Method and device for automatic extraction and volume measurement of ore granularity based on three-dimensional point cloud

Abstract

This invention discloses a method and apparatus for automatic extraction of ore particle size and volume measurement based on 3D point clouds, belonging to the field of mining and industrial automation technology. The method includes: acquiring 3D point cloud data of an ore material scene and preprocessing the 3D point cloud data; classifying each point in the preprocessed 3D point cloud data using an improved PointNet++ network, and outputting instance segmentation results for a single ore particle; wherein the improved PointNet++ network introduces an attention mechanism in the encoder and skip connections in the decoder; and calculating the volume and particle size information of the ore based on the instance segmentation results output by the improved PointNet++ network. Compared with existing methods, this invention has higher segmentation accuracy and volume measurement accuracy.

Description

Technical Field

[0001] This invention relates to the field of mining and industrial automation technology, and in particular to a method and apparatus for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds. Background Technology

[0002] With the rapid development of the economy and industry, the iron and steel industry faces the problem of excessive energy consumption in ore beneficiation. Iron and steel smelting requires a large amount of iron ore. When the grade and chemical properties of iron ore mined from the same deposit are constant, ore particle size, as a key parameter directly affecting the permeability, heat transfer, and mass transfer conditions of the furnace charge, has a significant impact on blast furnace smelting efficiency and energy consumption. Screening based on ore particle size can provide key indicators for selecting beneficiation methods and processes. Therefore, quantitative measurement of ore volumetric parameters is an important foundation for optimizing production processes and management.

[0003] Traditional methods for ore particle size analysis primarily employ sieving, which physically separates the ore using sieves with different aperture sizes. While this method is direct and reliable, it suffers from drawbacks such as high labor intensity, low efficiency, inability to perform online detection, and health risks to workers exposed to high dust environments.

[0004] In recent years, machine vision-based methods have been increasingly applied to ore particle size detection, with the main approaches as follows: I. Detection methods based on two-dimensional images Two-dimensional image processing methods acquire images of the ore surface using a camera, extract the ore contour using image segmentation algorithms (such as the watershed algorithm and edge detection), and estimate the granularity using the contour area. However, this method has the following fundamental drawbacks: 1. Two-dimensional images lose depth information, making it difficult to accurately estimate the true volume and particle size of irregularly shaped and stacked ores; 2. It is greatly affected by light conditions. Dust and water stains on the surface of the ore will change its appearance, making it difficult to divide. 3. Unable to handle severely occluded scenes; when ores are stacked on top of each other, the 2D image cannot distinguish the relationship between the front and back.

[0005] II. Three-dimensional point cloud detection methods

[0006] With the development of 3D sensing technology, some studies have applied 3D point clouds to ore particle size detection.

[0007] Patent application CN202310123456.7 discloses a method for detecting ore particle size based on three-dimensional point cloud. It uses traditional clustering algorithms (such as Euclidean distance clustering) to segment the ore point cloud. However, this method has low segmentation accuracy for cohesive ores and is prone to oversegmentation or undersegmentation. Furthermore, it is not optimized for the irregular shape of the ore.

[0008] Patent application CN202210987654.3 discloses an online ore particle size detection system and method, which uses deep learning to segment point clouds. However, the network structure is simple and does not fully consider the local geometric features of the point cloud (normal vector, curvature, etc.), resulting in limited segmentation accuracy.

[0009] Patent application CN202110345678.9 discloses a machine vision-based ore particle size analyzer that uses binocular vision to reconstruct the three-dimensional point cloud of the ore. However, the volume estimation uses a simple bounding box method or integration method, which does not consider the problem of point cloud incompleteness caused by occlusion, resulting in large measurement errors.

[0010] In summary, existing methods mainly suffer from the following three key technical bottlenecks: 1. Insufficient point cloud segmentation accuracy. The ore has irregular shape, rough surface, and is stacked and adhered to each other. Existing segmentation methods (Euclidean distance clustering, region growing, simple PointNet++, etc.) are difficult to accurately separate individual ore particles, resulting in serious oversegmentation and undersegmentation phenomena.

[0011] 2. Lack of effective models for incomplete volume estimation. During 3D imaging, due to occlusion and viewing angle limitations, the acquired mineral point clouds are usually incomplete (approximately 20%-40% surface missing). Existing methods have not established a quantitative relationship between the incompleteness rate and volume error, and directly using the slicing method or integration method to calculate the volume will produce significant errors (up to 15% or more).

[0012] 3. Lack of algorithm optimization for ore scenarios. Existing methods are mostly direct applications of general point cloud processing algorithms, without being specifically optimized for the geometric characteristics of ores (irregularity, multi-scale, surface roughness). Summary of the Invention

[0013] This invention provides a method and apparatus for automatic extraction of ore particle size and volume measurement based on three-dimensional point cloud, in order to solve the technical problems of low segmentation accuracy and large error in incomplete volume estimation in the prior art.

[0014] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, the present invention provides an automatic method for ore particle size extraction and volume measurement based on three-dimensional point clouds, the method comprising: Acquire 3D point cloud data of the ore material scene and preprocess the 3D point cloud data; An improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data and output the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder and a skip connection in the decoder. Based on the instance segmentation results output by the improved PointNet++ network, the volume and particle size information of the ore are calculated.

