbors connected to each node, thereby constructing a local structure diagram of the point cloud;

[0037] Use the K-nearest neighbor algorithm to determine the specific number of neighbors connected to each node;

[0038] Among them, the K-nearest neighbor algorithm involves the hyperparameter K, and the hyperparameter K is set manually according to the specific scenario, so the specific value of the hyperparameter K is not restricted.

[0039] Step 2, perform sparse coding. The specific method of the sparse coding is as follows:

[0040] For the ore powder point cloud map after point cloud denoising, each node uses its neighbor nodes as a dictionary and performs sparse coding again; Sparse coding finds a sparse coefficient vector by solving an optimization problem. This vector can linearly combine neighbor nodes to approximately represent the target node. The optimization goal is to maximize sparsity. Among them, the optimization goal can be achieved through L1 norm regularization. The expression of this optimization goal is as follows:

[0041] <>

[0042] Among them, is the neighbor point set of point x i , a i is the coefficient vector of point x i , is the reconstruction error term. This part ensures that the reconstructed x i is as similar as possible to the original x i , that is, the reconstruction error is as small as possible;

[0043] The sparse regularization term ‖a i ‖1 makes most elements zero through the l1 norm, thereby achieving sparsity;

[0044] Among them, based on efficient acceleration techniques such as FISTA (Fast Iterative Shrinkage-Thresholding Algorithm) to solve this 1-norm regularization optimization problem and speed up the speed of finding the sparse solution,

[0045] Therefore, this algorithm realizes encoding noise removal and reconstruction: using the reconstructed values obtained by sparse coding to replace the original data points, because these reconstructed values are jointly determined by local neighbors and can suppress isolated noise points. During the reconstruction process, since it focuses on using points similar to the target point, it naturally filters out those noise components that do not conform to the local structural characteristics.

[0046] Step 3: Perform data compression and standardization. The octree data structure is adopted for point cloud data compression. This method effectively retains key geometric information. Specifically:

[0047] The octree data structure first divides the entire three-dimensional space into a cube, and then recursively divides each cube into eight sub-cubes. Each sub-cube corresponds to a different region of the space. This process continues until the preset resolution and maximum tree depth are reached. Further, the point cloud data is then stored in the corresponding tree nodes, effectively classifying the originally unordered point cloud data into the corresponding sub-spaces. In the tree nodes of each sub-cube, a certain number of point cloud data can be stored. When the data volume exceeds the preset threshold, further segmentation will be triggered. Among them, the tree nodes can store the indices of the points to reduce data redundancy and accelerate the subsequent query and operation speeds. Further, duplicate and redundant points will be checked and removed in each sub-cube to reduce data redundancy. According to actual needs, outliers beyond the predetermined range can also be deleted to further optimize the data.

[0048] Therefore, there are no restrictions on the specific threshold for storing point cloud data in the nodes and the method of checking for duplicate and redundant points.

[0049] By adjusting the maximum depth of the octree, the resolution of the data can be controlled, and this resolution is related to the specific usage scenario.

[0050] Preferably, in each node, a specific data standardization process is applied to ensure that the data is on the same scale and within the same range, facilitating subsequent calculations and reducing calculation errors caused by scale differences.

[0051] Among them, the data standardization process can be executed according to the following expression:

[0052]

[0053] where y i,j is the j-th feature of the octree node i, μj is the average value of the j-th feature of all octree nodes, and it is easy to obtain the variance of the j-th feature of all octree nodes according to the sampling theorem as Therefore is the j-th feature of the standardized octree node i.

