Intelligent screening and conveying integrated system for rare earth ore beneficiation process

Abstract

The application provides an intelligent screening and conveying integrated system for a rare earth ore dressing process, which collects image features, near-infrared spectral features and spectral data of a rare earth ore stream; determines a mineral region and a particle size distribution according to the image features, the near-infrared spectral features and the spectral data, determines a screening parameter combination and adjusts a screening device to preliminarily screen mineral particles; scans and excites the screened rare earth ore stream; performs splicing and fusion based on atomic emission spectroscopy and near-infrared spectral features to obtain a bimodal feature vector of the rare earth ore stream, performs attention optimization identification on the modal feature vector, and then obtains a sorting decision signal; and adjusts parameters of a pneumatic sorting machine according to the sorting decision signal, so as to re-sort the rare earth ore stream. The scheme of the application can accurately identify the characteristics of rare earth minerals and perform efficient intelligent screening and conveying of rare earth ore dressing.

Description

Technical Field

[0001] This application relates to the field of screening technology, and more specifically, to an integrated intelligent screening and conveying system for rare earth ore beneficiation processes. Background Technology

[0002] Rare earth minerals are key strategic resources supporting the development of modern high-tech and green energy industries. Mineral processing is the core step in efficiently enriching target minerals from complex raw ores. Traditional mineral processing relies heavily on pre-set fixed parameters and human experience, resulting in low efficiency and significant resource waste.

[0003] Existing technologies attempt to incorporate single-spectral analysis (such as X-ray fluorescence or near-infrared spectroscopy) or machine vision into the sorting process to predict ore grade. However, these methods typically rely on a single information dimension, making it difficult to comprehensively and accurately identify rare earth minerals that look similar but have vastly different compositions. Furthermore, they cannot differentiate mineral surface characteristics (such as weathering and contamination levels) in real time on high-speed production lines, resulting in insufficient basis for sorting decisions. Simultaneously, traditional screening equipment has fixed parameters and cannot adaptively adjust to real-time changes in the ore flow, leading to a crude sorting process, loss of high-value ore, and low energy consumption. Therefore, accurately identifying rare earth mineral characteristics and implementing efficient intelligent screening and conveying for rare earth ore beneficiation has become a challenge for the industry. Summary of the Invention

[0004] This application provides an integrated intelligent screening and conveying system for rare earth ore beneficiation, which can accurately identify the characteristics of rare earth minerals and perform intelligent screening and conveying for efficient rare earth ore beneficiation.

[0005] This application provides an integrated intelligent screening and conveying system for rare earth ore beneficiation, the integrated intelligent screening and conveying system comprising:

[0006] The data acquisition module performs spectral detection and image acquisition on the rare earth ore flow on the conveyor belt, and obtains the image features, near-infrared spectral features and X-ray fluorescence spectral data of the rare earth ore flow;

[0007] The preliminary screening module determines the mineral regions and particle size distribution of rare earth minerals in the rare earth ore stream based on the image features, the near-infrared spectral features, and the spectral data, and then determines the combination of screening parameters and adjusts the screening equipment to perform preliminary screening of mineral particles with different surface properties.

[0008] The scanning excitation module uses a coaxially integrated laser-induced breakdown spectroscopy probe to scan and excite the screened rare earth ore flow and acquire atomic emission spectra.

[0009] The signal generation module splices and fuses the atomic emission spectrum and the near-infrared spectral features to obtain a dual-modal feature vector of the ore components in the rare earth ore stream. The module then performs attention optimization identification on the modal feature vector focusing on rare earth minerals to obtain a sorting decision signal.

[0010] The re-sorting module adjusts the parameters of the pneumatic separator according to the sorting decision signal, thereby re-sorting the rare earth ore stream.

[0011] In some embodiments, determining the mineral regions and grain size distribution of rare earth minerals in the rare earth ore stream based on the image features, the near-infrared spectral features, and the spectral data specifically includes:

[0012] The near-infrared spectral features were spectrally matched to obtain the potential regions of rare earth minerals;

[0013] Using the potential region as a spatial guide, the image features are subjected to deformable convolutional semantic segmentation to obtain a segmentation mask;

[0014] The rare earth mineral region is segmented based on all the segmentation masks, and the edges of all segmented regions are enhanced to obtain multiple segmented ore regions.