[0015] Furthermore, acquiring the three-dimensional point cloud data of the ore material scene includes: Three-dimensional cameras are installed at different angles and positions above the conveyor belt used to transport ore materials to form a ring-shaped multi-view coverage of the ore material scene to be photographed. A 3D camera was used to capture images of the ore material scene, resulting in 3D point cloud data of the ore material scene.

[0016] Furthermore, when using a 3D camera to photograph the ore material scene and obtain the 3D point cloud data of the ore material scene, when the ore material is photographed by 3D cameras from multiple perspectives at different locations, the iterative nearest point algorithm is used to register and fuse the point clouds from different perspectives. As the ore material moves on the conveyor belt, the speed of the conveyor belt is obtained through an encoder. And combined with camera frame rate Obtain the time difference from the image registration results. Motion compensation is performed on point clouds collected at different times, and the compensation formula is as follows: ; in, The original point cloud coordinates; The coordinates of the point cloud after compensation.

[0017] Further, the three-dimensional point cloud data is preprocessed, including: For the aforementioned 3D point cloud data, statistical filtering is used to remove outliers; For 3D point cloud data that has undergone statistical filtering, voxel filtering is used for downsampling.

[0018] Furthermore, the attention mechanism is a channel attention mechanism.

[0019] Furthermore, the improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data, outputting the instance segmentation result of a single ore, including: Calculate the normal vector and curvature information of each point in the preprocessed 3D point cloud data; The 3D coordinates, normal vector information, and curvature information of each point in the preprocessed 3D point cloud data are input into the improved PointNet++ network. The improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data and output the instance segmentation result of a single ore. Among them, the loss function of the improved PointNet++ network The calculation formula is: ; in, Cross-entropy loss for point-by-point classification; Weighted loss for boundary points; This is the point cloud density regularization loss; and This is the balance coefficient.

[0020] Furthermore, the calculation of ore volume and grain size information based on the instance segmentation results output by the improved PointNet++ network includes: Based on the preprocessed 3D point cloud data, a random sampling consensus algorithm is used to fit the reference plane that carries the ore material. The instance segmentation results output by the improved PointNet++ network are optimized by using an adaptive Euclidean distance resegmentation algorithm based on density clustering to obtain the instance point cloud of a single ore. Using the reference plane as the reference plane, the volume of each ore instance point cloud is estimated by slicing. A joint volume compensation model based on slice defect rate weighting and least squares surface fitting is constructed to correct the volume of the defect point cloud, thereby obtaining the volume of the ore. Based on the ore volume, the particle size information of the ore is obtained.

[0021] Furthermore, the optimization of the instance segmentation results output by the improved PointNet++ network using the density-based adaptive Euclidean distance resegmentation algorithm includes: Calculate the average point spacing within each point cloud block after segmentation. ; For two adjacent point cloud blocks, calculate the Euclidean distance between the nearest points of the blocks. ; like Then the two point cloud blocks will be merged; where, This is the merging coefficient; The average point spacing within each point cloud block The Eps parameter is used as the parameter for the DBSCAN algorithm, and the logarithm of the number of points within a block is used as the MinPts parameter. The DBSCAN algorithm is used to cluster a single point cloud block. If multiple sub-clusters exist within the same point cloud block after clustering and the distance between the sub-clusters is large, the clustering is considered successful. Then the corresponding point cloud block will be split into multiple independent ore point clouds; among them, is the splitting coefficient.

[0022] Furthermore, the construction of a joint volume compensation model based on slice incompleteness rate weighting and least squares surface fitting corrects the volume of the incomplete defect cloud to obtain the volume of the ore, including: The incompleteness rate of a single slice is calculated using the following formula: ; in, Indicates the first j The defect rate of each slice; For the first j The farthest height from the reference plane in a slice point cloud; For the first j The complete height of each slice; Calculate the weighted function for slice incompleteness rate : ; in, The incompleteness impact coefficient is obtained by calibration using standard samples. c The defect rate of the slice is defined, and its value ranges from [0,1]. Calculate the weighted volume : ; in, The volume estimated using the slicing method; For each ore point cloud, a quadratic surface equation is used for fitting, and the surface coefficients are solved using the least squares method to obtain the fitted surface. The fitted surface is then used to predict the height of occluded areas, and the height prediction values ​​are used to complete the missing point cloud. The volume of the completed point cloud is then calculated. ; Calculate the final volume of the ore. : ; in, To achieve weight fusion, the higher the completeness of the point cloud, the better. The larger the value, the better.

[0023] On the other hand, the present invention also provides an automatic ore particle size extraction and volume measurement device based on three-dimensional point clouds, the device comprising: The 3D point cloud acquisition module is used to acquire 3D point cloud data of ore material scenes; The data preprocessing module is used to preprocess the three-dimensional point cloud data; The point cloud segmentation module is used to classify each point in the preprocessed 3D point cloud data using an improved PointNet++ network and output the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder of the PointNet++ network and introduces skip connections in the decoder. The volume and grain size information calculation module is used to calculate the volume and grain size information of ore based on the instance segmentation results output by the improved PointNet++ network.

[0024] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0025] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above-described method.

[0026] The beneficial effects of the technical solution provided by this invention include at least the following: 1. High segmentation accuracy: This invention achieves accurate segmentation of complex stacked and adhered ores by improving the PointNet++ network (attention mechanism + skip connection + joint loss) and the adaptive resegmentation algorithm (merging and splitting based on density clustering), with a correct separation rate of over 94%.