[0054] Step 4: The point cloud feature extraction technology based on deep learning is adopted to perform feature extraction. The improved version of the PointNet network, PointNet++, is used to capture the geometric and topological features of the ore powder pile to more effectively process non-uniformly sampled data. Specifically:

[0055] PointNet++ learns the features of point cloud data through a hierarchical method, thereby improving the accuracy and processing ability of feature recognition. First, the point cloud is segmented into multiple local regions, and feature learning is independently performed within each region. This segmentation is based on the Multi-Scale Grouping (MSG) strategy, which allows the model to capture features at different scales. In each local region after segmentation, the model uses a small PointNet network to extract point features. This process involves multi-layer perceptron (MLP) and max pooling operations to extract features from each point and aggregate the information within the region. Subsequently, through the error backpropagation algorithm, the PointNet++ network recursively applies these processing steps to the multi-level structure of the entire point cloud, gradually expanding from local to global, and finally forming a comprehensive description of the point cloud features.

[0056] This method not only improves the model's ability to handle features of different sizes and complexities but also optimizes the efficiency of extracting useful information from highly irregular point cloud data.

[0057] Through this feature extraction technology, the present invention can effectively improve the accuracy and reliability of ore powder volume measurement, especially when dealing with ore powder piles with complex geometric structures.

[0058] S3. Reconstruct the preprocessed point cloud data based on the ore powder three-dimensional structure model construction module and using a three-dimensional reconstruction algorithm; the construction of the three-dimensional structure model includes using the Poisson surface reconstruction algorithm and a deep learning architecture based on Transformer. The method of reconstructing the point cloud data is as follows:

[0059] Generate a three-dimensional model of the ore powder pile based on the Poisson three-dimensional plane reconstruction algorithm. At the same time, construct a model based on the deep learning architecture of Transformer, extract data features through the Transformer architecture, and then use the Poisson surface reconstruction algorithm and the normal vector estimation and adjustment algorithm to generate a three-dimensional model of the ore powder pile.

[0060] More specifically, it can be divided into two algorithms: normal vector estimation and adjustment, and Poisson surface reconstruction.

[0061] The specific process of normal vector estimation and adjustment is as follows:

[0062] The feature information extracted based on the Transformer architecture is used to automatically estimate the normal vectors of the point cloud to obtain more accurate results. The Transformer architecture learns the internal structure of the point cloud and selects a certain number of neighboring points around each point to estimate the normal vectors. At the same time, the algorithm's robustness can be ensured based on the Transformer architecture, that is, the influence of noise and outliers in the data is fully filtered to ensure the stability of the results. After the normal vectors are estimated, a smoothing and adjustment algorithm is applied to ensure the consistency and correct direction of the normal vectors. By comparing the normal vectors of adjacent points, incorrect directions can be detected and corrected to ensure that the normal vector directions in the entire dataset are consistent, so as to obtain better surface quality in Poisson reconstruction. Subsequently, the Poisson surface reconstruction algorithm is used to convert the point cloud data after normal vector estimation and adjustment into a continuous three-dimensional surface, and the specific process of Poisson surface reconstruction is as follows:

[0063] Based on the position information and normal vectors of the points, a smooth and continuous three-dimensional surface is generated by solving the Poisson equation, and the octree obtained based on S2 is analyzed. In the octree, each node represents a spatial region and contains the normal vector information of the points within this region. For each node of the octree, the divergence is calculated according to its normal vector. The divergence reflects the flow direction and intensity of the vector field in a specific region.

[0064] Among them, the expression for calculating the divergence is as follows:

[0065]

[0066] Among them, N(p) is the neighborhood of point p, which contains other points within a certain range from point p. This distance is given by the actual scenario as a hyperparameter, and the present invention does not limit this distance;

[0067] is the normal vector of point p, and ∣q - p∣ is the Euclidean distance between point q and point p;

[0068] In Poisson surface reconstruction, the core lies in solving the Poisson equation;

[0069] Specifically, the Poisson equation defines a distributed vector field through the normal vectors and position information of the points;

[0070] The Poisson equation for calculating the continuous three-dimensional surface near point p is as follows:

[0071]

[0072] Among them, f(p) is the three-dimensional surface function of the point p to be generated, representing the position of the surface;

[0073] is the Laplace operator, representing the second-order derivative in space;

[0074] When solving the Poisson equation, the algorithm searches for a vector field that is consistent with the normal vector of the points in the entire space and forms a smooth surface on this vector field;

[0075] While generating a smooth surface, Poisson reconstruction can tolerate noise and holes in the data and is suitable for processing the irregular surface of the ore powder pile.

