A system, method, electronic device and storage medium for intelligent beneficiation

Abstract

The application provides a system and method for intelligent ore dressing, an electronic device and a storage medium, which comprises the following steps: a conveying device uniformly distributes and arranges the ore to be selected in a single layer and then sends it to a detection area; a primary selection device collects the geometric features and color distribution features of the ore, and divides the ore into easy-to-select types and difficult-to-select types according to the features, and then sends the ore into a first reselection channel or a second reselection channel; the first reselection channel directly determines the concentrate or tailings of the easy-to-select types of ore through geometric feature and / or color distribution feature recognition; the second reselection channel collects the image features and spectral features of the difficult-to-select types of ore, associates the image features and spectral features to the same target ore based on the geometric features as a spatial reference, fuses the features to obtain fused features, and calculates the confidence of the sorting action, so as to perform sorting. Through hierarchical processing and multi-modal association fusion based on geometric reference, the application improves the sorting efficiency and accuracy, and is particularly suitable for complex and difficult-to-select ores.

Description

Technical Field

[0001] This application relates to the field of intelligent mineral processing technology, and more specifically, to an intelligent mineral processing system, method, electronic device, and storage medium. Background Technology

[0002] In the field of mineral processing, most beneficiation systems employ a uniform testing process for all ores, such as relying solely on color or a single sensor for judgment. They lack the ability to pre-classify ores based on easily obtainable characteristics like ore geometry, color, and texture, and allocate processing channels of varying complexity. This results in easily sortable ores being unnecessarily sent to costly, multi-sensor, complex testing stages, leading to decreased processing efficiency and resource waste. Furthermore, even when existing technologies collect data from multiple sensors, they often employ simple, fixed rules, making accurate sorting difficult. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a smart mineral processing system, method, electronic device and storage medium to overcome the problems in the prior art.

[0004] In a first aspect, embodiments of this application provide an intelligent mineral processing system, comprising: A conveying device is used to evenly distribute and arrange the ore to be processed in a single layer before conveying it to the testing area. A preliminary selection device is used to collect the geometric features and color distribution features of the ore to be selected in the detection area, and determine the ore type and corresponding subsequent processing channel of the ore to be selected based on the geometric features and the color distribution features; wherein, the subsequent processing channel includes a first re-selection channel and a second re-selection channel, and the process complexity of the second re-selection channel is higher than that of the first re-selection channel. The first secondary selection channel is used to identify the geometric features and / or color distribution features of the ore to be selected of the easy-to-select type, and determine whether the ore to be selected is concentrate or tailings. The second selection channel is used to collect image features and spectral features of the difficult-to-select type of ore to be selected, and associate the image features and spectral features with the geometric features as a spatial reference to the same target ore; calculate the sorting action confidence of the target ore based on the fusion features of the target ore, and sort the target ore based on the sorting action confidence; wherein, the fusion features are obtained by fusing the geometric features, image features and spectral features.

[0005] In some technical solutions of this application, the above-mentioned determination of the ore type and corresponding subsequent processing channel of the ore to be selected based on the geometric features and the color distribution features includes: The complexity index of the ore to be selected is calculated based on the geometric features and the color distribution features; wherein the complexity index characterizes the complexity of the geometric shape and color texture of the ore to be selected. The complexity index is compared with the complexity threshold to determine whether the ore to be selected is of an easy type and enters the first secondary selection channel, or to determine whether the ore to be selected is of a difficult type and enters the second secondary selection channel.

[0006] In some technical solutions of this application, the calculation of the complexity index of the ore to be selected based on the geometric features and the color distribution features includes: The geometric and color distribution characteristics of the ore to be selected are compared with the geometric and color distribution characteristics of a preset reference ore to obtain geometric similarity and color similarity. Based on the obtained geometric similarity and color similarity, the complexity index of the ore to be selected is calculated.

[0007] In some technical solutions of this application, the aforementioned geometric features include the envelope contour, centroid coordinates, and volume of each ore to be selected extracted from a three-dimensional point cloud; The second multiplexing channel is used to associate the image features and spectral features with the same target ore based on the envelope contour and centroid coordinates; The volume is used to correct the correlated spectral features to obtain corrected spectral features; the corrected spectral features are used for feature fusion.

[0008] In some technical solutions of this application, the aforementioned image features include grayscale distribution features extracted from dual-energy X-ray transmission images and color distribution features extracted from color images; the spectral features include elemental intensity features extracted from micro-area X-ray fluorescence spectra and reflectance curve features extracted from short-wave infrared reflectance spectra; the fused features are obtained through the following methods: The element intensity feature and the reflectance curve feature are concatenated into a query vector, and the grayscale distribution feature and the color distribution feature are concatenated into a key vector and a value vector; Scaling dot product attention is performed on the query vector, key vector, and value vector to obtain attention output features; The attention output feature is concatenated with the geometric feature to obtain the fused feature.

