Ore grade detection method and system, electronic equipment and storage medium

Abstract

The invention provides an ore grade detection method and system, electronic equipment and a storage medium, and belongs to the technical field of ore detection.The method comprises the steps that ore weight features are collected, first ore features are obtained based on ore images, and second ore features are obtained based on three-dimensional point cloud data of ore; performing correlation analysis on the first ore feature and the second ore feature to obtain a correlation matrix; carrying out attention weight distribution on the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature distributed with an attention weight; inputting the ore weight features and the target ore features into an ore grade detection model to obtain an ore grade detection result; the ore grade detection model comprises a linear regression model and a decision tree model. According to the ore grade detection method and system, the electronic equipment and the storage medium provided by the invention, the accuracy of ore grade detection can be improved.

Description

Technical Field

[0001] The present application belongs to the field of ore detection technology, and more specifically, relates to an ore grade detection method and system, electronic equipment, and storage medium. Background Art

[0002] Ore grade testing is crucial in the mining industry, directly impacting ore value, mining feasibility, and selection of beneficiation processes. As resource development shifts toward lower-grade, complex, and inter-growth minerals, the shortcomings of traditional testing methods are becoming increasingly apparent.

[0003] While traditional chemical analysis offers high precision, it's cumbersome, requiring specialized personnel in a multi-step laboratory process. This process is destructive, consumes large amounts of sample, and is unsuitable for real-time production and rare mineral testing. Physical sorting techniques, while capable of initially separating ores of varying grades, suffer from low accuracy. Image processing methods also have drawbacks, being limited to large-scale ore data collection. This leads to inaccurate feature extraction and, consequently, low grade detection accuracy.

[0004] Therefore, there is an urgent need for more efficient and accurate ore grade detection methods. Summary of the Invention

[0005] The purpose of this application is to provide an ore grade detection method and system, electronic equipment, and storage medium to improve the accuracy of ore grade detection.

[0006] A first aspect of the embodiments of the present application provides a method for detecting ore grade, comprising: Collecting ore weight characteristics, obtaining a first ore characteristic based on an ore image, and obtaining a second ore characteristic based on three-dimensional point cloud data of the ore; performing correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation matrix; assigning attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature assigned with an attention weight; The ore weight characteristics and the target ore characteristics are input into an ore grade detection model to obtain an ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

[0007] A second aspect of the embodiments of the present application provides an ore grade detection system, comprising: an ore feature analysis module, configured to collect ore weight features, obtain a first ore feature based on an ore image, and obtain a second ore feature based on three-dimensional point cloud data of the ore; and perform correlation analysis on the first ore feature and the second ore feature to obtain a correlation matrix; an attention enhancement module, configured to assign attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature assigned with an attention weight; The ore grade detection module is used to input the ore weight characteristics and the target ore characteristics into the ore grade detection model to obtain the ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

[0008] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned ore grade detection method when executing the computer program.

[0009] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for detecting ore grade are implemented.

[0010] The beneficial effects of an ore grade detection method and system, electronic device, and storage medium provided in the embodiments of the present application are as follows: the embodiments of the present application cover all dimensions of surface, interior, and dynamics through two-dimensional images, three-dimensional point clouds, and weight features, thereby solving the problem of missing information from a single sensor, strengthening the association between physical quantities and grades, and realizing multi-dimensional feature fusion; on the other hand, the embodiments of the present application achieve feature optimization through correlation analysis and attention enhancement, so that more attention is paid to valuable features during grade detection, and the accuracy of grade detection can be improved while balancing information diversity and complexity; on the other hand, the linear regression and decision tree hybrid model adopted in the embodiments of the present application can complementarily capture linear and nonlinear patterns, and weighted fusion can improve the accuracy of grade detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0012] Figure 1 A schematic diagram of a flow chart of a method for detecting ore grade provided in one embodiment of the present application; Figure 2 This is a structural block diagram of an ore grade detection system provided in one embodiment of the present application; Figure 3 A schematic block diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0013] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0014] In order to make the purpose, technical solutions and advantages of this application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0015] Please refer to Figure 1 , Figure 1 This is a flow chart of a method for detecting ore grade provided in one embodiment of the present application. The method may be executed by an electronic device. Specifically, the method may include S101 to S103.

[0016] S101: Collect ore weight characteristics, obtain a first ore characteristic based on the ore image, and obtain a second ore characteristic based on the three-dimensional point cloud data of the ore; perform correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation matrix.

[0017] In this embodiment, obtaining the first ore feature based on the ore image specifically includes: performing enhancement processing on the collected ore image to obtain an enhanced ore image; The image edge detection algorithm is used to extract features from the enhanced ore image to obtain the ore contour features; Generate a gray level co-occurrence matrix based on the enhanced ore image, and extract ore texture features based on the gray level co-occurrence matrix; The ore contour feature and the ore texture feature are taken as the first ore feature.