[0076] S4. Based on the volume calculation module, use the volume calculation algorithm based on voxel representation to calculate the volume of the ore powder; the volume calculation includes using the voxel representation method based on Delaunay triangulation and Laplace smoothing. The specific method for calculating the volume of the ore powder is as follows:

[0077] Based on Delaunay triangulation and Laplace smoothing, convert the accurately constructed three-dimensional model into voxel representation to accurately calculate the volume of the model. Finally, perform volume calculation on the three-dimensional model to ensure the accuracy and reliability of the calculation results;

[0078] More specifically, it is divided into three parts: mesh generation, mesh optimization, and volume calculation.

[0079] 1) Mesh generation: Generate a three-dimensional mesh model from the surface reconstructed in S3 through the triangulation method. The specific method for mesh generation is as follows:

[0080] Generate a complete three-dimensional structure through the triangulation method. Specifically, use the Delaunay Triangulation algorithm to divide the three-dimensional surface into a series of triangles;

[0081] For any point P on the smooth surface generated by Poisson reconstruction, the circumcenter coordinates satisfying the Delaunay condition can be calculated according to the following expression:

[0082]

[0083] where, (x P , y P ) are the circumcenter coordinates satisfying the Delaunay condition; (x1, y1), (x2, y2), (x3, y3) are the coordinates of the three vertices of the Delaunay triangle;

[0084] Use the divide-and-conquer algorithm to implement the Delaunay triangulation algorithm. The steps are as follows: First, divide the point set into two parts based on the binary tree until it cannot be divided any further; then perform triangulation on each node of the binary tree; then merge the two parts based on the binary tree merging algorithm, and adjust the edges and triangle structures to ensure that the Delaunay condition is satisfied;

[0085] Optimize the mesh, delete redundant vertices and edges, and improve the stability and efficiency of the model.

[0086] 2) The method for grid optimization is specifically as follows:

[0087] Analyze the connection relationships of each vertex, and delete duplicate and redundant vertices;

[0088] Specifically, for the edges, ensure that each edge is connected to only two triangles, thus maintaining the coherence of the grid;

[0089] Subsequently, use Laplacian Smoothing to eliminate sharp corners and abnormal protrusions in the grid, improving the smoothness and stability of the grid. For any vertex vi on the smooth surface, the basic expression of its Laplacian Smoothing is:

[0090]

[0091] where N(ν i ) is the set of vertices adjacent to vertex ν i , |N(ν i )| represents the number of adjacent vertices, and ν′ i is the position of the vertex after smoothing;

[0092] Based on the basic expression of Laplacian Smoothing, the application process of Laplacian Smoothing can be obtained: First, for each vertex ν i on the smooth surface, calculate the average position of its adjacent vertices based on the basic expression of Laplacian Smoothing;

[0093] Subsequently, adjust the position of vertex ν i to the calculated average position until the given convergence value δ;

[0094] Among them, the convergence value δ is related to the actual scenario, so the present invention does not limit the magnitude of the convergence value δ.

[0095] 3) The specific method for volume calculation is as follows:

[0096] For the finally generated three-dimensional grid model, use the voxelization method to convert it into a voxel representation for accurate volume calculation;

[0097] Preferably, apply the mathematical integration method to calculate the volume of the entire model, that is, the volume of the ore powder pile. The specific process is as follows:

[0098] First, divide the three-dimensional grid after grid optimization into a series of tetrahedrons, and each tetrahedron is composed of a triangle and a reference point;

[0099] By calculating the volumes of these tetrahedrons and adding them up, the volume of the entire model is obtained;

[0100] The calculation of the tetrahedron volume is based on the coordinates of points in a three-dimensional grid model, and the calculation formula is:

[0101]

[0102] Among them, the coordinates of the four points involved in this formula are (x1, y1, z1), (x2, y2, z2), (x3, y3, z3), and (x4, y4, z4) respectively.