[0009] In some technical solutions of this application, the confidence level of the sorting action of the target ore calculated based on the fusion characteristics of the target ore includes: Based on the fusion features and the preset first weight vector and first bias, a first score is calculated; wherein, the first confidence level represents the probability that the target ore is a concentrate; A second score is calculated based on the fusion features and a preset second weight vector and second bias; wherein the first weight vector is different from the second weight vector; and the second confidence level represents the probability that the target ore is tailings. The first confidence level and the second confidence level are used as the confidence levels for the sorting action.

[0010] In some technical solutions of this application, the above system also includes a re-inspection channel; The second selection channel is also used for: Calculate the consistency index among the geometric features, image features, and spectral features associated with the same target ore. If the consistency index is less than a preset conflict threshold, the target ore is sent to the re-inspection channel. Alternatively, if the first confidence level is less than or equal to the first threshold and the second confidence level is less than or equal to the second threshold, then the ore to be selected is sent to the re-inspection channel.

[0011] Secondly, embodiments of this application provide a method for intelligent mineral processing, the method comprising: The ore to be processed is evenly distributed and arranged in a single layer by a conveying device and then transported to the testing area. The preliminary selection device collects the geometric features and color distribution features of the ore to be selected in the detection area, and determines the ore type and corresponding subsequent processing channel based on the geometric features and color distribution features; wherein, the subsequent processing channel includes a first re-selection channel and a second re-selection channel, and the process complexity of the second re-selection channel is higher than that of the first re-selection channel. The geometric features and / or color distribution features of the easily selectable ore are identified by the first multiple selection channel to determine whether the ore is concentrate or tailings. Image and spectral features of the difficult-to-select ore are acquired through a second multiplexing channel. Using the geometric features as a spatial reference, the image and spectral features are associated with the same target ore. The sorting action confidence of the target ore is calculated based on the fusion features of the target ore, and the target ore is sorted based on the sorting action confidence. The fusion features are obtained by fusing the geometric features, image features, and spectral features.

[0012] Thirdly, embodiments of this application provide an electronic device, a processor, a memory, and a bus. The memory stores machine instructions executed by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine instructions are executed by the processor, the steps of the above-described intelligent mineral processing method are performed.

[0013] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when run by a processor, executes the steps of the aforementioned intelligent mineral processing method.

[0014] The technical solutions provided by the embodiments of this application may include the following beneficial effects: This application provides an intelligent mineral processing system, comprising: a conveying device for uniformly distributing and arranging ore to be processed in a single layer and then conveying it to a detection area; a primary selection device for collecting the geometric features and color distribution features of the ore to be processed in the detection area, and determining the ore type and corresponding subsequent processing channel based on the geometric features and color distribution features; wherein the subsequent processing channel includes a first secondary selection channel and a second secondary selection channel, and the process complexity of the second secondary selection channel is higher than that of the first secondary selection channel; the first secondary selection channel is used for easily sorted... The candidate ore of the specified type is subjected to geometric feature and / or color distribution feature identification to determine whether the candidate ore is concentrate or tailings; a second re-selection channel is used to collect image features and spectral features of the candidate ore of the difficult-to-select type, and associate the image features and spectral features with the geometric features as a spatial reference to the same target ore; the sorting action confidence of the target ore is calculated based on the fusion features of the target ore, and the target ore is sorted based on the sorting action confidence; wherein, the fusion features are obtained by fusing the geometric features, image features and spectral features.

[0015] This application utilizes a primary sorting device to classify ores into easily sortable and difficult-to-sort types based on their geometric and color distribution characteristics. These types are then fed into a first or second secondary sorting channel with different process complexities. This allows easily sortable ores to be quickly identified as concentrates or tailings without requiring expensive multi-sensor detection, significantly improving processing efficiency and reducing equipment wear and energy consumption. The second secondary sorting channel uses geometric features as a spatial reference to accurately correlate image and spectral features with the same target ore, overcoming the misalignment problem caused by heterogeneous data formats and asynchronous sampling frequencies from multiple sensors, ensuring the consistency of fused data. Simultaneously, by fusing geometric, image, and spectral features to obtain fused features, and calculating the confidence level of sorting actions based on these features, refined and quantitative decision-making for difficult-to-sort ores is achieved, effectively improving sorting accuracy and resource recovery rate.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of an intelligent mineral processing system provided in an embodiment of this application is shown; Figure 2 A schematic flowchart of an intelligent mineral processing method provided in an embodiment of this application is shown; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0020] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0022] In the field of mineral processing, most beneficiation systems employ a uniform testing process for all ores, such as relying solely on color or a single sensor for judgment. They lack the ability to pre-classify ores based on easily obtainable characteristics like ore geometry, color, and texture, and allocate processing channels of varying complexity. This results in easily sortable ores being unnecessarily sent to costly, multi-sensor, complex testing stages, leading to decreased processing efficiency and resource waste. Furthermore, even when existing technologies collect data from multiple sensors, they often employ simple, fixed rules, making accurate sorting difficult.