[0018] In this embodiment, the ore weight feature refers to the ore weight distribution feature on the ore car, which is used to analyze the correlation between the ore accumulation morphology and the ore grade. The ore image is a two-dimensional visual data of the ore surface collected by industrial visual equipment, which is used to reflect the surface geometry and microtexture of the ore. The first ore feature refers to a multi-dimensional feature set composed of ore contour features and texture features. The second ore feature is a quantitative indicator extracted from the three-dimensional point cloud data that reflects the three-dimensional geometric structure, internal characteristics and dynamic behavior of the ore. The second ore feature may include the geometric and structural characteristics, dynamic characteristics, etc. of the ore. The correlation matrix is ​​a matrix that measures the degree of correlation between the first ore feature and the second ore feature. The matrix elements of the correlation matrix are the correlation indexes of the two features. If the first ore feature has M dimensions and the second ore feature has N dimensions, the matrix size is M×N; In this embodiment, the 3D point cloud data of the ore is a collection of 3D coordinates of discrete points on the ore surface acquired by a 3D scanning device. Each point contains (X, Y, Z) spatial coordinates. The 3D point cloud data can be used to restore the ore's true geometry through spatial sampling.

[0019] For example, this embodiment can utilize a lidar (lidar) to emit laser light and receive reflected signals, calculate distance using time-of-flight (TFO), and generate 3D point cloud coordinates based on the scanning angle. Alternatively, a structured light camera can be used to project fringes and analyze deformation to calculate depth information. Compared to 2D images, point cloud data can provide the 3D shape, internal structure, and dynamic changes of the ore, compensating for the lack of depth in 2D vision and providing key physical parameters such as volume, density, and porosity for grade detection.

[0020] In this embodiment, the image edge detection algorithm refers to extracting the boundary between the ore and the background by detecting the area with sudden grayscale changes in the image, which is used to quantify the geometric shape of the ore. The image edge detection algorithm may include the Canny algorithm, the Sobel algorithm, etc. The ore contour feature refers to the geometric parameters of the outer boundary of the ore extracted by edge detection, which is used to quantify the macroscopic shape characteristics of the ore. The grayscale co-occurrence matrix is ​​a matrix that statistics the joint distribution of the grayscale values ​​of pixel pairs in the image, and is used to quantify the microscopic characteristics of the ore surface texture. The ore texture feature is a statistic calculated based on the grayscale co-occurrence matrix, which can reflect the regularity and complexity of the grayscale distribution on the ore surface and is directly related to the mineral composition.

[0021] In this embodiment, the collected ore image is enhanced, specifically including: denoising the ore image using Gaussian filtering, contrast enhancing the denoised ore image using global histogram equalization, and binarizing the contrast-enhanced image to obtain an enhanced ore image.

[0022] For example, this embodiment can use an industrial area array camera and a ring-shaped LED fill light mounted on a mining vehicle to capture ore images. This embodiment can perform image enhancement processing on the ore images, including: using Gaussian blur to eliminate sensor noise; using CLAHE to enhance the grayscale difference between the ore and the background to improve image contrast; and using the Otsu thresholding method to separate the ore from the background.

[0023] The extraction of contour features in this embodiment may include: using the Canny algorithm to perform edge detection on the ore image, extracting the ore boundary, and outputting a binary edge map; this embodiment uses the findContours function of OpenCV to obtain the maximum connected area (i.e., the ore body) based on the binary edge map, and calculate the contour perimeter (number of pixels), contour area, and circularity.

[0024] The extraction of texture features in this embodiment may include calculating the weighted sum of squared grayscale value differences based on the grayscale co-occurrence matrix to obtain a contrast feature. The contrast feature can reflect the degree of grayscale value difference between adjacent pixels in the image and measure the clarity and layering of the texture. The higher the contrast, the more dramatic the grayscale variation in the ore surface texture, such as uneven particle size, which corresponds to the uneven mineral composition and is related to grade fluctuation. The ratio of the grayscale covariance to the standard deviation is then calculated to obtain a correlation feature. The correlation feature measures the linear correlation between the grayscale values ​​of adjacent pixels in the image and reflects the regularity of the texture. A high correlation indicates that the ore texture has strong directionality or periodicity, such as a layered structure, which is related to the mineral deposition or formation environment and can indirectly indicate the grade distribution characteristics. The sum of the squares of the elements of the grayscale co-occurrence matrix is ​​then calculated to obtain an energy feature. The energy feature is used to characterize the distribution concentration of elements in the GLCM matrix and reflects the uniformity of the texture. The higher the energy value, the more uniform the grayscale distribution on the ore surface, such as a fine-grained, dense structure, which corresponds to a higher-purity mineral composition and is related to grade stability. Finally, the information entropy of the probability distribution is calculated. Information entropy is used to measure the complexity and randomness of the texture in the image. The higher the information entropy value, the more complex the ore texture, such as multi-mineral mixing or pore development, corresponding to larger grade fluctuations or more impurities.

[0025] In this embodiment, the construction of the first ore feature may include: combining features such as contour perimeter, contour area, circularity and contrast image, correlation feature, texture energy, information entropy, etc. into a multidimensional feature vector as input for subsequent correlation analysis.

[0026] This embodiment ensures image quality through enhancement processing, extracts shape features through edge detection, and captures micro-texture differences through the grayscale co-occurrence matrix. Both describe the surface characteristics of the ore from the macro-shape and micro-structure dimensions respectively, and integrate key visual information related to grade, such as the degree of mineral crystallization affecting texture and the density affecting contour regularity, providing a reliable two-dimensional visual basis for multimodal feature fusion.

[0027] S102: Allocate attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature allocated with attention weights.

[0028] In this example, attention weight allocation refers to assigning importance weights to each feature based on the strength of the correlation between features, thereby highlighting key features and suppressing redundant or noisy features. The target ore features refer to the weighted feature set, which includes the filtered and weighted valid features from the first and second ore features, such as contour perimeter × 0.7 and porosity × 0.9.