[0103] S5. Analysis of calculation results, specifically: Based on historical picture data, generate the volume of the three-dimensional structure of the ore powder as the discriminator, and use the output of S4 as the generation end of the generative adversarial network (GAN). If the output result of the discriminator does not conform to the actual situation, S1 - S4 will be carried out again.

[0104] More specifically, analyze the calculation results of the ore powder based on the generative adversarial network (GAN);

[0105] Specifically, this GAN network verifies and improves the quality of the ore powder volume calculation results through the adversarial training of the generative model and the discriminative model. The specific steps are as follows:

[0106] First, the present invention trains the GAN network based on the data in the ore powder database. The data includes the real pictures and volume values of the ore powder pile; the present invention does not limit the implementation manner of the ore powder database;

[0107] The generator generates the volume of a certain ore powder pile based on S1 - S4, and the discriminator tries to distinguish between the generated data and the real data;

[0108] If the error between the volume of a certain ore powder pile generated by S1 - S4 and the volume output by the discriminator is less than ∈, it means that the real calculation result is relatively accurate. At this time, the result will be output, and this group of data will be transmitted to the database for storage to be used for offline training of the model later to improve the accuracy of subsequent analysis; otherwise, it indicates that there is a large error in the calculation result;

[0109] The error ∈ between the estimated volume of the ore powder pile and the volume output by the discriminator is related to specific requirements, and the present invention does not limit the specific value of the error ∈ between the estimated volume of the ore powder pile and the volume output by the discriminator;

[0110] At the same time, the calculation result analysis module transmits a recalculation instruction to the point cloud processing system to re - perform the point cloud reconstruction work.

[0111] It can be seen Figure 3 As shown, a method for measuring the volume of ore powder based on three - dimensional shape data analysis further includes a three - dimensional shape data measurement system for measuring the volume of ore powder, specifically:

[0112] a. A scanning module configured to collect the point cloud data of the ore powder pile using three - dimensional scanning technology;

[0113] b. A data processing module, configured to denoise, filter, and register the collected point cloud data. Further, the data processing module is configured to apply a graph-based sparse coding denoising algorithm and an octree data compression algorithm;

[0114] c. A feature extraction and model construction module, configured to process the preprocessed point cloud data using a deep learning-based algorithm and construct a three-dimensional structure model. Further, the feature extraction and model construction module is configured to use a Poisson surface reconstruction algorithm and a Transformer-based deep learning architecture;

[0115] d. A volume calculation module, configured to calculate the volume of the mineral powder pile based on the constructed three-dimensional model. Further, the volume calculation module is configured to use a voxel representation method based on Delaunay triangulation and Laplacian smoothing;

[0116] e. An analysis module, configured to analyze and optimize the calculation results using a generative adversarial network (GAN). The generative adversarial network (GAN) includes a generative model to generate the calculation results of the mineral powder volume and a discriminative model to distinguish the generated calculation results from the actual calculation results.

[0117] Technical terms that need to be explained and are helpful for understanding the present invention

[0118] Octree: An octree is a tree-shaped data structure based on space partitioning, often used for data organization and management in three-dimensional space. In the three-dimensional scanning of a mineral powder pile, the point cloud data is often extremely large and dense, requiring effective compression and standardization strategies for subsequent processing and analysis. When processing point cloud data, the data volume is usually extremely large, directly affecting storage and processing efficiency. The octree effectively organizes and compresses data by recursively dividing the three-dimensional space into smaller subspaces, while providing the ability for fast query and search.