[0023] Based on this, embodiments of this application provide a smart mineral processing system, method, electronic device, and storage medium, which are described below through embodiments. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0024] Figure 1 The diagram shows a schematic of an intelligent mineral processing system provided in an embodiment of this application, including a conveying device, a primary processing device, a first secondary processing channel, and a second secondary processing channel.

[0025] The conveying device is used to evenly distribute and arrange the ore to be processed in a single layer before transporting it to the testing area. In practice, the conveying device typically includes a vibrating feeder and a belt conveyor. The vibrating feeder continuously and evenly distributes the ore from the hopper onto the surface of the belt conveyor. The feed rate can be controlled by adjusting the vibration frequency and amplitude to prevent ore accumulation or flow interruption. The belt conveyor consists of a drive roller, idlers, and a conveyor belt. The belt speed can be adjusted according to the processing volume and testing requirements, for example, set to 1.5–2.0 m / s. To ensure that the ore enters the testing area in a single layer without overlapping, an aligning device is installed above the belt. This aligning device can be multiple rows of parallel combs or adjustable-height baffles. The gap between the combs is slightly larger than the maximum particle size of the ore. As the ore moves with the belt, overlapping ore is separated by the combs or baffles, forming a single layer arrangement. The aligned ore is evenly distributed in the width direction of the belt, and adjacent ore pieces maintain a certain distance, thus avoiding testing errors caused by overlap or obstruction. The transmission device provides a stable and reliable ore flow to the subsequent primary and secondary selection channels in the manner described above.

[0026] After the ore to be processed is transported to the testing area, to avoid wasting resources, this embodiment of the application sets up a multi-stage screening method. First, a preliminary screening device quickly distinguishes between easily processed and difficult-to-process ores. Then, more testing resources are allocated to the difficult-to-process ores, while the easily processed ores are processed using only a simple channel. The preliminary screening device is used to collect the geometric and color distribution characteristics of the ore to be processed in the testing area, and based on these characteristics, determine the ore type and corresponding subsequent processing channel. The subsequent processing channels include a first secondary screening channel and a second secondary screening channel, with the second secondary screening channel having a higher process complexity than the first secondary screening channel.

[0027] In practice, geometric features are acquired using a 3D topography sensor. This sensor can be a laser profilometer or a structured light camera. As the ore moves along the conveyor belt, the sensor continuously scans the height profile of each belt section, stitching together 3D point cloud data for each piece of ore on the entire belt. From the point cloud, each ore particle can be segmented, and its envelope profile, centroid coordinates, and volume can be extracted. The envelope profile refers to the convex polygon projected onto the conveyor belt plane, used to define the spatial extent of the ore; the centroid coordinates are the two-dimensional position of the center of this polygon; and the volume is obtained by integrating the point cloud voxels. Color distribution features are acquired using a color industrial camera. The camera captures RGB images of the ore surface under LED illumination, and then calculates statistical quantities such as color histograms, color moments, or texture entropy from the images. These statistical quantities include the mean and variance of the three channels, as well as the entropy value based on the gray-level co-occurrence matrix. These statistical quantities reflect the tonal distribution and texture complexity of the ore surface.

[0028] For example, point cloud data ,in .

[0029] For each individual ore : Calculate the centroid coordinates ; Calculate volume (Voxenon method); Extracting the envelope contour (The convex hull polygon projected onto the xy plane); Output: Set of individual ore particles Each individual contains .

[0030] When determining the ore type and corresponding subsequent processing channel of the candidate ore based on the geometric features and color distribution features, the initial selection device first calculates the complexity index of each candidate ore based on the collected geometric features and color distribution features. This complexity index is a dimensionless value used to quantitatively characterize the complexity of the ore's geometric shape and color texture: the more irregular the geometric shape (e.g., low convexity, large aspect ratio) and the more chaotic the color texture (e.g., large color variance, high texture entropy), the larger the complexity index, indicating that the ore is more likely to belong to a difficult-to-select type; conversely, ores with regular shapes and uniform colors have a smaller complexity index, indicating that they belong to an easy-to-select type.