[0015] Determine the grain size distribution histogram based on all the segmented ore regions;

[0016] By cross-correlating the particle size distribution histogram with the potential region, the mineral regions and particle size distribution of rare earth minerals in the rare earth ore flow can be obtained.

[0017] In some embodiments, determining the combination of screening parameters and adjusting the screening equipment to perform preliminary screening of mineral particles with different surface properties specifically includes:

[0018] Multiple surface properties of minerals in the rare earth ore flow were extracted based on the near-infrared spectral characteristics.

[0019] The parameter combinations for multiple screenings are determined based on all surface properties and the particle size distribution.

[0020] All parameter combinations are sent to the screening equipment, and the servo mechanism of the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface characteristics.

[0021] In some embodiments, scanning and exciting the screened rare earth ore flow using a coaxially integrated laser-induced breakdown spectroscopy probe and obtaining atomic emission spectra specifically includes:

[0022] The preliminary screened rare earth ore stream is then introduced into the secondary conveying unit;

[0023] Above the secondary conveying unit, a coaxially integrated laser-induced breakdown spectroscopy probe is used to scan and excite the pre-screened rare earth ore stream.

[0024] Obtain the atomic emission spectra obtained from scanning excitation.

[0025] In some embodiments, the method of splicing and fusing the atomic emission spectrum and the near-infrared spectral features to obtain the dual-modal feature vector of the ore composition in the rare earth ore stream specifically includes:

[0026] Extract the characteristic spectral lines of rare earth elements from the atomic emission spectrum;

[0027] Determine the rare earth mineral-related absorption band characteristics in the near-infrared spectral features;

[0028] The characteristic spectral lines and absorption band features are spliced ​​together in their entirety to obtain the specific mineral information of rare earth minerals.

[0029] The specific mineral information is subjected to feature optimization and fusion to obtain a dual-modal feature vector of ore composition in the rare earth ore flow.

[0030] In some embodiments, focusing attention optimization identification on rare earth minerals on the modal feature vectors to obtain sorting decision signals specifically includes:

[0031] Multiple bias weights are assigned to the rare earth element spectral lines and characteristic functional groups in the dual-modal eigenvectors;

[0032] Attention optimization is performed on the bimodal feature vector based on all the bias weights to obtain multiple feature dimensions focusing on rare earth minerals;

[0033] By performing multi-task analysis and grade estimation on all feature dimensions, multiple grade estimates and mineral categories of rare earth minerals in the screened rare earth ore stream are obtained.

[0034] Sorting decision signals are generated based on the mineral category and all grade estimates.

[0035] In some embodiments, adjusting the parameters of the pneumatic separator according to the sorting decision signal to further sort the rare earth ore stream specifically includes:

[0036] The sorting decision signal is analyzed to obtain the spatial coordinates, action commands, and predicted grade values ​​of the target particles;

[0037] Based on the spatial coordinates and the conveyor belt speed, calculate the precise delay for triggering the corresponding sorting nozzles;

[0038] Based on the action command and the predicted grade value, the blowing air pressure and pulse width are dynamically set;

[0039] When the precise delay arrives, the corresponding nozzles are controlled to perform blowing with the set parameters to separate the particles into the corresponding concentrate, middlings or tailings collection tanks.

[0040] In some embodiments, image features of rare earth ore flows are acquired using an industrial camera.

[0041] In some embodiments, X-ray fluorescence spectrometry is used to acquire spectral data of rare earth ore flow.

[0042] In some embodiments, the near-infrared spectral characteristics of rare earth ore flows are acquired using a near-infrared spectrometer.

[0043] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0044] This application provides an integrated intelligent screening and conveying system for rare earth ore beneficiation. First, the rare earth ore stream on the conveyor belt undergoes spectral detection and image acquisition to obtain image features, near-infrared spectral features, and X-ray fluorescence spectral data. Based on the image features, near-infrared spectral features, and spectral data, the mineral regions and particle size distribution of rare earth minerals in the rare earth ore stream are determined. Then, the screening parameter combination is determined, and the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface properties. The screened rare earth ore stream is scanned and excited using a coaxially integrated laser-induced breakdown spectroscopy probe, and atomic emission spectra are acquired. Based on the atomic emission spectra and near-infrared spectral features, a dual-modal feature vector of the ore components in the rare earth ore stream is obtained. Attention optimization identification focusing on rare earth minerals is performed on the modal feature vector to obtain a sorting decision signal. Based on the sorting decision signal, the parameters of the pneumatic separator are adjusted to further sort the rare earth ore stream.