[0029] In this example, the correlation between ore features reflects the degree of information overlap and complementarity. Strongly correlated features repeatedly describe the same attribute and require redundancy removal; moderately correlated features reflect surface and internal structure respectively and need to be retained and weighted; weakly correlated features independently influence grade and require basic weighting. By assigning weights, we can optimize the balance between information content and complexity within the feature set, thereby improving the predictive accuracy and generalization capabilities of subsequent models.

[0030] S103: Inputting the ore weight characteristics and the target ore characteristics into an ore grade detection model to obtain an ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

[0031] In this example, target ore features refer to a comprehensive feature set after attention weighting, where each feature has been assigned a weight. A linear regression model is a statistical model that shows a linear relationship between features and grade, with parameters representing the weight coefficients of each feature. A decision tree model is a tree-structured model that predicts grade by recursively partitioning the feature space.

[0032] In this embodiment, a complex relationship between ore grade and features exists, both linear and nonlinear. Linear regression is used to capture simple linear patterns. Decision trees are used to capture nonlinear rules, such as a 2% grade decrease when the porosity is greater than 10% and the texture entropy is greater than 5. This allows for adaptation to complex mineral distribution scenarios. The ore grade detection model in this embodiment is a hybrid model, using linear regression to process basic associations and decision trees to supplement nonlinear details. The resulting weighted fusion output balances bias and variance, improving prediction accuracy.

[0033] For example, the model's input features may include weight features, target shape features, and multi-dimensional weighted features such as contour area × 0.6, texture energy × 0.5, and porosity × 0.8. This embodiment performs a weighted summation of the linear regression output and the decision tree output to obtain the final grade detection result.

[0034] For example, linear regression model training can include setting the objective function to minimize the mean squared error between the predicted grade and the actual grade. Using gradient descent and adding L2 regularization to prevent overfitting, the model ultimately obtains the weight coefficients and bias terms for each feature.

[0035] Decision tree model training can include: Splitting criteria: selecting features that minimize the variance of the grade of the child nodes, such as "porosity less than or equal to 8%". After splitting, the grade distribution of the child nodes is more concentrated; Pruning strategy: Determine the maximum depth by minimizing the error in the validation set to avoid overfitting; Output form: Generates a rule tree, such as "a porosity less than or equal to 8% and a texture entropy less than or equal to 3 can result in a grade of 65%", which directly reflects the impact of feature combinations on grade.

[0036] It can be concluded from the above that the ore grade detection method provided in this embodiment achieves high-precision and intelligent detection of ore grade through multimodal data fusion, feature optimization and hybrid model building.

[0037] On the one hand, this embodiment uses two-dimensional images, three-dimensional point clouds and weight features to cover all dimensions of surface, interior and dynamics, solves the problem of missing information from a single sensor, strengthens the correlation between physical quantities and grades, and realizes multi-dimensional feature fusion; on the other hand, this embodiment eliminates redundancy and retains complementary features through correlation analysis, realizes feature optimization, and then combines image enhancement and point cloud segmentation to improve data quality, suppress noise, and balance information diversity and complexity; on the other hand, the linear regression and decision tree hybrid model adopted in this embodiment can complement each other to capture linear and nonlinear patterns, and weighted fusion improves the accuracy of grade detection.

[0038] In one embodiment of the present application, obtaining a second ore feature based on three-dimensional point cloud data of the ore includes: Dividing the three-dimensional point cloud data of the ore into multiple voxel grids, and selecting seed voxels from the multiple voxel grids; Taking seed voxels as the basic unit, the 3D point cloud data of the ore is segmented into ore solid areas and hole areas based on the region growing algorithm; Calculating the volume of the ore solid area and the volume of the hole area respectively, and calculating the ore porosity based on the volume of the ore solid area and the volume of the hole area; The ore porosity is taken as the second ore characteristic.

[0039] In this embodiment, the three-dimensional point cloud data of the ore is divided into multiple voxel grids, specifically including: determining a first length based on the volume of the ore and the point cloud data density of the three-dimensional point cloud data of the ore, and dividing the three-dimensional point cloud data of the ore into multiple voxel grids with a side length of the first length.

[0040] In this embodiment, the voxel grid refers to dividing the three-dimensional space into uniform cubic grids. The side length of each grid can be adjusted according to the size of the ore and the density of the point cloud. The ore volume and the point cloud density are positively correlated with the first length of the voxel grid. Ensure that the voxel grid is of appropriate size so that each voxel grid contains an appropriate amount of point cloud data while avoiding too many divided voxel grids that increase the amount of calculation. The seed voxel refers to the starting grid for regional growth and can be randomly selected. The region growing algorithm refers to starting from the seed voxel and merging adjacent and similar grids. Similarity refers to the point cloud direction being similar and the spatial distance being close, thereby distinguishing the ore entity area and the internal hole area. The porosity refers to the proportion of the internal hole volume of the ore to the total volume. A high porosity corresponds to a low-grade ore with a loose structure, and a low porosity corresponds to a dense high-grade ore.