[0031] When calculating the complexity index of the ore to be selected based on the geometric features and the color distribution features, this application provides a variety of different methods. In specific implementation, different calculation methods can be used for different scenarios.

[0032] The first approach uses a pre-trained lightweight classifier, such as XGBoost or a logistic regression model. First, a large number of known easy-to-select and difficult-to-select ore samples are collected, and their geometric and color distribution features are extracted. Easy-to-select samples are labeled 0, and difficult-to-select samples are labeled 1. Then, these samples are used to train a classifier that can output a probability value between 0 and 1 based on the input feature vector; this probability value is the complexity index. In actual operation, the geometric and color distribution features of the current ore to be selected are input into the trained classifier to directly obtain the complexity index. This method is suitable for mines with relatively fixed ore types and sufficient sample data, and has the advantages of fast computation speed and high accuracy.

[0033] The second approach uses a distance metric. Geometric and color distribution features are pre-extracted from typical easily processed ores, and their mean vectors are calculated as the easily processed feature centers. For the current ore to be processed, its geometric and color distribution features are combined into a feature vector, and the Euclidean distance between this feature vector and the easily processed feature centers is calculated. A larger distance indicates that the ore deviates further from the easily processed center, and the higher the complexity. This approach is suitable for scenarios where easily processed ores have concentrated features, while difficult-to-process ores are more dispersed, and can be implemented without requiring a large number of labeled difficult-to-process samples.

[0034] The third approach is the deviation method based on principal component analysis. First, principal component analysis is performed on a batch of samples representing easily processed ores to extract the first principal component, and the mean score μ and standard deviation σ of the easily processed samples on this principal component are calculated. For the current ore to be processed, its geometric features and color distribution features are projected onto the same principal component direction to obtain a score x. Then, the deviation d = |x - μ| / σ is calculated, and the magnitude of the deviation is mapped to a complexity index. This method is suitable for scenarios with high feature dimensionality and correlation, effectively compressing information and highlighting key differences.

[0035] The fourth method is template comparison. The geometric and color distribution characteristics of the candidate ore are compared with those of a preset reference ore to obtain geometric and color similarity scores. Based on these scores, the complexity index of the candidate ore is calculated. In practice, the preset reference ore can be a pre-selected typical easily processed ore, such as pure quartz vein or a single high-grade chalcopyrite block. The system obtains its geometric and color distribution characteristics from a database. For the current candidate ore, its geometric feature vector is compared with the geometric feature vector of the reference ore, for example, using cosine similarity, to obtain geometric similarity. Similarly, its color distribution feature vector is compared with the color distribution feature vector of the reference ore to obtain color similarity. Both similarity scores are between 0 and 1, with higher values ​​indicating closer similarity to the reference ore. Then, the geometric and color similarity scores are combined and converted into a complexity index. A simple conversion formula is: Complexity Index = 1 (Geometric similarity × Color similarity). Thus, when the candidate ore is highly similar to the reference ore in both geometry and color, the product is close to 1, and the complexity index is close to 0, indicating it is very easy to select. Conversely, if there are significant differences in both aspects, the product is close to 0, and the complexity index is close to 1, indicating it is difficult to select. A weighted average can also be used: Complexity Index = 1 (α·geometric similarity + β·color similarity), where α + β = 1, and α and β can be adjusted according to the relative importance of easy-to-select ores based on geometric and color features. The template comparison method does not require a large number of labeled samples for training; it only needs one or a few representative templates to work. It is especially suitable for production scenarios with stable mineral sources and clear characteristics of easy-to-select ores. It is simple to calculate and convenient to implement.

[0036] In practical applications, any of the above methods can be selected based on factors such as the stability of the ore properties, the availability of historical data, and computing resources, or multiple methods can be weighted and integrated.

[0037] After obtaining the complexity index, the initial selection device compares it with a preset complexity threshold. This threshold is determined based on the ore properties and on-site tests, and can be set to, for example, 0.3. If the complexity index is less than the threshold, the ore is determined to be easy to select and assigned to the first secondary selection channel; if the complexity index is greater than or equal to the threshold, it is determined to be difficult to select and assigned to the second secondary selection channel. Through this classification strategy based on the complexity index, the system can automatically guide ores with simple surface features and easy identification to the low-cost, high-efficiency first secondary selection channel, while guiding ores with complex surfaces that may require multi-sensor joint judgment to the second secondary selection channel with more complete equipment and more refined algorithms. This maximizes the saving of equipment resources and processing time while ensuring sorting accuracy.