[0045] Therefore, in the intelligent screening and conveying method for rare earth ore beneficiation, this application first performs spectral detection and image acquisition on the rare earth ore flow on the conveyor belt to obtain image features, near-infrared spectral features, and X-ray fluorescence spectral data of the rare earth ore flow. Based on the image features, near-infrared spectral features, and spectral data, the mineral regions and particle size distribution of rare earth minerals in the rare earth ore flow are determined. Then, the combination of screening parameters is determined, and the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface properties. The particle size distribution histogram is used to statistically analyze the proportion of ore particles within different equivalent diameter ranges, providing a visual representation of the overall particle size composition. The mineral regions and particle size distribution are used to comprehensively guide the preliminary physical screening strategy, integrating chemical... The joint analysis results of spatial correlation between component enrichment probability information and physical size statistics directly indicate the spatial range requiring focused sorting and enrichment, avoiding indiscriminate processing of the entire material stream and improving the targeting of the operation. Secondly, a coaxially integrated laser-induced breakdown spectroscopy probe is used to scan and excite the screened rare earth ore stream, acquiring atomic emission spectra. Based on the splicing and fusion of the atomic emission spectra and near-infrared spectral features, a dual-modal feature vector of the ore components in the rare earth ore stream is obtained. Attention optimization identification focusing on rare earth minerals is performed on the modal feature vectors to obtain a sorting decision signal. The parameters of the pneumatic separator are adjusted according to the sorting decision signal to further sort the rare earth ore stream. This scheme can accurately identify the characteristics of rare earth minerals and perform intelligent screening and conveying for efficient rare earth ore beneficiation. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the structure of an integrated intelligent screening and conveying system for rare earth ore beneficiation process according to some embodiments of this application, wherein 100 is a data acquisition module, 200 is a preliminary screening module, 300 is a scanning excitation module, 400 is a signal generation module, and 500 is a secondary sorting module.

[0048] Figure 2 This is an exemplary flowchart of obtaining atomic emission spectra according to some embodiments of this application;

[0049] Figure 3 This is an exemplary flowchart illustrating the determination of bimodal feature vectors according to some embodiments of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0052] refer to Figure 1 The figure is a schematic diagram of the structure of an integrated intelligent screening and conveying system for rare earth ore beneficiation, according to some embodiments of this application. This integrated intelligent screening and conveying system for rare earth ore beneficiation mainly includes: a data acquisition module 100, a preliminary screening module 200, a scanning excitation module 300, a signal generation module 400, and a secondary sorting module 500, which are described below:

[0053] The data acquisition module 100 is used to perform spectral detection and image acquisition on the rare earth ore flow on the conveyor belt, and obtain the image features, near-infrared spectral features and X-ray fluorescence spectral data of the rare earth ore flow.

[0054] In specific implementation, the rare earth ore flow on the conveyor belt is subjected to spectral detection and image acquisition to obtain image features, near-infrared spectral features, and X-ray fluorescence spectral data. This can be achieved in the following way: First, an integrated detection unit consisting of an industrial camera, a near-infrared spectrometer, and a micro-area X-ray fluorescence spectrometer is fixedly installed above the conveyor belt. When the ore flow passes through, it is triggered by a unified timing controller, and the industrial camera captures images of the rare earth ore flow to record its image features such as color, texture, and macroscopic morphology. The near-infrared spectrometer simultaneously acquires the surface reflectance spectrum of the ore to obtain near-infrared spectral features. The micro-area X-ray fluorescence spectrometer excites the ore through an X-ray tube and detects characteristic X-rays to obtain X-ray fluorescence spectral data characterizing the types and contents of elements in the rare earth ore flow. Other embodiments may also use other methods, which are not limited here.

[0055] It should be noted that the image features in this application are a set of digital image information used to characterize the macroscopic physical morphology and surface texture of ore particles, the near-infrared spectral features are continuous spectral curve data used to reflect the vibrational state of mineral molecular bonds (such as hydroxyl, water molecules, carbonate ions) in the ore, and the spectral data are characteristic X-ray energy and intensity distribution data used for qualitative and semi-quantitative analysis of the types and contents of elements contained in the ore.