[0027] 2. Accurate volume measurement: This invention proposes a joint compensation model of incompleteness rate weighting and surface fitting, which reduces the volume measurement error from more than 15% in traditional methods to less than 5% to address the point cloud incompleteness problem caused by occlusion in 3D imaging.

[0028] 3. High adaptability: Key parameters such as merging coefficient, splitting coefficient, and fusion weight are adaptively adjusted according to the point cloud density and degree of incompleteness, without the need for manual intervention, and adaptable to ores with different particle size distributions and stacking degrees.

[0029] 4. High industrial availability: Supports multi-view fusion, motion compensation and protective design, and can achieve continuous operation 24 / 7 in harsh mining environments, meeting the online detection needs of mineral processing plants.

[0030] 5. Significant economic benefits: Compared to manual screening, the technical solution provided by this invention improves efficiency by a hundredfold; compared to two-dimensional vision methods, it significantly improves accuracy. Through real-time particle size detection and feedback, crushing and grinding process parameters can be optimized, reducing energy consumption by 5%-10% and increasing recovery rate by 2%-3%. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the execution flow of the automatic extraction and volume measurement method for ore particle size based on three-dimensional point cloud provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of training sample annotation provided in an embodiment of the present invention; Figure 3 This is a diagram showing the improved PointNet++ network segmentation effect provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the adaptive Euclidean distance re-segmentation optimization strategy provided in an embodiment of the present invention; Figure 5 These are optimized segmentation effect diagrams provided by embodiments of the present invention; where (a) is before improvement and (b) is after improvement; Figure 6 This is a schematic diagram of the volume calculation using the slicing method provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the defect rate model provided in an embodiment of the present invention; Figure 8 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0034] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0035] First Embodiment

[0036] To address the issues of low accuracy and susceptibility to stacking occlusion in existing 2D vision methods, and insufficient segmentation accuracy and lack of effective models for incomplete volume estimation in existing 3D point cloud methods, this embodiment provides an automatic ore particle size extraction and volume measurement method based on 3D point clouds. First, 3D point cloud data of the ore scene is acquired and preprocessed. Second, a point cloud semantic segmentation based on an improved PointNet++ multi-scale feature fusion network is proposed, combining normal vector enhancement and attention mechanisms to achieve instance segmentation of individual ore particles. Then, an adaptive Euclidean distance re-segmentation algorithm based on density clustering is proposed for secondary optimization segmentation of adhered ores. Finally, a volume compensation model based on slice incompleteness rate weighting and least-squares surface fitting is proposed for high-precision volume estimation of incomplete point clouds. This method can be implemented by an electronic device, such as a terminal or server. The execution flow of this method is as follows: Figure 1 As shown, it includes the following steps: S1, acquire three-dimensional point cloud data of the ore material scene, and preprocess the three-dimensional point cloud data; Specifically, in this embodiment, the implementation process of S1 is as follows: S11, three-dimensional cameras are installed at different angles and positions above the conveyor belt used to transport ore materials to form a ring-shaped multi-view coverage of the ore material scene to be photographed. S12 uses a 3D camera to photograph the ore and obtain 3D point cloud data of the ore material scene; In the process of capturing ore material scenes using a 3D camera to obtain 3D point cloud data of the ore material scene, when the ore material is captured by 3D cameras from multiple perspectives at different locations, an iterative nearest-point algorithm is used to register and fuse the point clouds from different perspectives. The objective function of the iterative nearest-point algorithm is:

[0037] in, Let be a rotation matrix. It is a translation vector. and For corresponding point pairs, singular value decomposition is used to solve for the optimal transformation.

[0038] In addition, the conveyor belt speed is obtained via an encoder as the ore material moves on the conveyor belt. And combined with camera frame rate Based on the image registration results, motion compensation is performed on the point clouds acquired at different times, using the following formula: ; in, The time difference is the time interval between the time of the original point cloud acquisition and the time of the target reference. The original point cloud coordinates; The coordinates of the point cloud after compensation.

[0039] S13, statistical filtering is used to remove outliers; where, for each point Search its radius r Neighborhood k Calculate the average distance from each of the three nearest points. Let the global average distance be... The standard deviation is For average distances exceeding the global average distance Add λ times the standard deviation Points that are identified as outliers are removed; that is: if Then determine Outliers are identified and removed. This is the threshold coefficient, typically ranging from 1.0 to 3.0; S14 uses voxel filtering for downsampling, dividing the point cloud space into sections with sides of length [missing information]. l A voxel grid is constructed, with the centroid point within each voxel retained as a representative point. Among these, l The value is taken as 1 / 10 to 1 / 5 of the smallest size of the ore.

[0040] S2 uses an improved PointNet++ network to classify each point in the preprocessed 3D point cloud data and outputs the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder of the PointNet++ network and a skip connection in the decoder. It should be noted that the original point cloud input is... In this embodiment, the improved PointNet++ network input point cloud data includes normal vector information and curvature information. The normal vector is calculated by using principal component analysis to calculate the neighborhood covariance matrix of each point, and using the eigenvector corresponding to the smallest eigenvalue as the normal vector of that point. The curvature is calculated using the eigenvalues ​​of the covariance matrix. The calculation process is as follows: Let the point... The neighborhood point set is covariance matrix Solve eigenvalues The corresponding feature vector , , . ,in, Let be the three eigenvalues ​​of the covariance matrix; for example, with (The eigenvector corresponding to the smallest eigenvalue) is used as the normal vector. Curvature calculation: The three components of the normal vector and curvature value As an additional data dimension and point cloud coordinates Together they form a seven-dimensional input feature vector.