[0038] Regarding the first secondary selection channel, in this embodiment, the first secondary selection channel is mainly used to process ore samples determined by the primary selection device to be of an easily selectable type. Since easily selectable ores have regular geometric shapes and uniform color textures, concentrates and tailings can be reliably distinguished solely by the geometric features and / or color distribution features acquired during the primary selection stage, thereby saving equipment resources and energy consumption. In specific implementation, after an ore sample is assigned to the first secondary selection channel by the primary selection device, the system directly utilizes the previously collected geometric and color distribution features for rapid classification. For example, if the color distribution features show that the ore's RGB mean falls in brass (R≈180~220, G≈100~150, B≈30~80) and its texture entropy is low (less than a certain threshold), while the convexity in the geometric features is close to 1 (regular shape), then it is determined to be concentrate, and the air valve is controlled to blow it into the concentrate bin. Conversely, if the average color value is whitish or light gray (high R, G, and B values), the texture entropy is very low, and the volume and convexity are large, it is identified as tailings and allowed to fall into the tailings bin. To prevent missed or false positives, multiple decision rules can be set or a simple threshold classifier can be used. For the very few cases where a clear judgment cannot be made, conservative rules can be set (e.g., defaulting to tailings or sending it to the re-inspection channel), but the identification accuracy for easily sortable types of ores can usually reach over 95%. Through this first re-selection channel, the system can significantly improve processing speed and reduce operating costs while ensuring sorting accuracy.

[0039] Regarding the second selection channel, in this embodiment, it is used to process ores that the initial selection device determines to be difficult to process. These ores are typically irregular in shape, complex in color and texture, or have inconsistent surface features with internal composition. A single sensor cannot accurately determine their composition, thus requiring the acquisition of richer multimodal information. Specifically, the second selection channel first acquires image and spectral features of the difficult-to-process ores. Image features include grayscale distribution features extracted from dual-energy X-ray (XRF) transmission images and color distribution features extracted from color images. Dual-energy X-ray transmission images can reflect the atomic number density distribution within the ore. Grayscale distribution features are numerical representations obtained after convolution or texture analysis of the image, such as local binary patterns or gray-level co-occurrence matrix statistics. The color image is an RGB three-channel color image, and color distribution features can be color histograms, color moments, or cluster centers in the color space. Spectral features include elemental intensity features extracted from micro-area X-ray fluorescence spectra and reflectance curve features extracted from short-wave infrared reflectance spectra. Micro-area X-ray fluorescence spectroscopy performs a dot matrix scan on the ore surface, and each detection point outputs the characteristic X-ray fluorescence intensity of several target elements (such as Cu, Pb, Zn, Au, As, etc.). The element intensity characteristics are the numerical set of these intensities. Short-wave infrared reflectance spectroscopy acquires spectral curves in the 1000nm to 2500nm band. The reflectance curve characteristics can be a one-dimensional spectral waveform or the principal component score after principal component dimensionality reduction.

[0040] Because image features and spectral features have different data formats (image features are dense area array data, while XRF in spectral features is sparse point detection and hyperspectral features are line scans), and the spatial sampling frequencies of each sensor are inconsistent, the second selection channel needs to use geometric features as a spatial reference to associate these features with the same target ore. Geometric features are obtained by the three-dimensional topography sensor in the primary selection device, including the envelope contour, centroid coordinates, and volume of each ore to be selected.

[0041] The association process is as follows: For image features (area array data), the spatial coordinates of each pixel are compared with the envelope contours of all ores. If the pixel coordinates are inside a certain envelope contour, the grayscale distribution features and color distribution features corresponding to that pixel are assigned to the ore corresponding to that contour. For XRF point detection data in spectral features, the Euclidean distance between the spatial coordinates of each detection point and the centroid coordinates of each ore is calculated. The ores corresponding to the centroids with a distance less than a preset threshold (e.g., 2 mm) are selected as the assigned ores, and the element intensity features of that detection point are assigned to that ore. For shortwave infrared scanning data, the spatial coordinates of each point on each scan line are obtained, and the reflectance curve features corresponding to the scan points located inside the envelope contour are assigned to the ores corresponding to that contour. Through the above steps, all image features and spectral features are accurately associated with the same target ore.

[0042] After association, the second selection channel further corrects the associated spectral features using volume from the geometric features. Since XRF fluorescence intensity is positively correlated with ore volume, directly using the original intensity would lead to large, low-grade ores being misclassified as high-grade. The correction model is as follows:

[0043] in, Target ore XRF fluorescence intensity (normalized to average). For the volume of the ore, These are coefficients determined through calibration experiments (using linear regression fitting with standard ore samples of known grades), for example, the calibration coefficient for copper is... , The estimated grade per unit volume of each target ore element is obtained. And use it as the spectral feature obtained after correction.