[0056] The preliminary screening module 200 is used to determine the mineral regions and particle size distribution of rare earth minerals in the rare earth ore stream based on the image features, the near-infrared spectral features and the spectral data, and then determine the screening parameter combination and adjust the screening equipment to perform preliminary screening of mineral particles with different surface properties.

[0057] In some embodiments, determining the mineral regions and grain size distribution of rare earth minerals in the rare earth ore stream based on the image features, the near-infrared spectral features, and the spectral data can be achieved using the following steps:

[0058] The near-infrared spectral features were spectrally matched to obtain the potential regions of rare earth minerals;

[0059] Using the potential region as a spatial guide, the image features are subjected to deformable convolutional semantic segmentation to obtain a segmentation mask;

[0060] The rare earth mineral region is segmented based on all the segmentation masks, and the edges of all segmented regions are enhanced to obtain multiple segmented ore regions.

[0061] Determine the grain size distribution histogram based on all the segmented ore regions;

[0062] By cross-correlating the particle size distribution histogram with the potential region, the mineral regions and particle size distribution of rare earth minerals in the rare earth ore flow can be obtained.

[0063] In specific implementation, the near-infrared spectral features are spectrally matched to obtain potential regions of rare earth minerals. This can be achieved in the following way: A feature database containing standard near-infrared spectra of various rare earth minerals (such as bastnaesite) and common gangue minerals (such as quartz and feldspar) is acquired. When the rare earth ore flow on the conveyor belt passes the detection point, the near-infrared spectrometer collects the spectral curve of each region. A spectral angle filling algorithm is used to measure the similarity between each acquired spectrum and each standard spectrum in the database in the characteristic band by calculating the cosine of the angle between two spectral vectors in multidimensional space. The core of this algorithm is to measure the overall consistency of the spectral curve shape and is not sensitive to changes in light intensity. By setting a similarity threshold (e.g., a threshold of 0.95), all pixels with a similarity to the standard spectrum of rare earth minerals higher than the threshold are selected, and the set of all the above pixels in two-dimensional space is labeled to generate a binary heat map, i.e., the potential regions of rare earth mineral enrichment. Other embodiments can also be implemented in other ways, which are not limited here.

[0064] In specific implementation, the potential region is used as a spatial guide to perform deformable convolution semantic segmentation on the image features. The resulting segmentation mask can be achieved as follows: the potential region is used as a spatial guide, and pixel-level spatial alignment and overlay are performed with images synchronously acquired by an industrial camera. Subsequently, the image with the overlaid spatial guide information is input into a pre-trained semantic segmentation neural network. The core encoder of this network uses deformable convolutional layers. The deformable convolution learns the positional offset of each sampling point through a parallel sub-network, enabling the convolutional kernel to adaptively adjust the shape and size of its receptive field according to the image content (such as irregular, blurred-edge ore shapes), thereby improving... The network accurately matches the true boundaries of the target. During the training phase, it learns from a large number of labeled rare earth ore images. Its loss function encourages the network to pay attention to and segment the highlighted areas of potential regions, thereby outputting a classification label (e.g., "background" or "ore particle") for each pixel of the input image. All pixels predicted as ore particles are grouped according to their spatial location to form a segmentation mask image of the same size as the original image. Multiple boundaries obtained from the grouping of the segmentation mask image are used as segmentation masks. Each independent connected region of the segmentation mask image represents a segmented ore particle. Other embodiments may also use other methods, which are not limited here.

[0065] In specific implementation, the rare earth mineral region is segmented according to all segmentation masks, and the edges of all segmented regions are enhanced to obtain multiple segmented ore regions. This can be achieved in the following way: using a connected component analysis algorithm, each segmentation mask in the segmentation mask image is used to identify an independent closed region representing the ore particle, and a unique identifier is assigned to each region; then, for each identified region, its minimum bounding rectangle or fitted ellipse is calculated, and the initial segmented ore region is defined based on this; then, a pixel-width buffer is defined on both the outer and inner perimeters of the outlines of all initial segmented ore regions. Within this buffer, a hybrid algorithm combining anisotropic diffusion and local contrast enhancement is used to smooth the texture noise inside the region through anisotropic diffusion, while preserving and sharpening the intensity gradient at the boundary of the region; subsequently, for the pixels on both sides of the boundary, the statistical difference in grayscale or texture within their neighborhood is calculated, and the display intensity or contrast of the boundary pixels is adjusted accordingly, thereby obtaining a series of segmented ore regions with clear and sharp boundaries and a higher degree of fit with the actual physical contour of the ore; other embodiments may also use other methods, which are not limited here.