[0119] Transformer architecture: The Transformer mainly consists of a multi-head self-attention mechanism, a feedforward network, and a layer normalization layer. The self-attention mechanism can extract global and local features from the point cloud data and identify complex relationships between points. This helps to capture the specific structure and shape of the mineral powder pile.

[0120] Structure and Principle of GAN: GAN is a deep learning architecture composed of a Generator and a Discriminator. The Generator is used to generate data similar to real data, and the Discriminator is used to distinguish between generated data and real data. The two improve the quality of the generated data through adversarial training. The goal of the Generator is to generate results of mineral powder volume calculation similar to real data. The input of the Generator is usually a noise vector, and through a multi-layer neural network, it generates an output similar to the real calculation result. The task of the Discriminator is to distinguish between generated data and real data. It receives the output of the Generator and the actual calculation result, and analyzes the features of the data through a multi-layer neural network to determine whether the input data is real

[0121] As described above, it is only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent replacements made by those of ordinary skill in the art to the technical solution of the present invention should be covered within the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention

Claims

1. A method for calculating the volume of ore powder based on three-dimensional shape data analysis, characterized in that: It includes the following steps: S1. Based on the ore powder pile scanning module, scan the ore powder pile and collect three-dimensional point cloud data; S2. Based on the point cloud preprocessing module, preprocess the collected point cloud data. The methods of point cloud data preprocessing include denoising, filtering, and registration to ensure the quality and accuracy of the data. The specific methods of point cloud data preprocessing are as follows: Use the graph-based sparse coding denoising algorithm for denoising, and remove model noise and redundant details by maximizing the sparsity optimization function. At the same time, based on the octree data compression algorithm, perform data compression. Apply the denoised three-dimensional model data to the octree data compression algorithm, and represent the high-density data in the form of a hierarchical octree to reduce the storage and processing overhead of the data. And based on the Gaussian correction method, perform standardization processing on the compressed data to ensure the consistency and comparability of the data, and provide a unified data format and standard for subsequent analysis and calculation; S3. Based on the ore powder three-dimensional structure model construction module and using the three-dimensional reconstruction algorithm, reconstruct the preprocessed point cloud data. The method of point cloud data reconstruction is as follows: Generate a three-dimensional model of the ore powder pile based on the Poisson three-dimensional plane reconstruction algorithm. At the same time, construct a model based on the deep learning architecture of Transformer, extract data features through the Transformer architecture, and then use the Poisson surface reconstruction algorithm and the normal vector estimation and adjustment algorithm to generate a three-dimensional model of the ore powder pile; S4. Based on the volume calculation module, use the volume calculation algorithm based on voxel representation to calculate the volume of the ore powder. The specific method of ore powder volume calculation is as follows: Based on Delaunay triangulation and Laplace smoothing, convert the accurately constructed three-dimensional model into voxel representation to accurately calculate the volume of the model. Finally, calculate the volume of the three-dimensional model to ensure the accuracy and reliability of the calculation results; S5. Analysis of calculation results, specifically: Based on historical picture data, generate the volume of the three-dimensional structure of the ore powder as a discriminator, and use the output of S4 as the generation end of the generative adversarial network (GAN). If the output result of the discriminator does not conform to the actual situation, repeat S1 - S4.