[0041] The attention mechanism module is a channel attention module (SENet structure), used to adaptively weight different feature channels. Its calculation formula is as follows:

[0042] in, Input feature map (dimension) GAP stands for Global Average Pooling (output dimension). MLP stands for Multilayer Perceptron. For the Sigmoid activation function, This indicates element-wise multiplication.

[0043] The PointNet++ encoder consists of multiple Set Abstraction Layers (SA layers) stacked together, with each SA layer containing three sub-modules: Sampling Layer: The center point is selected using Farthest Point Sampling (FPS). Grouping Layer: Query neighboring points within a specified radius, using each center point as the center; PointNet layer: applies shared MLP and max pooling to each local neighborhood to extract local features; The encoder in this embodiment adopts a three-stage SA structure, and the specific parameters are shown in Table 1: Table 1 Encoder Structure

[0044] In this embodiment, the channel attention module (SENet structure) is inserted in the following three locations: Location 1: After the PointNet layer of each SA layer and before max pooling; That is, after feature extraction of each local neighborhood is completed and before local neighborhood aggregation is performed, for each point... Channel attention weighting is applied to the 3D feature vectors.

[0045] Let the input of a certain SA layer be ( One central point, 3D coordinates (Dimensional features), after sampling and grouping, to obtain Each of the following local neighborhoods contains There are points, and the feature dimension is [number]. The feature matrix of each neighborhood is .

[0046] The channel attention process for handling this feature matrix is ​​as follows: Step 1: Feature Aggregation Global average pooling is performed on the feature matrix of each local neighborhood to obtain the global feature descriptor for that neighborhood. :

[0047] Step 2: Learning the Importance of Channels

[0048] Channel weights are learned through a two-layer fully connected network:

[0049] in: The first layer is fully connected, achieving dimensionality reduction. (reduction rate); It is a ReLU activation function; The second layer is fully connected, enabling dimensionality enhancement; The Sigmoid activation function outputs... ; Step 3: Feature Recalibration Learned channel weights Multiply back the original feature matrix channel by channel:

[0050] Step 4: Aggregate Output

[0051] Max pooling is then applied to the recalibrated features to obtain the final features for this local neighborhood.

[0052] This feature is input into the next SA layer.

[0053] Location 2: Feature transfer point between SA layers

[0054] That is, before the output features of the previous layer SA are input into the next layer SA, channel attention recalibration is performed on the global features.

[0055] Let the feature matrix output by the previous layer SA be... ,in, The number of the center point, This determines the output feature dimension. The channel attention processing flow is as follows: Step 1: For Perform global average pooling to obtain global feature descriptors. :

[0056] Step 2: Learn channel weights using the SENet structure :

[0057] Step 3: Perform channel weighting on the features of each center point:

[0058] Position 3: Before the final output of the encoder

[0059] That is, after the output features of the last layer SA (SA3) and before the input to the decoder, channel attention recalibration is performed on the final global features of the encoder. The processing flow is the same as at position 2, and the output is the weighted global features.

[0060] In addition, features from the corresponding layer of the encoder are introduced into the decoder, and multi-scale feature fusion is achieved through skip connections, which enhances the segmentation capability of boundaries and small-sized ores.

[0061] In this embodiment, the PointNet++ decoder uses a feature propagation (FP) structure, which gradually restores the sparse features output by the encoder to the original point cloud density through distance-based interpolation, and finally outputs the classification result for each point.

[0062] Specifically, the decoder in this embodiment adopts a three-level feature propagation layer structure, which forms a symmetrical "U-shaped" structure with the encoder's three-level set abstraction layer (SA1, SA2, SA3).

[0063] 1. FP3 layer (first-level feature propagation)

[0064] enter: High-level features from SA3: ,in , ; Skip connection characteristics from SA2: ,in , ; Processing steps: Step 1: Distance-based interpolation upsampling The features of 32 points in SA3 are upsampled to 128 points (the number of points in SA2). For the upsampling target points... The interpolation feature calculation formula is as follows:

[0065] in: The number of nearest neighbors. ; Distance weights ; Step 2: Feature stitching The interpolated features are concatenated with the skip connection features of SA2 along the channel dimension:

[0066] Step 3: Feature Transformation

[0067] The spliced ​​features are transformed and dimensionality reduced using a multilayer perceptron (MLP):

[0068] in, The structure is as follows: Layer 1: Linear layer (768 → 384) + BatchNorm + ReLU; Layer 2: Linear layer (384 → 256) + BatchNorm + ReLU; The role of skip connections is to fuse the local detail features of SA2 with the global semantic features of SA3, enabling the decoder to obtain both global contextual information and local geometric details, thereby enhancing its ability to identify ore boundaries.

[0069] 2. FP2 layer (second-level feature propagation)

[0070] enter: Features from FP3: ; Skip connection characteristics from SA1: ,in , ; Processing steps: Step 1: Interpolation Upsampling The features of 128 points in FP3 are upsampled to 512 points (the number of points in SA1). The same interpolation method as FP3 is used. (inversely weighted by distance).