[0066] In specific implementation, the particle size distribution histogram can be determined based on all segmented ore regions in the following way: For each segmented ore region, calculate its pixel area and convert the pixel area into the actual physical projection area according to the camera calibration parameters (i.e., the actual physical size corresponding to each pixel in the image); secondly, assume that the ore particles are approximately spherical or ellipsoidal, and calculate their equivalent diameter based on the projection area; traverse all segmented ore particles, calculate and record the equivalent diameter of each particle, and then, according to a pre-defined set of continuous particle size intervals, such as 0-5mm, 5-10mm, 10-20mm, etc., count the number of ore particles falling into each particle size interval, and plot a bar chart with the particle size interval as the horizontal axis and the number or percentage of particles in the corresponding interval as the vertical axis, thus obtaining the particle size distribution histogram; other embodiments may also use other methods, which are not limited here.

[0067] In specific implementation, the particle size distribution histogram and the potential region are cross-correlated to obtain the mineral regions and particle size distribution of rare earth minerals in the rare earth ore flow. This can be achieved in the following way: First, for any segmented ore region, using the coordinates of all pixels in the segmented ore region as an index, backtrack to the binary heatmap of the potential region and read the spectral matching confidence (i.e., the cosine value of the spectral angle) corresponding to all the above coordinate positions; then, take the average of the confidence scores of all pixels in the segmented ore region as the regional mineral enrichment score of the segmented ore region, to quantify the overall probability that a single ore particle in the segmented ore region belongs to a rare earth mineral; secondly... All segmented ore regions are categorized into different grain size intervals of a preset grain size distribution histogram based on their equivalent diameter. Within each grain size interval, a weighted statistical analysis is performed based on a threshold set according to the regional mineral enrichment score. This yields the overall rare earth enrichment degree of the washed rare earth ore within that grain size interval, ultimately generating a two-dimensional grain size-enrichment correlation distribution map: the horizontal axis represents the grain size interval, and the vertical axis can represent the proportion of high-enrichment particles or the average enrichment score. Simultaneously, the original potential area heat map and the weighted grain size distribution histogram together constitute a complete description of the rare earth mineral distribution in the current ore flow, i.e., mineral regions and grain size distribution. Other embodiments may also employ other methods, which are not limited here.

[0068] It should be noted that the potential region in this application is a spatial probability map calculated based on the matching of near-infrared spectra and standard spectral libraries, used to initially indicate the possible enrichment locations of rare earth minerals; the segmentation mask is a binarized image generated by a deformable convolutional semantic segmentation algorithm, used to accurately separate the contours of individual ore particles from the ore image; the segmented ore region is an image region characterized by the shape and area of ​​a single complete ore particle, after edge feature enhancement processing; the particle size distribution histogram is a chart used to statistically analyze the proportion of ore particles in different equivalent diameter ranges, visually displaying the overall particle size composition; and the mineral region and particle size distribution are joint analysis results after spatially correlating chemical composition enrichment probability information with physical size statistics, used to comprehensively guide the preliminary physical screening strategy.

[0069] In some embodiments, determining the combination of screening parameters and adjusting the screening equipment to perform preliminary screening of mineral particles with different surface properties can be achieved by the following steps:

[0070] Multiple surface properties of minerals in the rare earth ore flow were extracted based on the near-infrared spectral characteristics.

[0071] The parameter combinations for multiple screenings are determined based on all surface properties and the particle size distribution.

[0072] All parameter combinations are sent to the screening equipment, and the servo mechanism of the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface characteristics.