[0041] In this example, the density of an ore's internal structure is directly related to its grade. High-grade ores, due to their pure mineral composition, tend to have a dense structure and low porosity. Low-grade ores, however, may contain more impurities or weathering-induced pores and have a high porosity. By dividing the 3D point cloud into a voxel grid and gradually merging similar grids starting with a seed voxel, the solid and void regions can be precisely segmented. The porosity can then be calculated, reflecting the ore's internal quality and providing a key basis for grade detection.

[0042] For example, this embodiment preprocesses the point cloud data, removing discrete points that significantly deviate from the overall structure and retaining the point cloud that truly reflects the ore shape. If the point cloud is too dense, such as in a small ore sample, the number of points can be reduced while retaining the overall shape, improving subsequent processing efficiency.

[0043] This example first sets the grid length based on the overall size of the ore and the density of the point cloud. For example, large ore loaded on trucks uses a grid with a side length of 5-10 cm, while small laboratory samples use a grid with a side length of 1-3 cm. This ensures that each grid contains 50-200 points. This avoids missing hole details due to overly large grids or a surge in computational complexity due to overly small grids. This example divides the ore's 3D point cloud data into sections based on the set grid lengths, with each grid containing the point cloud for the corresponding area.

[0044] This embodiment calculates the surface flatness (curvature) of the point cloud within each grid cell and selects the flattest grid cell with the largest number of point clouds as the seed voxel, ensuring that segmentation begins within the main ore area. Starting from the seed voxel, the point clouds of adjacent grid cells are checked for similar orientations, such as being approximately on the same plane, and for reasonable spatial distances, such as no more than twice the grid side length. If these conditions are met, the cells are merged into a solid area; otherwise, they are marked as holes. This process continues until all voxel grids have been processed.

[0045] This example calculates the volume of the solid area by counting the number of grids and multiplying it by the volume of a single grid. Similarly, the volume of the void area is calculated. The void volume is divided by the total ore volume to obtain the porosity as a percentage, which is a key characteristic reflecting the internal structure of the ore.

[0046] In this embodiment, the three-dimensional point cloud data of the ore includes three-dimensional point cloud data of the ore continuously acquired at multiple moments during the transportation process of the ore-carrying vehicle loaded with the ore; Obtaining a second ore feature based on the three-dimensional point cloud data of the ore further includes: constructing a surface model of the ore based on the three-dimensional point cloud data of the ore, and calculating a surface area and a surface curvature of the ore based on the surface model of the ore; Calculate the volume of ore based on the 3D point cloud data of the ore; Calculate the relative volume change rate of the ore based on the 3D point cloud data of the ore at adjacent moments; The ore surface area, ore surface curvature, ore volume and ore relative volume change rate are taken as the second ore characteristics.

[0047] In this embodiment, continuous three-dimensional point cloud data refers to a set of three-dimensional coordinates of the ore surface acquired by continuous scanning at a fixed time interval of once every 10 minutes using equipment such as a laser radar during ore transportation, which can record the morphological changes of the ore over time. A surface model refers to fitting a discrete point cloud into a continuous three-dimensional surface, such as a triangular mesh, for accurate calculation of geometric features. Surface area refers to the total area of ​​the ore surface model, reflecting the size and shape complexity of the ore. Large pieces of ore generally have a larger surface area. Surface curvature is used to measure the local unevenness of the ore surface, such as corners and hole edges. A larger curvature value indicates a rougher surface, such as a high curvature of the edge of a crushed ore. The relative volume change rate refers to the ratio of change in the volume of the ore at adjacent moments, reflecting changes in the crushing or stacking state of the ore due to vibration and collision during transportation.

[0048] In this example, physical changes in ore during transportation, such as crushing and rolling, alter its three-dimensional geometric characteristics, which are indirectly correlated with grade. The increased surface area and curvature of the crushed ore expose more of its internal structure. If the high-grade portion of the ore is harder and less susceptible to crushing, this surface area change can be used to correlate the degree of crushing with the grade distribution. A rapid decrease in volume indicates increased ore crushing, leading to the mixing of ores of different grades and affecting overall grade stability. By continuously monitoring these dynamic characteristics, the grade prediction model can be refined in real time, improving detection accuracy.

[0049] Exemplarily, this embodiment aligns the coordinate systems of point cloud data at different times to ensure that data at adjacent times can be directly compared. This embodiment uses methods such as median filtering to remove abnormal points, making the point cloud smoother and facilitating subsequent surface modeling. This embodiment can use Poisson reconstruction or triangulation algorithm to convert the point cloud into a continuous triangular mesh model, sum up the areas of all faces of the triangular mesh, and obtain the surface area of ​​the ore. For each vertex of the mesh, take the neighboring points within a radius of 5 cm around it, calculate the vertex curvature after fitting the plane, and take the overall average as the curvature of the ore surface. This embodiment calculates the volume of the ore by using the point cloud bounding box volume or voxelization method. Take the volume data of two adjacent moments, calculate the change and then divide it by the initial volume to obtain the rate of change in percentage form.

[0050] This embodiment uses the calculated surface area, surface curvature, volume, and relative volume change rate as the second ore feature, and combines them with static features such as porosity to form a multidimensional feature set that reflects the geometric morphology and dynamic behavior of the ore, which is input into subsequent correlation analysis and grade prediction models.

[0051] This embodiment extracts multi-dimensional secondary ore features from 3D point cloud data, enabling non-contact, real-time dynamic monitoring of ore structure and morphological changes. Porosity reflects internal quality, while dynamic features such as surface area correlate with crushing. Combined with continuous data, this allows for real-time correction of grade prediction models, improving detection accuracy and efficiency, reducing labor costs, and providing a comprehensive and reliable basis for ore quality assessment.