[0071] Step 2: Feature stitching

[0072] Concatenate the interpolated features with the features of SA1:

[0073] Step 3: Feature Transformation

[0074] Structure: Layer 1: Linear layer (384 → 256) + BatchNorm + ReLU; Layer 2: Linear layer (256 → 128) + BatchNorm + ReLU; 3. FP1 layer (third-level feature propagation) enter: Features from FP2: ; Original input point cloud coordinates: ,in, ; Processing steps: Step 1: Interpolation Upsampling The features of 512 points in FP2 are upsampled to the original input point cloud of 4096 points.

[0075] Step 2: Feature stitching

[0076] The interpolated features are then concatenated with the original input point cloud coordinates.

[0077] Step 3: Feature Transformation

[0078] Structure: Layer 1: Linear layer (131 → 128) + BatchNorm + ReLU; Layer 2: Linear layer (128 → 64) + BatchNorm + ReLU; 4. Output Layer Input: FP1; Output ; deal with:

[0079]

[0080] in, The number of classification categories (ore / background, or multiple defect classifications).

[0081] Structure: Linear layer (64 → 32) + BatchNorm + ReLU; Linear layer (32 → ); Output: The probability of each point belonging to each category. The category with the highest probability is taken as the predicted label for that point.

[0082] The improved PointNet++ network uses a joint loss function, which includes cross-entropy loss, boundary-aware loss, and density consistency loss. The calculation formula is as follows:

[0083] in, For point-by-point classification cross-entropy loss, Weighted loss for boundary points, The loss is the point cloud density regularization loss. and This is the balance coefficient. Values , Values .

[0084] Cross-entropy loss: Used for point-by-point classification; among which, For the first i The ground truth label of each point takes the value of 0 or 1 (binary classification: 0 represents non-ore / background, 1 represents ore / defect) or multi-class one-hot encoded vector; For the first i The predicted probability of a point is the probability value of that point belonging to the positive class as output by the softmax layer of the network, and the value range is [0,1]. Boundary-aware loss: Boundary points are given higher weights. (like );in, For the first i The boundary weight coefficient for each point is used to increase the importance of boundary points in the loss function. Its value selection rules are as follows: For a point located on the boundary of an ore grain (i.e., where both ore and background exist in the neighborhood of that point), w i =2.0 5.0; For non-boundary interior points, w i =1.0; Boundary point determination method: If there are two different types of points in the neighborhood of radius r around the point, then it is determined to be a boundary point; For the first i The cross-entropy loss at point i represents the classification error at a single point. Density uniformity loss: The constraint ensures that the predicted point cloud density matches the actual density; among which, N The number of local regions (or the number of voxel grids) in the point cloud is calculated by dividing the point cloud space into N local regions / voxels and then comparing their densities. The predicted point cloud density for i local regions is the number of positive points predicted by the network to belong to that region divided by the volume of that region. The ground truth point cloud density of the i-th local region is the number of positive points in the region in the ground truth annotation divided by the volume of the region.

[0085] S3, based on the instance segmentation results output by the improved PointNet++ network, calculates the volume and particle size information of the ore; Specifically, in this embodiment, the implementation process of S3 is as follows: S31, based on the preprocessed 3D point cloud data, a random sampling consensus algorithm is used to fit the reference plane carrying the ore material, obtaining the equation of the reference plane and achieving preliminary separation of the background and the ore; specifically as follows: Determine the plane equation by randomly selecting three points. Calculate the distance from all points to the plane, and count the distances to interior points. ) quantity, iteration The plane with the most interior points is selected as the optimal plane. The final equation of the reference plane is:

[0086] S32 employs an adaptive Euclidean distance re-segmentation algorithm based on density clustering to optimize the instance segmentation results output by the improved PointNet++ network. It adaptively determines the clustering threshold by calculating the average point spacing within point cloud blocks, performing secondary separation on adhered ores to extract precise single-ores instance point clouds. The adaptive Euclidean distance re-segmentation optimization strategy is as follows: Figure 4 As shown, the details are as follows: S321, Calculate the average point spacing within each segmented point cloud block. ; For the segmented point cloud blocks The average internal point spacing is:

[0087] S322, For two adjacent point cloud blocks, calculate the Euclidean distance between the nearest points of the blocks. ; For adjacent blocks and Minimum distance between blocks .

[0088] S323, if Then merge and ;in, This is the merging coefficient, with a value ranging from 1.2 to 2.0; S324, for blocks DBSCAN clustering was used to... For radius, Clustering is performed to minimize the number of points. If the clustering result produces multiple clusters and the distance between the sub-clusters is... Then it splits. Take a value of 2.5~4.0. Take 2 to 5. The splitting coefficient ranges from 2.5 to 4.0. and It adapts to the average grain size and shape characteristics of the ore.

[0089] The optimized segmentation effect is as follows Figure 5 As shown.

[0090] S33, using the reference plane as the datum plane, the volume of the instance point cloud of each ore is estimated by slicing, and a joint volume compensation model based on slice defect rate weighting and least square surface fitting is constructed to correct the volume of the defect point cloud, obtain the volume of the ore, and obtain the particle size information of the ore based on the ore volume. Among them, volume estimation is performed using the slicing method, such as... Figure 6 As shown, specifically: along the direction perpendicular to the reference plane (z-axis), with thickness The ore point cloud is sliced ​​into equal-thickness sections. For each section, the point cloud within the section is projected onto a horizontal plane. The area of ​​the polygon is obtained by calculating the convex hull of the projected points or by Delaunay triangulation, and this polygon is used as the area of ​​the upper or lower surface of the section. The frustum approximation formula is used to calculate the first... j slice volume The initial volume of the ore is obtained by summing the volumes of all slices. .