[0073] In specific implementation, extracting multiple surface characteristics of minerals in the rare earth ore flow based on the near-infrared spectral features can be achieved in the following way: First, calculate the depth and area of ​​the absorption band of the near-infrared spectral features near 1400 nm, quantify it as a hydroxyl index, and use it to characterize the degree of hydration of the mineral surface and the content of clay minerals; second, analyze the reflection characteristics of the near-infrared spectral features near 900 nm related to iron oxides, calculate the maximum value of the first derivative of its reflectance, and define it as the iron oxide index to reflect the degree of surface oxidation; finally, identify the characteristic absorption combination bands of near-infrared spectral features in the range of 2300 to 2500 nm belonging to carbonate minerals, and generate a carbonate index by fitting the symmetry and width of its absorption valleys, which is used to indicate the abundance of carbonate-containing rare earth minerals such as bastnaesite; and then combine the above three indices... The number of indexes represents multiple surface characteristics of minerals in a rare earth ore stream. Determining multiple parameter combinations for screening based on all surface characteristics and the particle size distribution can be achieved by inputting the surface characteristic indexes and particle size distribution data into a matrix. The core of this matrix is ​​a set of preset mapping rules based on process knowledge. For example, when a certain ore particle has a high hydroxyl index and a fine particle size, the rules indicate that its viscosity may be high, easily leading to screen clogging. Therefore, a high-frequency, low-amplitude vibrating screen parameter combination is matched for it, supplemented by a higher airflow to assist in screen cleaning. Conversely, for particles with a high carbonate index and coarse particle size, they are judged to be valuable rare earth mineral coarse particles, and a parameter combination with a small screen inclination angle and slow vibration speed is matched for them to prevent them from passing through the screen too quickly and not being adequately inspected. The system iterates through all identified different characteristic-particle size combinations, generating a set of parameter combination instructions for each combination, including specific values ​​such as vibration frequency, amplitude, tilt angle, and airflow velocity. All parameter combinations are sent to the screening equipment, and the servo mechanism of the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface characteristics. This can be achieved as follows: the central controller sends the generated parameter combination instruction set to the servo motor controller of the vibrating screen and the frequency converter of the airflow fan via a real-time industrial network. For the vibrating screen, the servo motor precisely adjusts the speed and phase of the drive eccentric block according to the instructions. For the airflow separation section, the frequency converter adjusts the fan speed to control the airflow velocity in the duct. During the screening process, online weighing sensors and image sensors installed at the screen box outlet and the air-separated product outlet provide real-time feedback on the flow rate and particle distribution changes of different products. The control system compares this feedback with the expected target and fine-tunes the servo mechanism to achieve preliminary screening of mineral particles with different surface characteristics. Other embodiments may also use other methods, which are not limited here.

[0074] It should be noted that the surface characteristics in this application refer to a set of quantifiable spectral characteristic parameters obtained through near-infrared spectroscopy non-destructive testing technology, which are directly related to the surface chemical composition and microstructure of mineral particles. These parameters reflect the key physicochemical behavioral differences of minerals during the beneficiation process. Specifically, the potential mineral regions identified through near-infrared spectral matching directly indicate the spatial range requiring focused sorting and enrichment, avoiding indiscriminate processing of the entire material flow and improving the targeting of operations. Simultaneously, accurate particle size distribution statistics can pre-determine the passability and behavioral differences of materials of different particle sizes, providing key input parameters for selecting or adjusting screening equipment (such as determining screen aperture and airflow velocity). The spatial correlation between these two types of information transforms preliminary screening from a simple size-based separation into a targeted pre-enrichment operation, effectively improving the recovery rate of useful minerals and reducing ineffective energy consumption. The process of determining the combination of screening parameters helps screening by providing a set of dynamic equipment control instructions that match the surface characteristics and particle size of the ore, ensuring the effectiveness of the preliminary screening process.

[0075] The scanning excitation module 300 is used to scan and excite the screened rare earth ore flow through a coaxially integrated laser-induced breakdown spectroscopy probe and obtain atomic emission spectra.

[0076] In some embodiments, reference Figure 2 The diagram is an exemplary flowchart for obtaining atomic emission spectra in some embodiments of this application. In this embodiment, the atomic emission spectra of the screened rare earth ore stream are obtained by scanning and exciting it using a coaxially integrated laser-induced breakdown spectroscopy probe, which can be achieved through the following steps:

[0077] First, in step S31, the preliminary screened rare earth ore stream is introduced into the secondary conveying unit;

[0078] Then, in step S32, above the secondary conveying unit, the pre-screened rare earth ore flow is scanned and excited by a coaxially integrated laser-induced breakdown spectroscopy probe.

[0079] Finally, in step S33, the atomic emission spectrum obtained by scanning excitation is acquired.