2. The method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 1, wherein: The specific content of S2 is as follows: Step 1. Perform point cloud denoising. The specific method of point cloud denoising is as follows: Based on the graph-based sparse coding denoising algorithm, construct a graph between data points to represent the local structure of the data, and then use sparse coding to recover and remove noise. Further, the graph-based sparse coding denoising algorithm regards data points as nodes in the graph, and the connections between nodes represent the similarity and distance relationships between data points, effectively removing the noise introduced by mechanical vibrations and environmental factors during the scanning process; Step 2. Perform sparse coding. The specific method of sparse coding is as follows: For the ore powder point cloud map after point cloud denoising, each node uses its neighbor nodes as a dictionary and performs sparse coding again; Sparse coding finds a sparse coefficient vector by solving an optimization problem. This vector can linearly combine neighbor nodes to approximately represent the target node, and the optimization goal is to maximize sparsity; Step 3: Perform data compression and standardization. An octree data structure is adopted to compress the point cloud data. This method effectively preserves the key geometric information. Specifically: The octree data structure first divides the entire three-dimensional space into a cube, and then recursively divides each cube into eight sub-cubes. Each sub-cube corresponds to a different region of the space. This process continues until the preset resolution and maximum tree depth are reached. Further, the point cloud data is then stored in the corresponding tree nodes, effectively classifying the originally unordered point cloud data into the corresponding sub-spaces; in the tree nodes of each sub-cube, a certain amount of point cloud data can be stored. When the data volume exceeds the preset threshold, further segmentation will be triggered. Among them, the tree nodes can store the indices of the points to reduce data redundancy and accelerate the subsequent query and operation speeds. Further, in each sub-cube, duplicate and redundant points will be checked and removed to reduce data redundancy. According to actual needs, outliers beyond the predetermined range can also be deleted to further optimize the data; By adjusting the maximum depth of the octree, the resolution of the data can be controlled. This resolution is related to the specific usage scenario. Preferably, in each node, a specific data standardization process is applied to ensure that the data is on the same scale and within the same range, facilitating subsequent calculations and reducing the calculation errors caused by scale differences; Step 4: Adopt a point cloud feature extraction technology based on deep learning to perform feature extraction. The improved version of the PointNet network, PointNet++, is used to capture the geometric and topological features of the ore powder pile to more effectively process non-uniformly sampled data. Specifically: PointNet++ learns the features of point cloud data through a hierarchical method, thereby improving the accuracy and processing ability of feature recognition. First, the point cloud is segmented into multiple local regions, and feature learning is independently performed within each region. This segmentation is based on the multi-scale grouping (MSG) strategy, allowing the model to capture features at different scales; in each of the segmented local regions, the model uses a small PointNet network to extract point features. This process covers multi-layer perceptron (MLP) and max pooling operations to extract features from each point and aggregate the information within the region; subsequently, through the error backpropagation algorithm, the PointNet++ network recursively applies these processing steps to the multi-level structure of the entire point cloud, gradually expanding from local to global, and finally forming a comprehensive description of the point cloud features.

3. The method for measuring the volume of ore powder based on three-dimensional shape data analysis according to claim 2, wherein: In the graph-based sparse coding denoising algorithm in the above Step 1, each point in the ore powder point cloud map is first regarded as a node of the graph. Subsequently, based on the distance and similarity between points, edges are established to connect adjacent points; and the 2-norm based on Lagrangian penalty is used to measure the similarity between points. Its expression is as follows: where x j are the three-dimensional coordinates of points i and j respectively, and λ is the penalty factor. This strategy effectively cancels out the noise caused by mechanical vibration; After performing similarity measurement, the system selectively sets thresholds and nearest neighbor methods to determine the specific number of neighbors connected to each node, thereby constructing a local structure diagram of the point cloud; Use the K-nearest neighbor algorithm to determine the specific number of neighbors connected to each node.

4. The method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 2, wherein: The optimization objective in step 2 can be achieved through L1-norm regularization, and the expression of this optimization objective is as follows: Among them, is the neighbor point set of point x i , a i is the coefficient vector of point x i , is the reconstruction error term, which ensures that the reconstructed x i is as similar as possible to the original x i , that is, the reconstruction error is as small as possible; The sparse regularization term ‖a i ‖1 promotes most elements to be zero through the l1 norm, thus achieving sparsity.

5. A method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 2, characterized in that: The data normalization process in step 3 can be performed according to the following expression: where y i,j is the j-th feature of the octree node i, and μj is the average value of the j-th feature of all octree nodes. It is easy to obtain from the sampling theorem that the variance of the j-th feature of all octree nodes is Therefore is the j-th feature of the standardized octree node i.