[0091] Defect rate model such as Figure 7 As shown, the definition and calculation formula for the defect rate are as follows: Let the first In the point cloud slice, the farthest height from the reference plane is The minimum heights at the left and right boundaries of the slice are respectively and ; complete slice height Then the first Incompleteness of each slice ;in, This indicates that the slice is complete. The closer it is to 1, the more severe the defect.

[0092] The effect of incompleteness on volume is non-linear. A quadratic weighting function is used: ,in, c The completeness ratio of the slice, with a value range of [0,1], is used to quantitatively describe the completeness of the point cloud data of the slice. The incompleteness influence coefficient is obtained through standard sample calibration. The specific calibration steps are as follows: Preparation The preliminary volume of each of the standard ore samples with known volumes was calculated using the method of this invention. , with actual volume Comparison, solving optimization problems To obtain the optimal value.

[0093] Weighted volume: ;in, This is the volume estimated using the slicing method.

[0094] The least-squares surface fitting volume compensation model includes: for each ore point cloud, using a quadratic surface equation. Perform fitting, where, For horizontal coordinates, For the height; solve the surface coefficients using the least squares method; establish the error function: ;right By taking partial derivatives and setting them to zero, we solve the system of linear equations to obtain the surface coefficients. We then use the fitted surface to predict the height of the occluded region, and use these height predictions to complete the missing point cloud. Finally, we calculate the volume of the completed point cloud. ; The slicing volume and the surface fitting volume are weighted and fused using the following formula: ;in, To integrate the weighting coefficients, The value ranges from 0.5 to 0.8, and is adaptively adjusted according to the completeness of the point cloud. When the completeness is high... Take the largest value (e.g., 0.8) when the defects are severe. Take the smaller value (e.g., 0.5).

[0095] Ore particle size information is calculated using the equivalent sphere diameter or the equivalent cube side length: Equivalent sphere diameter ; Equivalent cube side length The sieve grade is determined according to the pre-set particle size classification standards (such as <10mm, 10-20mm, 20-30mm, 30-40mm, >40mm), and the particle size distribution histogram and statistical report are output.

[0096] In summary, this embodiment provides an automatic ore grain size extraction and volume measurement method based on 3D point clouds. It employs an improved PointNet++ semantic segmentation as the initial segmentation, followed by adaptive Euclidean distance resegmentation for optimization. The merging and splitting thresholds are dynamically adjusted based on the average ore grain size, solving the adaptability problem of fixed threshold methods. Furthermore, this invention improves PointNet++ in three aspects: expanding the input features from 3D to 7D enhances the network's ability to perceive local geometric structures; introducing a channel attention mechanism to achieve adaptive feature weighting; and adding skip connections to fuse multi-scale features. Experiments show that the segmentation IoU of the improved network increases from 78% to 89%. In addition, this invention proposes a joint volume compensation model: on the one hand, it performs nonlinear compensation of the slice volume through a residual rate weighting function; on the other hand, it performs geometric completion of occluded regions through least-squares surface fitting; finally, the two results are weighted and fused.

[0097] Based on the above, the automatic ore particle size extraction and volume measurement method of this embodiment has the following technical advantages: Segmentation accuracy advantage: Improved PointNet++ network (attention mechanism + skip connection + joint loss) + adaptive resegmentation algorithm to achieve end-to-end accurate ore instance segmentation with segmentation intersection-over-union (IoU) ratio of over 89%, which is 11 percentage points higher than the standard PointNet++.

[0098] Advantages in volume measurement accuracy: The unique defect rate weighting + surface fitting joint compensation model reduces the relative error of volume measurement from 15% to less than 5%, meeting the requirements of industrial applications.

[0099] Advantages of scene adaptability: The adaptive clustering threshold is dynamically adjusted according to the point cloud density, the surface fitting parameters are optimized according to the degree of incompleteness, and the fusion weight is adaptively selected according to the integrity, without the need for manual parameter tuning.

[0100] Industrial practical advantages: It supports multi-view point cloud registration and fusion, dynamic motion compensation and dust and shock protection, and can operate stably in harsh mining environments.

[0101] Experiments show that this model reduces the average error in volume measurement from 15% to less than 5%.

[0102] Second Embodiment

[0103] This embodiment uses a practical application case to illustrate the implementation effect of the method of the present invention.

[0104] The apparatus and method of this invention were installed at an iron ore beneficiation plant to perform online particle size detection on iron ore. The ore particle size range was 5-80 mm, the conveyor belt speed was 0.5 m / s, and the output was 200 tons / hour.

[0105] I. Hardware Deployment

[0106] 3D camera: Ruben Technology RVC-X structured light camera, with a depth resolution of 1280×1024, an accuracy of ±0.5mm@1m, and a frame rate of 30fps.

[0107] Installation method: Three cameras are installed at a 120° angle, 1.2m above the conveyor belt, to form a ring-shaped multi-view coverage and reduce point cloud incompleteness caused by single-view occlusion.

[0108] Protection: The camera is equipped with an IP65 protective housing and features compressed air cooling and purging.

[0109] Computing platform: Intel Core i9-12900K CPU (16 cores), NVIDIA RTX 4080 GPU (16GB VRAM), 64GB DDR5 memory.