[0080] In specific implementation, the atomic emission spectrum obtained by scanning excitation is acquired as follows: The ore stream, which has undergone preliminary screening and is relatively enriched with target mineral particles, is guided to a secondary conveying unit with higher detection accuracy via a guide plate or secondary conveyor belt. Above this unit, a coaxial probe integrating laser-induced breakdown spectroscopy and near-infrared spectroscopy is fixedly installed. The galvanometer system within the probe controls the laser beam to perform rapid point scanning on the surface of the moving ore particles. When the high-energy pulsed laser is focused on the ore surface, it excites plasma containing elemental information. After a precisely controlled delay, the spectrometer collects specific wavelengths of light emitted by rare earth element atoms or ions during plasma cooling, forming an atomic emission spectrum. To overcome the random error of a single measurement, the same area is usually excited multiple times, and multiple spectra are averaged to obtain a representative atomic emission spectrum curve with a higher signal-to-noise ratio. Other embodiments may also employ other methods, which are not limited here.

[0081] It should be noted that the atomic emission spectrum in this application is a standardized spectral curve generated by laser-induced breakdown plasma and subjected to multiple averaging and noise reduction processes for quantitative analysis of the types and contents of rare earth elements in the ore.

[0082] The signal generation module 400 is used to splice and fuse the atomic emission spectrum and the near-infrared spectral features to obtain a dual-modal feature vector of the ore components in the rare earth ore stream, and to perform attention optimization identification on the modal feature vector focusing on rare earth minerals to obtain a sorting decision signal.

[0083] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart of determining the dual-modal feature vector in some embodiments of this application. In this embodiment, the dual-modal feature vector of the ore composition in the rare earth ore stream is obtained by splicing and fusing the atomic emission spectrum and the near-infrared spectral features using the following steps:

[0084] First, in step S41, the characteristic spectral lines of rare earth elements are extracted from the atomic emission spectrum;

[0085] Then, in step S42, the absorption band characteristics related to rare earth minerals in the near-infrared spectral features are determined;

[0086] Secondly, in step S43, the characteristic spectral lines and the absorption band features are spliced ​​together in full spectrum to obtain the specific mineral information of rare earth minerals;

[0087] Finally, in step S44, the specific mineral information is subjected to feature optimization and fusion to obtain the dual-modal feature vector of the ore composition in the rare earth ore flow.

[0088] In specific implementation, the extraction of rare earth element characteristic spectral lines from the atomic emission spectrum can be achieved in the following manner: The acquired laser-induced breakdown spectrum is preprocessed, including wavelength calibration, continuous background subtraction, and noise smoothing; subsequently, a preset rare earth element characteristic wavelength database is used to identify characteristic emission lines belonging to rare earth elements in the atomic emission spectrum; for example, for neodymium (Nd), the system will locate its characteristic peaks at 401.225 nm, 430.357 nm, etc.; for each identified characteristic spectral line, its net peak intensity is calculated; finally, a set of rare earth element characteristic spectral lines and their corresponding intensity values ​​are output; and the absorption bands related to rare earth minerals in the near-infrared spectral features are determined. The features can be implemented in the following way: preprocessing the near-infrared reflectance spectrum, including multivariate scattering correction to eliminate the influence of scattering and derivative processing; secondly, analyzing the characteristic absorption bands in the near-infrared reflectance spectrum related to the crystal structure of specific rare earth minerals: for example, for bastnaesite, its carbonate groups will produce characteristic combination absorption bands around 2330 nm and 2500 nm; for each characteristic absorption band, extracting its absorption depth, absorption width, and the precise location of the absorption valley after second derivative transformation, and combining all the above parameters to constitute the rare earth mineral-related absorption band features in the near-infrared spectrum features; other embodiments can also be implemented in other ways, which are not limited here.

[0089] In specific implementation, the characteristic spectral lines and absorption band features are spliced ​​together across the entire spectrum to obtain the specific mineral information of rare earth minerals. This can be achieved in the following way: the characteristic spectral lines and absorption band features are combined into two one-dimensional feature vectors: one vector contains the normalized spectral line intensity values ​​of all rare earth elements, and the other vector contains the feature parameter values ​​of all target absorption bands; to ensure data scale consistency, the two vectors are subjected to max-min normalization; then, the two normalized feature vectors are spliced ​​together end-to-end in a predefined order to form a unified one-dimensional composite feature vector; this composite feature vector integrates elemental fingerprint information from atomic emission spectra and molecular structure information from near-infrared spectra, thereby constituting specific mineral information that can characterize the dual elemental-structural properties of rare earth minerals; other embodiments may also use other methods, which are not limited here.