[0052] In one embodiment of the present application, a correlation analysis is performed on the first ore characteristic and the second ore characteristic to obtain a correlation matrix, including: Performing a correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation index between the first ore characteristic and the second ore characteristic; Construct a correlation matrix based on the correlation index.

[0053] In this embodiment, the first ore characteristics include ore contour characteristics and ore texture characteristics; the second ore characteristics include ore porosity, ore surface area, ore surface curvature, ore volume and ore relative volume change rate; performing a correlation analysis on the first ore feature and the second ore feature to obtain a correlation index between the first ore feature and the second ore feature, specifically comprising: extracting an ore contour perimeter, an ore contour area, and an ore circularity based on the ore contour feature; and constructing a contour feature subvector based on the ore contour perimeter, the ore contour area, and the ore circularity; Constructing texture feature sub-vectors based on ore texture features; Combining the contour feature sub-vector and the texture feature sub-vector to obtain a first feature vector; constructing a second eigenvector based on the ore porosity, the ore surface area, the ore surface curvature, the ore volume, and the relative volume change rate of the ore; The mutual information value between the first eigenvector and the second eigenvector is calculated, and the mutual information value is used as a correlation index.

[0054] In this embodiment, the mutual information value refers to an indicator that measures the dependency between features. A larger mutual information value indicates a stronger correlation, such as a strong positive correlation between contour area and volume, and a strong negative correlation between texture energy and porosity. A contour feature subvector refers to a set of geometric parameters extracted from the ore's two-dimensional image contour, used to quantify the ore's macroscopic shape characteristics. Contour feature subvectors can include contour perimeter, contour area, and circularity. Texture feature subvectors refer to a set of statistics extracted based on the gray-level co-occurrence matrix of the ore image, used to describe the microscopic texture characteristics of the ore surface. Texture feature subvectors can include contrast, energy, entropy, etc.

[0055] In this embodiment, the first eigenvector refers to a multidimensional vector that integrates contour and texture features and represents a comprehensive representation of 2D image features. It describes the ore surface characteristics from two dimensions: macroscopic shape and microscopic texture, providing 2D visual input for subsequent correlation analysis with 3D features. The second eigenvector, a multidimensional vector extracted from the ore's 3D point cloud data, reflects the ore's 3D structure, internal properties, and dynamic behavior.

[0056] In this example, the ore's two-dimensional image features (contour and texture) are intrinsically correlated with its three-dimensional structural features (porosity, surface area, and dynamic changes). The area of ​​the ore's contour is positively correlated with its three-dimensional volume; the energy of the ore's texture is negatively correlated with its internal porosity, with the more uniform the texture (i.e., the higher the energy), the lower the internal porosity. The circularity of the ore's contour is positively correlated with its volume change rate during transportation; irregular (lower circularity) ore is more prone to breakage, resulting in a higher volume change rate during transportation. This example uses mutual information analysis to quantify these nonlinear correlations and identify redundant and complementary features, such as the strong correlation between contour area and volume and the weak correlation between texture entropy and surface curvature. This provides a scientific basis for subsequent weight assignment and prevents invalid features from interfering with model accuracy and efficiency.

[0057] Exemplarily, this embodiment performs edge detection on the enhanced ore image, obtains the contour perimeter and contour area, and calculates the circularity. Based on the grayscale co-occurrence matrix, this embodiment calculates the contrast, energy, and entropy in each direction, and takes the mean as the texture feature. This embodiment combines the contour perimeter, area, circularity and texture features to form a 6-dimensional first eigenvector. This embodiment extracts the porosity, surface area, surface curvature, volume, and relative volume change rate from the three-dimensional point cloud processing results to form a 5-dimensional second eigenvector. This embodiment divides continuous features such as perimeter and volume into 5-10 intervals, for example, perimeter <1000 pixels is interval 1, 1000-2000 pixels is interval 2, and so on.

[0058] This embodiment permutes and combines the first eigenvector and all vectors in the second eigenvector to obtain multiple feature pairs. The number of times each feature pair co-occurs in different intervals is counted to form a joint frequency table. Simultaneously, the frequency of occurrence of individual features in each interval is counted to form an edge frequency table. This embodiment calculates the mutual information value for each feature pair using the joint frequency and edge frequency. A larger value indicates a stronger correlation.

[0059] This example pairs the six features of the first eigenvector with the five features of the second eigenvector, calculates mutual information values, and forms a 6-row, 5-column correlation matrix. For example, the mutual information value between contour area and volume is 0.8 (strong positive correlation); the mutual information value between texture energy and porosity is -0.6 (strong negative correlation); and the mutual information value between circularity and relative volume change rate is 0.2 (weak correlation).

[0060] This embodiment quantifies the nonlinear correlation between 2D image and 3D structural features through mutual information analysis, accurately identifying redundant and complementary features and avoiding interference from invalid information. The correlation matrix constructed in this embodiment provides a scientific basis for feature weight assignment, optimizes feature set quality, improves subsequent model prediction accuracy and operational efficiency, and enables efficient fusion and utilization of multimodal features.