[0110] II. Calibration

[0111] Camera intrinsic parameter calibration: A 12×9 checkerboard calibration board (30mm side length) was used to take 30 calibration images from different angles. The intrinsic parameter matrix and distortion coefficients were calculated using the Zhang Zhengyou calibration method.

[0112] Multi-camera extrinsic calibration: A standard sphere (50mm in diameter) is placed in the field of view, and point clouds from each camera are acquired. The rotation matrix between cameras is solved by minimizing the center deviation of the standard sphere's point cloud. Translation vector .

[0113] III. Model Training

[0114] Dataset Construction: Collect ore point clouds from 2000 different stacked scenarios (5-20 ore particles per scenario), and manually label the instance segmentation labels of each ore particle (using CloudCompare software), such as... Figure 2 As shown in the figure. Among them, there are 1600 training scenarios, 200 validation scenarios, and 200 testing scenarios.

[0115] Network parameters: Input points: 4096 points / scene Multi-scale grouping radii: 0.02, 0.04, 0.08 Batch size: 16 Initial learning rate: 0.001 (Adam optimizer, cosine annealing decay) Number of training rounds: 300 Training results: Validation set mIoU: 89.2% Test set mIoU: 87.6% Single frame segmentation time: approximately 45ms (GPU) IV. Operational Testing In a complete test, the conveyor belt transported approximately 500 ore particles, and the system processed them in real time; the segmentation effect was as follows. Figure 3 As shown in Table 2, the comparison results between the method of the present invention and the prior art are obtained.

[0116] Table 2 Comparison results of the method of the present invention and the prior art

[0117] V. Verification of Volume Compensation Effect

[0118] Fifty regularly shaped ore samples were selected, and their actual volumes were measured using the displacement method. The results of comparing different methods with existing volume calculation methods are shown in Table 3.

[0119] Table 3 Comparison results between the method of the present invention and existing volume calculation methods

[0120] VI. System Output

[0121] System real-time output: Particle size distribution histogram (5-10mm: 12%, 10-20mm: 35%, 20-30mm: 28%, 30-40mm: 15%, >40mm: 10%) Statistical indicators such as average particle size, D50, and D90 Oversized particle alarm (particles > 80mm) The test results are stored in a database, supporting historical queries and trend analysis.

[0122] Third Embodiment

[0123] This embodiment provides an automatic ore particle size extraction and volume measurement device based on three-dimensional point clouds. The device includes the following modules: The 3D point cloud acquisition module is used to acquire 3D point cloud data of ore material scenes; The data preprocessing module is used to preprocess the three-dimensional point cloud data; The point cloud segmentation module is used to classify each point in the preprocessed 3D point cloud data using an improved PointNet++ network and output the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder of the PointNet++ network and introduces skip connections in the decoder. The volume and grain size information calculation module is used to calculate the volume and grain size information of ore based on the instance segmentation results output by the improved PointNet++ network.

[0124] Furthermore, the apparatus of this embodiment may further include: The calibration module is used to perform intrinsic and extrinsic parameter calibration on the 3D camera. The intrinsic parameter calibration adopts the Zhang Zhengyou calibration method, and the extrinsic parameter calibration obtains the coordinate transformation matrix by collecting the point cloud of a standard sphere or standard block and comparing it with the known dimensions. The results display and storage module is used to display ore particle size distribution histograms and statistical reports, and store the test results in the database; The protective module includes a dustproof and waterproof housing, an active cooling system, and a vibration-resistant mounting bracket to adapt to the dusty, humid, and vibrating environment of a mining site.

[0125] The 3D point cloud acquisition module can use a structured light 3D camera, a binocular stereo vision camera, or a lidar. Specifically, in this embodiment, the 3D point cloud acquisition module includes multiple 3D cameras, which are installed above the conveyor belt at different angles and positions to form multi-view coverage, thereby reducing the point cloud incompleteness rate caused by occlusion.

[0126] It should be noted that the automatic ore particle size extraction and volume measurement device based on three-dimensional point clouds in this embodiment corresponds to the automatic ore particle size extraction and volume measurement method based on three-dimensional point clouds in the first embodiment described above. The functions implemented by each functional module in the automatic ore particle size extraction and volume measurement device based on three-dimensional point clouds in this embodiment correspond one-to-one with the process steps in the automatic ore particle size extraction and volume measurement method based on three-dimensional point clouds in the first embodiment described above; therefore, they will not be repeated here.

[0127] Fourth embodiment

[0128] This embodiment provides an electronic device, such as... Figure 8As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0129] Below, in conjunction with Figure 8 A detailed introduction to each component of this electronic device is provided below: The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0130] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 8 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0131] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0132] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or exist independently, and may be accessed through the interface circuit of the electronic device ( Figure 8 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0133] The transceiver may include a receiver and a transmitter. Figure 8 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and can be connected through the interface circuit of the electronic device (…). Figure 8 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0134] In addition, it should be noted that, Figure 8 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0135] Fifth embodiment

[0136] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0137] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0138] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0140] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0141] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0142] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0143] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0144] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0145] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments of the present invention have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make several improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds, characterized in that, include: Acquire 3D point cloud data of the ore material scene and preprocess the 3D point cloud data; An improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data and output the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder and a skip connection in the decoder. Based on the instance segmentation results output by the improved PointNet++ network, the volume and particle size information of the ore are calculated.

2. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 1, characterized in that, The acquisition of 3D point cloud data of the ore material scene includes: Three-dimensional cameras are installed at different angles and positions above the conveyor belt used to transport ore materials to form a ring-shaped multi-view coverage of the ore material scene to be photographed. A 3D camera was used to capture images of the ore material scene, resulting in 3D point cloud data of the ore material scene.

3. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 2, characterized in that, When using a 3D camera to capture the ore material scene and obtain the 3D point cloud data of the ore material scene, when the ore material is captured by 3D cameras from multiple perspectives at different locations, the iterative nearest point algorithm is used to register and fuse the point clouds from different perspectives. As the ore material moves on the conveyor belt, the speed of the conveyor belt is obtained through an encoder. And combined with camera frame rate Obtain the time difference from the image registration results. Motion compensation is performed on point clouds collected at different times, and the compensation formula is as follows: ； in, The original point cloud coordinates; The coordinates of the point cloud after compensation.

4. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 1, characterized in that, Preprocessing of the 3D point cloud data includes: For the aforementioned 3D point cloud data, statistical filtering is used to remove outliers; For 3D point cloud data that has undergone statistical filtering, voxel filtering is used for downsampling.

5. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 1, characterized in that, The attention mechanism is a channel attention mechanism.

6. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 1, characterized in that, The improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data, outputting the instance segmentation result of a single ore, including: Calculate the normal vector and curvature information of each point in the preprocessed 3D point cloud data; The 3D coordinates, normal vector information, and curvature information of each point in the preprocessed 3D point cloud data are input into the improved PointNet++ network. The improved PointNet++ network is used to classify each point in the preprocessed 3D point cloud data and output the instance segmentation result of a single ore. Among them, the loss function of the improved PointNet++ network The calculation formula is: ； in, Cross-entropy loss for point-by-point classification; Weighted loss for boundary points; This is the point cloud density regularization loss; and This is the balance coefficient.

7. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point clouds as described in claim 1, characterized in that, The calculation of ore volume and grain size information based on the instance segmentation results output by the improved PointNet++ network includes: Based on the preprocessed 3D point cloud data, a random sampling consensus algorithm is used to fit the reference plane that carries the ore material. An adaptive Euclidean distance resegmentation algorithm based on density clustering was used to optimize the instance segmentation results output by the improved PointNet++ network, resulting in the instance point cloud of a single ore. Using the reference plane as the reference plane, the volume of each ore instance point cloud is estimated by slicing. A joint volume compensation model based on slice defect rate weighting and least squares surface fitting is constructed to correct the volume of the defect point cloud, thereby obtaining the volume of the ore. Based on the ore volume, the particle size information of the ore is obtained.

8. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point cloud as described in claim 7, characterized in that, The optimization of instance segmentation results output by the improved PointNet++ network using an adaptive Euclidean distance resegmentation algorithm based on density clustering includes: Calculate the average point spacing within each point cloud block after segmentation. ; For two adjacent point cloud blocks, calculate the Euclidean distance between the nearest points of the blocks. ; like Then the two point cloud blocks will be merged; where, This is the merging coefficient; The average point spacing within each point cloud block The Eps parameter is used as the parameter for the DBSCAN algorithm, and the logarithm of the number of points within a block is used as the MinPts parameter. The DBSCAN algorithm is used to cluster a single point cloud block. If multiple sub-clusters exist within the same point cloud block after clustering and the distance between the sub-clusters is large, the clustering is considered successful. Then the corresponding point cloud block will be split into multiple independent ore point clouds; among them, is the splitting coefficient.

9. The method for automatic extraction of ore particle size and volume measurement based on three-dimensional point cloud as described in claim 7, characterized in that, The construction of a joint volume compensation model based on slice incompleteness rate weighting and least squares surface fitting corrects the volume of the incomplete defect cloud to obtain the volume of the ore, including: The incompleteness rate of a single slice is calculated using the following formula: ； in, Indicates the first j The defect rate of each slice; For the first j The farthest height from the reference plane in a slice point cloud; For the first j The complete height of each slice; Calculate the slice incompleteness weighting function : ； in, The incompleteness impact coefficient is obtained by calibration using standard samples. c The defect rate of the slice is defined, and its value ranges from [0,1]. Calculate the weighted volume : ； in, The volume estimated using the slicing method; For each ore point cloud, a quadratic surface equation is used for fitting, and the surface coefficients are solved using the least squares method to obtain the fitted surface. The fitted surface is then used to predict the height of occluded areas, and the height prediction values ​​are used to complete the missing point cloud. The volume of the completed point cloud is then calculated. ; Calculate the final volume of the ore. : ； in, To achieve weight fusion, the higher the completeness of the point cloud, the better. The larger the value, the better.

10. An automatic ore particle size extraction and volume measurement device based on three-dimensional point clouds, characterized in that, include: The 3D point cloud acquisition module is used to acquire 3D point cloud data of ore material scenes; The data preprocessing module is used to preprocess the three-dimensional point cloud data; The point cloud segmentation module is used to classify each point in the preprocessed 3D point cloud data using an improved PointNet++ network and output the instance segmentation result of a single ore. The improved PointNet++ network introduces an attention mechanism in the encoder of the PointNet++ network and introduces skip connections in the decoder. The volume and grain size information calculation module is used to calculate the volume and grain size information of ore based on the instance segmentation results output by the improved PointNet++ network.

Images