6. The method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 1, characterized in that: S3 can be divided into two algorithms: normal vector estimation and adjustment, and Poisson surface reconstruction; the specific process of normal vector estimation and adjustment is as follows: Based on the feature information extracted by the Transformer architecture, the normal vector of the point cloud is automatically estimated to obtain more accurate results. The Transformer architecture learns the internal structure of the point cloud and selects a certain number of neighboring points around each point to estimate the normal vector. At the same time, based on the transformer architecture, the robustness of the algorithm can be guaranteed, that is, the influence of data noise and outliers can be fully filtered out to ensure the stability of the results. After normal vector estimation, a smoothing and adjustment algorithm is applied to ensure the consistency and correct direction of the normal vector. By comparing the normal vectors of adjacent points, incorrect directions can be detected and corrected to ensure that the normal vector directions in the entire dataset are consistent, so as to obtain better surface quality in Poisson reconstruction. Subsequently, the Poisson surface reconstruction algorithm is used to convert the point cloud data after normal vector estimation and adjustment into a continuous three-dimensional surface; And the specific process of Poisson surface reconstruction is as follows: Based on the position information and normal vector of the points, a smooth and continuous three-dimensional surface is generated by solving the Poisson equation, and the octree obtained from S2 is analyzed. In the octree, each node represents a spatial region and contains the normal vector information of the points within this region. For each node of the octree, the divergence is calculated according to its normal vector. The divergence reflects the flow direction and intensity of the vector field in a specific region.

7. A method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 6, characterized in that: The expression of the divergence calculation is as follows: Among them, N(p) is the neighborhood of point p, which contains other points within a certain range from point p, and this distance is given as a hyperparameter by the actual scenario; is the normal vector of point p, and ∣q - p∣ is the Euclidean distance between point q and point p; In Poisson surface reconstruction, the core lies in solving the Poisson equation; Specifically, the Poisson equation defines a distributed vector field through the normal vector and position information of the points; The Poisson equation for calculating the continuous three-dimensional surface near point p is as follows: Among them, f(p) is the three-dimensional surface function of point p to be generated, representing the position of the surface; is the Laplace operator, representing the second derivative in space; When solving the Poisson equation, the algorithm searches for a vector field consistent with the normal vector of the points in the entire space and forms a smooth surface on this vector field; While generating a smooth surface, Poisson reconstruction can tolerate noise and holes in the data and is suitable for processing the irregular surface of the ore powder pile.

8. A method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 1, characterized in that: S4 is divided into three parts: mesh generation, mesh optimization, and volume calculation, 1) Mesh generation generates a three-dimensional mesh model from the surface reconstructed in S3 through the triangulation method. The method of mesh generation is as follows: Generate a complete three-dimensional structure through triangulation. Specifically, use the Delaunay Triangulation algorithm to divide the three-dimensional surface into a series of triangles; For any point P on the smooth surface generated by Poisson reconstruction, the coordinates of the circumcenter of the circumcircle that satisfies the Delaunay condition can be calculated according to the following expression: Among them, (x P , y P ) are the coordinates of the circumcenter of the circumcircle that satisfies the Delaunay condition; (x1, y1), (x2, y2), and (x3, y3) are the coordinates of the three vertices of the Delaunay triangle; Use the divide-and-conquer algorithm to implement the Delaunay triangulation algorithm. The steps are as follows: First, divide the point set into two parts based on a binary tree until it cannot be divided any further; then perform triangulation on each node of the binary tree; then merge the two parts based on the binary tree merging algorithm, and adjust the edges and triangle structures to ensure that the Delaunay condition is satisfied; Optimize the mesh, delete redundant vertices and edges, and improve the stability and efficiency of the model; 2) The method of mesh optimization is specifically as follows: Analyze the connection relationships of each vertex and delete duplicate and redundant vertices; Specifically, for the edges, ensure that each edge is only connected to two triangles, thus maintaining the coherence of the mesh; Subsequently, use Laplacian Smoothing to eliminate sharp corners and abnormal protrusions in the mesh, and improve the smoothness and stability of the mesh. For any vertex vi on the smooth surface, the basic expression of its Laplacian smoothing is: where, N(ν i ) is the set of vertices adjacent to vertex ν i , |N(ν i )| represents the number of adjacent vertices, and ν′ i is the position of the vertex after smoothing; Based on the basic expression of Laplacian smoothing, the application process of Laplacian smoothing can be obtained: First, for each vertex ν on the smooth surface i , calculate the average position of its adjacent vertices based on the basic expression of Laplacian smoothing; Subsequently, the position of vertex ν i is adjusted to the calculated average position until a given convergence value δ. 3) The specific method of volume calculation is as follows: For the finally generated three-dimensional mesh model, use the voxelization method to convert it into a voxel representation for accurate volume calculation; Preferably, apply the mathematical integration method to calculate the volume of the entire model, that is, the volume of the ore powder pile. The specific process is as follows: First, divide the three-dimensional mesh after mesh optimization into a series of tetrahedrons. Each tetrahedron is composed of a triangle and a reference point; Calculate the volumes of these tetrahedrons and add them up to obtain the volume of the entire model; The calculation of the tetrahedron volume is based on the coordinates of the points in the three-dimensional mesh model, and the calculation formula is: Among them, the coordinates of the four points involved in this formula are (x1, y1, z1), (x2, y2, z2), (x3, y3, z3), (x4, y4, z4).