[0061] In one embodiment of the present application, attention weights are assigned to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature assigned with attention weights, including: Traversing the correlation matrix, adding the first ore feature and the second ore feature that meet a first condition to the first candidate feature set; the first condition being: the absolute value of the correlation index between the first ore feature and the second ore feature is less than a first threshold and greater than a second threshold; adding the first ore feature and the second ore feature that meet the second condition to the second candidate feature set; the second condition being that the absolute value of the correlation index between the first ore feature and the second ore feature is less than a third threshold; Redundant feature screening is performed on the first ore feature and the second ore feature that meet a third condition to obtain a third candidate feature set; the third condition is that the absolute value of the correlation index between the first ore feature and the second ore feature is greater than a first threshold; the first threshold is greater than the second threshold, and the second threshold is greater than the third threshold; Adding the feature of the first ore feature and the second ore feature that does not satisfy the first condition, the second condition and the third condition at the same time to the fourth candidate feature set; Attention weights are allocated to the first candidate feature set, the second candidate feature set, the third candidate feature set, and the fourth candidate feature set respectively to obtain target ore features allocated with attention weights.

[0062] In this embodiment, the first candidate set includes moderately correlated feature pairs, whose absolute values ​​of the correlation index are between the second threshold and the first threshold, and the information is complementary but not redundant. The second candidate set includes weakly correlated feature pairs, whose absolute values ​​of the correlation index are less than the third threshold, which will independently affect the grade or contain noise. The third candidate set includes strongly correlated feature pairs, whose absolute values ​​of the correlation index are greater than the first threshold, and the information is highly overlapping. The fourth candidate set includes unclassified independent features, which have extremely low correlation with other features and need to be processed separately. The first threshold is a preset strong correlation critical value, the second threshold is a preset medium correlation lower limit, and the third threshold is a preset weak correlation upper limit. The thresholds can be set through experiments or domain knowledge.

[0063] This embodiment removes redundancy from strongly correlated features, retaining primary features and eliminating redundant features to avoid model overfitting. This embodiment also weights and complements moderately correlated features. For example, texture energy (surface uniformity) and porosity (internal structure) are moderately correlated, reflecting different dimensions and therefore need to be retained and assigned a moderate weight. This embodiment performs noise suppression on weakly correlated features. For example, weakly correlated features with no direct correlation to grade are assigned low weights to reduce interference. This embodiment retains basic weights for independent features, such as contour perimeter and surface curvature, which, despite low correlation, provide unique information and are therefore assigned basic weights. The specific distribution of weights can be preset in advance.

[0064] For example, this embodiment calculates a correlation index (i.e., mutual information value) for 30 feature pairs between the first and second feature vectors. Based on the mutual information value of each feature pair, this embodiment performs classification according to a preset threshold to obtain a first candidate feature set, a second candidate feature set, a third candidate feature set, and a fourth candidate feature set.

[0065] Among them, the first candidate feature set: feature pairs with an absolute value of correlation index greater than 0.8, such as contour area and volume, are marked as strongly correlated. The first candidate feature set: feature pairs with an absolute value of correlation index greater than 0.5 and less than or equal to 0.8, such as texture energy and porosity, are marked as moderately correlated. The second candidate feature set: feature pairs with an absolute value of correlation index less than or equal to 0.2, such as circularity and volume change rate, are marked as weakly correlated. The fourth candidate feature set: the remaining feature pairs, such as contour circumference and surface curvature, have a correlation index of 0.3-0.5 and are marked as independent features.

[0066] The weight allocation strategy for each feature candidate set in this embodiment includes: Third candidate set (strong correlation): For each feature pair, calculate the Pearson coefficient between the individual feature and ore grade, and retain the feature with the highest correlation. For example, if both contour area and volume are positively correlated with ore grade, with a correlation of 0.7 for volume and 0.6 for contour area, retain the volume and remove the contour area. The primary feature weight is set to 1.0, and redundant features are removed.

[0067] First candidate set (moderate correlation): Weights are assigned based on the mutual information value ratio. For example, texture energy (mutual information 0.6) and hole ratio (mutual information 0.7) have a total weight of 1.3, giving texture energy and hole ratio 0.46 and 0.54, respectively.

[0068] Second candidate set (weak correlation): All features are assigned the same low weight, such as 0.3, or adjusted according to the univariate importance of the feature to the grade. For example, the univariate importance of circularity is low, so it is weighted 0.2; the volume change rate is slightly higher, so it is weighted 0.35.

[0069] The fourth candidate set (independent features) is assigned a medium weight, such as 0.5, because its information is independent and can supplement dimensions not covered by other features, such as the contour perimeter reflecting the complexity of the two-dimensional shape.

[0070] This embodiment combines the processed features of each candidate set according to the weights to form the final target feature set. For example: The third candidate set retains the volume with a weight of 1.0; The first candidate set retains texture energy with a weight of 0.46 and hole rate with a weight of 0.54; The second candidate set retains the volume change rate with a weight of 0.35; The fourth candidate set retains the contour perimeter with a weight of 0.5 and the surface curvature with a weight of 0.5.

[0071] The final target feature set is: [volume × 1.0, texture energy × 0.46, porosity × 0.54, volume change rate × 0.35, contour perimeter × 0.5, surface curvature × 0.5].

[0072] This embodiment achieves feature optimization through a candidate set classification strategy. Strongly correlated features are de-redundant to prevent overfitting, moderately correlated features are weighted and complementary to preserve multidimensional information, weakly correlated features suppress noise, and independent features preserve essential information. Ultimately, this achieves efficient fusion of multimodal features, improving feature set quality, enhancing model prediction accuracy and generalization, and optimizing information utilization efficiency.