[0044] Then, the second selection channel fuses geometric features, image features, and corrected spectral features into a fused feature. Specifically, a deep neural network based on cross-modal attention is used: the corrected elemental intensity features and reflectance curve features are concatenated into a query vector, and the grayscale distribution features and color distribution features are concatenated into a key vector and a value vector; scaling dot product attention is performed on the query vector, key vector, and value vector, i.e., the dot product of the query and key is calculated, divided by a scaling factor, weighted using Softmax, and then the values ​​are weighted and summed to obtain the attention output feature; finally, this attention output feature is concatenated with the geometric features to form the final fused feature. This attention mechanism allows spectral features to actively guide the network to focus on spatially related regions in the image features, thereby enhancing the structural information of mineralized areas.

[0045] After obtaining the fusion feature, the second selection channel calculates the confidence level of the target ore sorting action based on the fusion feature. Specifically, the system pre-trains using logistic regression to obtain a first weight vector, a first bias (corresponding to the concentrate category), a second weight vector, and a second bias (corresponding to the tailings category). The dimensions of these parameters are the same as the fusion feature. The dot product of the fusion feature and the first weight vector, plus the first bias, is calculated to obtain the first score. This score is then mapped to a first confidence level between 0 and 1 using the Sigmoid function, representing the probability that the ore is concentrate. Similarly, the second confidence level is calculated, representing the probability that the ore is tailings. If the first confidence level is greater than a preset first threshold (e.g., 0.7), it is determined to be concentrate, and the control system drives the air valve to blow the ore into the concentrate bin. If the second confidence level is greater than a second threshold (e.g., 0.7), it is determined to be tailings, and the ore falls into the tailings bin. If neither confidence level is greater than its respective threshold, it is determined to be awaiting re-inspection, and the ore is sent to the re-inspection channel for further processing (e.g., crushing and re-inspection). Through the above-mentioned confidence calculation of sorting action based on fusion features, the second re-sorting channel can realize refined and quantitative decision-making for difficult-to-sort ores, significantly improving sorting accuracy and resource recovery rate.

[0046] In the second selection channel, this application also introduces a conflict detection mechanism to identify ores that are inconsistent in appearance and internal structure. Specifically, after associating geometric features, image features, and spectral features with the same target ore, the system calculates a consistency index among these modal features. This consistency index is used to quantitatively assess the degree of agreement between different sensors' judgments of the same ore. For example, a comprehensive vote can be made based on the density category (high density / low density) predicted by dual-energy X-ray transmission images, the grade category (high grade / low grade) predicted by X-ray fluorescence elemental intensity, and the surface category (rich ore / poor ore) predicted by color and geometric features. If the predictions of the three are consistent (e.g., all pointing to concentrate or all pointing to tailings), the consistency index is high (close to 1); if there is a discrepancy (e.g., the surface shows poor ore while the interior shows a high-density ore body), the consistency index is low (close to 0).

[0047] One specific calculation method involves having M modes (e.g., geometric features, XRT image features, color image features, etc.) for the same target ore. Each mode outputs a sorting confidence score for determining whether the ore is a concentrate. The confidence level of the mode with the highest stability (e.g., XRF) is taken as the reference confidence level. Then the formula for calculating the consistency index λj is: The value of λj ranges from [0, +∞). A smaller value indicates greater consistency in modal judgments; a larger value indicates more severe discrepancies. The system presets a conflict threshold (e.g., 0.5). When λj is greater than or equal to this threshold, it indicates a serious discrepancy between the surface and internal structure of the ore (e.g., the surface color shows low-grade ore while the internal X-ray transmission image shows a high-density ore body). In this case, subsequent feature fusion and confidence calculation are not performed, and the target ore is directly sent to the re-inspection channel. The re-inspection channel is usually connected to a crusher, and the ore is crushed and then re-enters the system for inspection. When λj is less than the conflict threshold, the modal judgments are basically consistent, and the system continues to perform volume correction, feature fusion, and confidence calculation for sorting actions. Through the above conflict detection mechanism, this application effectively avoids erroneous sorting caused by contradictions in multimodal data, improving the system's robustness to complex ores.

[0048] Figure 2 The diagram illustrates a flow chart of an intelligent mineral processing method provided in an embodiment of this application, wherein the method includes steps S101-S104; specifically: S101. The ore to be processed is evenly distributed and arranged in a single layer by a conveying device and then transported to the detection area. S102. The preliminary selection device collects the geometric features and color distribution features of the ore to be selected in the detection area, and determines the ore type and corresponding subsequent processing channel of the ore to be selected based on the geometric features and color distribution features; wherein, the subsequent processing channel includes a first re-selection channel and a second re-selection channel, and the process complexity of the second re-selection channel is higher than that of the first re-selection channel. S103. The geometric features and / or color distribution features of the easily selectable ore are identified by the first multiple selection channel to determine whether the ore is concentrate or tailings. S104. Image features and spectral features of the difficult-to-select type of ore are collected through the second multiple selection channel. Using the geometric features as a spatial reference, the image features and spectral features are associated with the same target ore. The sorting action confidence of the target ore is calculated based on the fusion features of the target ore, and the target ore is sorted based on the sorting action confidence. The fusion features are obtained by fusing the geometric features, image features and spectral features.