9. A method for measuring the volume of mineral powder based on three-dimensional shape data analysis according to claim 1, characterized in that: The specific content of S5 is as follows: Analyze the calculation results of the ore powder based on the Generative Adversarial Network (GAN); Specifically, this GAN network verifies and improves the quality of the ore powder volume calculation results through the adversarial training of the generative model and the discriminative model. The specific steps are as follows: First, train the GAN network based on the data in the ore powder database. The data includes the real pictures and volume values of the ore powder pile; The generator generates the volume of a certain ore powder pile based on S1-S4, and the discriminator tries to distinguish between the generated data and the real data; If the error between the volume of a certain ore powder pile generated by S1-S4 and the volume output by the discriminator is less than ∈, it means that the real calculation result is relatively accurate. At this time, output the result, transmit this group of data to the database for storage for subsequent offline training of the model to improve the accuracy of subsequent analysis; otherwise, it indicates that there is a large error in the calculation result; The error ∈ between the estimated volume of the ore powder pile and the volume output by the discriminator is related to specific requirements; Meanwhile, the calculation result analysis module transmits a recalculation instruction to the point cloud processing system to re - perform the point cloud reconstruction work.

10. A method for calculating the volume of ore powder based on three-dimensional shape data analysis according to claim 1, characterized in that: It also includes a three - dimensional shape data measurement system for measuring the volume of mineral powder, specifically: a. A scanning module configured to collect point cloud data of the mineral powder pile using three - dimensional scanning technology; b. A data processing module configured to denoise, filter, and register the collected point cloud data. Among them, the data processing module is further configured to apply a graph - based sparse coding denoising algorithm and an octree data compression algorithm; c. A feature extraction and model construction module configured to process the pre - processed point cloud data using a deep - learning - based algorithm and construct a three - dimensional structure model. Among them, the feature extraction and model construction module is further configured to use a Poisson surface reconstruction algorithm and a deep - learning architecture based on Transformer; d. A volume calculation module configured to calculate the volume of the mineral powder pile based on the constructed three - dimensional model. Among them, the volume calculation module is further configured to use a voxel representation method based on Delaunay triangulation and Laplacian smoothing; e. An analysis module configured to analyze and optimize the calculation results using a generative adversarial network (GAN). The generative adversarial network (GAN) includes a generative model to generate the calculation results of the mineral powder volume and a discriminative model to distinguish the generated calculation results from the actual calculation results.

Images