[0073] Corresponding to an ore grade detection method of the above embodiment, Figure 2 This is a structural block diagram of an ore grade detection system provided in one embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 2 The ore grade detection system 20 includes: an ore feature analysis module 21, an attention enhancement module 22 and an ore grade detection module 23.

[0074] The ore feature analysis module 21 is used to collect ore weight features, obtain a first ore feature based on the ore image, and obtain a second ore feature based on the 3D point cloud data of the ore; perform correlation analysis on the first ore feature and the second ore feature to obtain a correlation matrix; an attention enhancement module 22 for allocating attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature allocated with the attention weight; The ore grade detection module 23 is used to input the ore weight characteristics and target ore characteristics into the ore grade detection model to obtain the ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

[0075] In one embodiment of the present application, the ore feature analysis module 21 is specifically configured to perform enhancement processing on the collected ore image to obtain an enhanced ore image; The image edge detection algorithm is used to extract features from the enhanced ore image to obtain the ore contour features; Generate a gray level co-occurrence matrix based on the enhanced ore image, and extract ore texture features based on the gray level co-occurrence matrix; The ore contour feature and the ore texture feature are taken as the first ore feature.

[0076] In one embodiment of the present application, the ore feature analysis module 21 is further configured to divide the three-dimensional point cloud data of the ore into a plurality of voxel grids, and select seed voxels from the plurality of voxel grids; Taking seed voxels as the basic unit, the 3D point cloud data of the ore is segmented into ore solid areas and hole areas based on the region growing algorithm; Calculating the volume of the ore solid area and the volume of the hole area respectively, and calculating the ore porosity based on the volume of the ore solid area and the volume of the hole area; The ore porosity is taken as the second ore characteristic.

[0077] In one embodiment of the present application, the three-dimensional point cloud data of the ore includes three-dimensional point cloud data of the ore continuously acquired at multiple moments during the transportation of the ore by the ore-carrying vehicle; the ore feature analysis module 21 is further configured to construct a surface model of the ore based on the three-dimensional point cloud data of the ore, and calculate the surface area and the surface curvature of the ore based on the surface model of the ore; Calculate the volume of ore based on the 3D point cloud data of the ore; Calculate the relative volume change rate of the ore based on the 3D point cloud data of the ore at adjacent moments; The ore surface area, ore surface curvature, ore volume and ore relative volume change rate are taken as the second ore characteristics.

[0078] In one embodiment of the present application, the ore characteristic analysis module 21 is further configured to perform a correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation index between the first ore characteristic and the second ore characteristic; Construct a correlation matrix based on the correlation index.

[0079] In one embodiment of the present application, the first ore feature includes an ore contour feature and an ore texture feature; the second ore feature includes an ore porosity, an ore surface area, an ore surface curvature, an ore volume, and a relative volume change rate of the ore; the ore feature analysis module 21 is further configured to extract an ore contour perimeter, an ore contour area, and an ore circularity based on the ore contour feature; and construct a contour feature subvector based on the ore contour perimeter, the ore contour area, and the ore circularity. Constructing texture feature sub-vectors based on ore texture features; Combining the contour feature sub-vector and the texture feature sub-vector to obtain a first feature vector; constructing a second eigenvector based on the ore porosity, the ore surface area, the ore surface curvature, the ore volume, and the relative volume change rate of the ore; The mutual information value between the first eigenvector and the second eigenvector is calculated, and the mutual information value is used as a correlation index.

[0080] In one embodiment of the present application, the attention enhancement module 22 is specifically configured to traverse the correlation matrix and add the first ore feature and the second ore feature that meet a first condition to the first candidate feature set; the first condition is that the absolute value of the correlation index between the first ore feature and the second ore feature is less than a first threshold and greater than a second threshold; adding the first ore feature and the second ore feature that meet the second condition to the second candidate feature set; the second condition being that the absolute value of the correlation index between the first ore feature and the second ore feature is less than a third threshold; Redundant feature screening is performed on the first ore feature and the second ore feature that meet a third condition to obtain a third candidate feature set; the third condition is that the absolute value of the correlation index between the first ore feature and the second ore feature is greater than a first threshold; the first threshold is greater than the second threshold, and the second threshold is greater than the third threshold; Adding the feature of the first ore feature and the second ore feature that does not satisfy the first condition, the second condition and the third condition at the same time to the fourth candidate feature set; Attention weights are allocated to the first candidate feature set, the second candidate feature set, the third candidate feature set, and the fourth candidate feature set respectively to obtain target ore features allocated with attention weights.

[0081] See also Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided in one embodiment of the present application. Figure 3 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned system embodiments, such as Figure 2 The functions of the ore feature analysis module 21, the attention enhancement module 22 and the ore grade detection module 23 are shown.

[0082] It should be understood that in the embodiment of the present application, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0083] The input device 302 may include a touchpad, a fingerprint collection sensor (for collecting user fingerprint information and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.

[0084] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information on ore data.

[0085] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiment of the present application can execute the implementation method described in an embodiment of an ore grade detection method provided in the embodiment of the present application, and can also execute the implementation method of the electronic device 300 described in the embodiment of the present application, which will not be repeated here.