[0049] The process of determining the ore type and corresponding subsequent processing channel of the candidate ore based on the geometric features and color distribution features includes: The complexity index of the ore to be selected is calculated based on the geometric features and the color distribution features; wherein the complexity index characterizes the complexity of the geometric shape and color texture of the ore to be selected. The complexity index is compared with the complexity threshold to determine whether the ore to be selected is of an easy type and enters the first secondary selection channel, or to determine whether the ore to be selected is of a difficult type and enters the second secondary selection channel.

[0050] The step of calculating the complexity index of the ore to be selected based on the geometric features and the color distribution features includes: The geometric and color distribution characteristics of the ore to be selected are compared with the geometric and color distribution characteristics of a preset reference ore to obtain geometric similarity and color similarity. Based on the obtained geometric similarity and color similarity, the complexity index of the ore to be selected is calculated.

[0051] The geometric features include the envelope contour, centroid coordinates, and volume of each ore sample extracted from the three-dimensional point cloud; Based on the envelope contour and centroid coordinates, the image features and spectral features are associated with the same target ore; The volume is used to correct the correlated spectral features to obtain corrected spectral features; the corrected spectral features are used for feature fusion.

[0052] The image features include grayscale distribution features extracted from dual-energy X-ray transmission images and color distribution features extracted from color images; the spectral features include elemental intensity features extracted from micro-area X-ray fluorescence spectra and reflectance curve features extracted from short-wave infrared reflectance spectra; the fused features are obtained through the following method: The element intensity feature and the reflectance curve feature are concatenated into a query vector, and the grayscale distribution feature and the color distribution feature are concatenated into a key vector and a value vector; Scaling dot product attention is performed on the query vector, key vector, and value vector to obtain attention output features; The attention output feature is concatenated with the geometric feature to obtain the fused feature.

[0053] The step of calculating the sorting confidence level of the target ore based on its fusion characteristics includes: Based on the fusion features and the preset first weight vector and first bias, a first score is calculated; wherein, the first confidence level represents the probability that the target ore is a concentrate; A second score is calculated based on the fusion features and a preset second weight vector and second bias; wherein the first weight vector is different from the second weight vector; and the second confidence level represents the probability that the target ore is tailings. The first confidence level and the second confidence level are used as the confidence levels for the sorting action.

[0054] The method also includes Calculate the consistency index among the geometric features, image features, and spectral features associated with the same target ore. If the consistency index is less than a preset conflict threshold, the target ore is sent to the re-inspection channel. Alternatively, if the first confidence level is less than or equal to the first threshold and the second confidence level is less than or equal to the second threshold, then the ore to be selected is sent to the re-inspection channel.

[0055] like Figure 3 As shown, this application provides an electronic device for executing the intelligent mineral processing method described in this application. The device includes a memory, a processor, a bus, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the intelligent mineral processing method described above.

[0056] Specifically, the aforementioned memory and processor can be general-purpose memory and processor, without any specific limitations. When the processor runs the computer program stored in the memory, it can execute the aforementioned intelligent mineral processing method.

[0057] Corresponding to the intelligent mineral processing method in this application, this application embodiment also provides a computer storage medium storing a computer program, which is executed by a processor to perform the steps of the above-described intelligent mineral processing method.

[0058] Specifically, the storage medium can be a general-purpose storage medium, such as a portable disk or hard disk. When the computer program on the storage medium is run, it can execute the aforementioned intelligent mineral processing method.

[0059] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0060] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; 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, depending on actual needs.

[0061] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0062] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application. 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.

[0063] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0064] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A smart mineral processing system, characterized in that, include: A conveying device is used to evenly distribute and arrange the ore to be processed in a single layer before conveying it to the testing area. A preliminary selection device is used to calculate a complexity index of the ore to be selected in the detection area based on the collected geometric features and color distribution features; compare the complexity index with a complexity threshold to determine whether the ore to be selected is of an easy-to-select type and enters a first secondary selection channel, or to determine whether the ore to be selected is of a difficult-to-select type and enters a second secondary selection channel; wherein, the complexity index characterizes the complexity of the geometric shape and color texture of the ore to be selected; the geometric shape and color texture are both surface features; The first selection channel is used to identify the geometric features and / or color distribution features of the easily selectable type of ore to determine whether the ore is concentrate or tailings; the geometric features include the envelope contour, centroid coordinates and volume of each ore extracted from the three-dimensional point cloud; The second selection channel is used to collect image features and spectral features of the difficult-to-select type of ore to be selected; the image features include grayscale distribution features extracted from dual-energy X-ray transmission images and color distribution features extracted from color images; the spectral features include elemental intensity features extracted from micro-area X-ray fluorescence spectra and reflectance curve features extracted from short-wave infrared reflectance spectra. Based on the envelope contour and centroid coordinates, the image features and spectral features are associated with the same target ore, and the volume is used to correct the associated spectral features to obtain the corrected spectral features; The corrected element intensity feature and reflectivity curve feature are concatenated into a query vector, and the grayscale distribution feature and color distribution feature are concatenated into a key vector and a value vector; scaling dot product attention is performed on the query vector, key vector, and value vector to obtain the attention output feature; the attention output feature is concatenated with the geometric feature to obtain the fused feature; The confidence level of the sorting action of the target ore is calculated based on the fusion characteristics of the target ore, and the target ore is sorted based on the confidence level of the sorting action.

2. The system according to claim 1, characterized in that, The step of calculating the complexity index of the ore to be selected based on the geometric features and the color distribution features includes: The geometric and color distribution characteristics of the ore to be selected are compared with the geometric and color distribution characteristics of a preset reference ore to obtain geometric similarity and color similarity. Based on the obtained geometric similarity and color similarity, the complexity index of the ore to be selected is calculated.

3. The system according to claim 1, characterized in that, The step of calculating the sorting confidence level of the target ore based on its fusion characteristics includes: Based on the fusion features and the preset first weight vector and first bias, a first score is calculated; wherein, the first confidence level represents the probability that the target ore is a concentrate; A second score is calculated based on the fusion features and a preset second weight vector and second bias; wherein the first weight vector is different from the second weight vector; the second confidence level represents the probability that the target ore is tailings; The first confidence level and the second confidence level are used as the confidence levels for the sorting action.

4. The system according to claim 3, characterized in that, It also includes the re-inspection channel; The second selection channel is also used for: Calculate the consistency index among the geometric features, image features, and spectral features associated with the same target ore. If the consistency index is less than a preset conflict threshold, the target ore is sent to the re-inspection channel. Alternatively, if the first confidence level is less than or equal to the first threshold and the second confidence level is less than or equal to the second threshold, then the ore to be selected is sent to the re-inspection channel.

5. A method for intelligent mineral processing, characterized in that, The method includes: The ore to be processed is evenly distributed and arranged in a single layer by a conveying device and then transported to the testing area. In the detection area, the preliminary selection device calculates the complexity index of the ore to be selected based on the collected geometric and color distribution characteristics. The complexity index is then compared with a complexity threshold to determine whether the ore is easy to select and enters the first secondary selection channel, or whether it is difficult to select and enters the second secondary selection channel. The complexity index characterizes the complexity of the geometric shape and color texture of the ore to be selected; both the geometric shape and color texture are surface features. The geometric features and / or color distribution features of the easily selectable ore are identified through the first selection channel to determine whether the ore is concentrate or tailings; the geometric features include the envelope contour, centroid coordinates and volume of each ore piece extracted from the three-dimensional point cloud; Image and spectral features of the difficult-to-select ore are acquired through the second multiple selection channel; the image features include grayscale distribution features extracted from dual-energy X-ray transmission images and color distribution features extracted from color images; the spectral features include elemental intensity features extracted from micro-area X-ray fluorescence spectra and reflectance curve features extracted from short-wave infrared reflectance spectra. Based on the envelope contour and centroid coordinates, the image features and spectral features are associated with the same target ore, and the volume is used to correct the associated spectral features to obtain the corrected spectral features; The corrected element intensity feature and reflectivity curve feature are concatenated into a query vector, and the grayscale distribution feature and color distribution feature are concatenated into a key vector and a value vector; scaling dot product attention is performed on the query vector, key vector, and value vector to obtain the attention output feature; the attention output feature is concatenated with the geometric feature to obtain the fused feature; The confidence level of the sorting action of the target ore is calculated based on the fusion characteristics of the target ore, and the target ore is sorted based on the confidence level of the sorting action.

6. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine instructions that the processor executes. When the electronic device is running, the processor communicates with the memory via the bus. When the machine instructions are executed by the processor, the steps of the intelligent mineral processing method as described in claim 5 are performed.

7. A computer storage medium, characterized in that, The computer storage medium stores a computer program that, when executed by a processor, performs the steps of the intelligent mineral processing method as described in claim 5.

Images