[0086] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, all or part of the process of the method in the above embodiment is implemented. The computer program can also be used to instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments are implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0087] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the computer-readable storage medium can include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or is about to be output.

[0088] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0089] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or can be an electrical, mechanical or other form of connection.

[0091] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0092] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0093] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for detecting ore grade, characterized in that: include: Collecting ore weight characteristics, obtaining a first ore characteristic based on the ore image, and obtaining a second ore characteristic based on the ore's three-dimensional point cloud data; performing a correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation matrix; assigning attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature assigned with an attention weight; The ore weight characteristics and the target ore characteristics are input into an ore grade detection model to obtain an ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

2. The method for detecting ore grade according to claim 1, wherein: The obtaining of the first ore feature based on the ore image includes: Performing enhancement processing on the collected ore image to obtain an enhanced ore image; The image edge detection algorithm is used to extract features from the enhanced ore image to obtain the ore contour features; generating a gray level co-occurrence matrix based on the enhanced ore image, and extracting ore texture features based on the gray level co-occurrence matrix; The ore contour feature and the ore texture feature are used as first ore features.

3. The method for detecting ore grade according to claim 1, wherein: The obtaining of the second ore feature based on the three-dimensional point cloud data of the ore includes: Dividing the three-dimensional point cloud data of the ore into a plurality of voxel grids, and selecting seed voxels from the plurality of voxel grids; Using the seed voxel as a basic unit, the three-dimensional point cloud data of the ore is segmented into ore solid areas and hole areas based on a region growing algorithm; Calculating the volume of the ore solid area and the volume of the hole area respectively, and calculating the ore porosity based on the volume of the ore solid area and the volume of the hole area; The ore porosity is used as the second ore characteristic.

4. The method for detecting ore grade according to claim 3, wherein: The three-dimensional point cloud data of the ore includes three-dimensional point cloud data of the ore continuously acquired at multiple moments during the transportation process of the ore-carrying vehicle loaded with the ore; The step of obtaining a second ore feature based on the three-dimensional point cloud data of the ore further includes: constructing a surface model of the ore based on the three-dimensional point cloud data of the ore, and calculating the surface area and surface curvature of the ore based on the surface model of the ore; Calculating the volume of the ore based on the three-dimensional point cloud data of the ore; Calculate the relative volume change rate of the ore based on the 3D point cloud data of the ore at adjacent moments; The ore surface area, the ore surface curvature, the ore volume and the relative volume change rate of the ore are used as second ore characteristics.

5. The method for detecting ore grade according to claim 1, wherein: The performing correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation matrix includes: performing a correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation index between the first ore characteristic and the second ore characteristic; A correlation matrix is ​​constructed based on the correlation index.

6. The method for detecting ore grade according to claim 5, wherein: The first ore characteristics include ore contour characteristics and ore texture characteristics; the second ore characteristics include ore porosity, ore surface area, ore surface curvature, ore volume and ore relative volume change rate; The performing correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation index between the first ore characteristic and the second ore characteristic includes: Extracting an ore contour perimeter, an ore contour area, and an ore circularity based on the ore contour feature; and constructing an outline feature subvector based on the ore contour perimeter, the ore contour area, and the ore circularity; Constructing a texture feature sub-vector based on the ore texture feature; Combining the contour feature sub-vector and the texture feature sub-vector to obtain a first feature vector; constructing a second feature vector based on the ore porosity, the ore surface area, the ore surface curvature, the ore volume, and the relative volume change rate of the ore; A mutual information value between the first eigenvector and the second eigenvector is calculated, and the mutual information value is used as the correlation index.

7. The method for detecting ore grade according to claim 1, wherein: The allocating attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature allocated with attention weights includes: Traversing the correlation matrix, adding a first ore feature and a second ore feature that meet a first condition to a first candidate feature set; the first condition being that an absolute value of a correlation index between the first ore feature and the second ore feature is less than a first threshold and greater than a second threshold; adding the first ore feature and the second ore feature that meet a second condition to a second candidate feature set; wherein the second condition is that the absolute value of the correlation index between the first ore feature and the second ore feature is less than a third threshold; performing redundant feature screening on the first ore feature and the second ore feature that meet a third condition to obtain a third candidate feature set; the third condition being that an absolute value of a correlation index between the first ore feature and the second ore feature is greater than a first threshold; the first threshold is greater than the second threshold, and the second threshold is greater than the third threshold; adding the features of the first ore feature and the second ore feature that do not satisfy the first condition, the second condition, and the third condition at the same time to a fourth candidate feature set; Attention weights are allocated to the first candidate feature set, the second candidate feature set, the third candidate feature set, and the fourth candidate feature set respectively to obtain target ore features allocated with attention weights.

8. An ore grade detection system, characterized in that: include: An ore feature analysis module is used to collect ore weight features, obtain a first ore feature based on an ore image, and obtain a second ore feature based on the ore's three-dimensional point cloud data; performing a correlation analysis on the first ore characteristic and the second ore characteristic to obtain a correlation matrix; an attention enhancement module, configured to assign attention weights to the first ore feature and the second ore feature based on the correlation matrix to obtain a target ore feature assigned with an attention weight; The ore grade detection module is used to input the ore weight characteristics and the target ore characteristics into the ore grade detection model to obtain the ore grade detection result; the ore grade detection model includes a linear regression model and a decision tree model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images