[0090] In specific implementation, the unique mineral information is subjected to feature optimization and fusion to obtain the bimodal feature vector of the ore composition in the rare earth ore stream. This can be achieved in the following way: The unique mineral information is input into a feature optimization module. This module first performs dimensionality reduction processing on the high-dimensional original fused information using principal component analysis to remove noise and redundancy, and extracts a few principal components that can represent most of the data variance. Then, a lightweight attention network is used to perform secondary optimization on these principal component features. Through training, the network assigns appropriate weights to different principal components, automatically enhancing those feature dimensions that contribute significantly to distinguishing different rare earth mineral types and grades (e.g., the intensity of a key spectral line of a certain rare earth element or the depth of a specific absorption band), while suppressing dimensions that contribute less or have greater interference. After the weighting and filtering process, a refined feature vector that integrates elemental and molecular information is output, which is the bimodal feature vector of the ore composition in the rare earth ore stream. Other embodiments can also be implemented in other ways, which are not limited here.

[0091] It should be noted that the characteristic spectral lines in this application are intensity or area data of specific wavelength spectral peaks extracted from atomic emission spectra, identifying the presence of specific rare earth elements; absorption band features are depth, width, or shape parameters of absorption valleys within a specific wavelength range extracted from near-infrared spectra, reflecting the molecular structure and crystal information of rare earth minerals; specific mineral information is fused data formed by splicing characteristic spectral lines and absorption band features at the full spectrum level, comprehensively describing the element-structure properties of the target rare earth mineral; and dual-modal feature vectors are low-dimensional digital feature sets obtained after feature optimization and dimensionality reduction of specific mineral information for efficient discrimination by the input intelligent recognition model. Among these, the deep fusion of elemental information from the atomic emission spectrum of laser-induced breakdown spectroscopy with molecular structural information from near-infrared spectroscopy, this element-molecule dual-modal feature, compared to single spectral features, can more comprehensively and resiliently characterize the essence of the ore. It effectively distinguishes ores with similar elemental compositions but different mineral phases (such as bastnaesite and certain calcium-bearing gangues), significantly improving the robustness and accuracy of subsequent recognition models.

[0092] In some embodiments, the following steps can be used to perform attention-optimized identification focusing on rare earth minerals on the modal feature vectors to obtain sorting decision signals:

[0093] Multiple bias weights are assigned to the rare earth element spectral lines and characteristic functional groups in the dual-modal eigenvectors;

[0094] Attention optimization is performed on the bimodal feature vector based on all the bias weights to obtain multiple feature dimensions focusing on rare earth minerals;

[0095] By performing multi-task analysis and grade estimation on all feature dimensions, multiple grade estimates and mineral categories of rare earth minerals in the screened rare earth ore stream are obtained.

[0096] Sorting decision signals are generated based on the mineral category and all grade estimates.

[0097] In specific implementation, assigning multiple bias weights to the bimodal feature vector related to rare earth element spectral lines and feature functional groups can be achieved in the following way: A lightweight weight generation network is preset, which takes the bimodal feature vector as input. Internally, this network first learns the global contextual relationships of the features through fully connected layers, and then automatically generates a set of importance weights for different dimensions of the feature vector through a dynamic attention modulation module. Specifically, the dynamic attention modulation module identifies and strengthens the weight values ​​corresponding to the feature spectral line intensity dimensions of preset key rare earth elements (such as neodymium, cerium, and praseodymium) and the molecular functional group feature dimensions related to rare earth minerals (such as the carbonate absorption band of bastnaesite), while reducing the weights of features related to common gangue minerals (such as quartz and feldspar). Finally, the network outputs multiple bias weights, each corresponding one-to-one with the dimensions of the input bimodal feature vector, with values ​​ranging from 0 to 1, to characterize the relative importance of each feature dimension for rare earth mineral identification. Other implementation methods can also be used in other embodiments, which are not limited here.

[0098] In specific implementation, attention optimization is performed on the bimodal feature vector based on all bias weights to obtain multiple feature dimensions focusing on rare earth minerals. This can be achieved by multiplying the bias weight vector obtained in the previous step element-wise with the original bimodal feature vector (i.e., the Hadamard product). As a result, the value of the feature dimension with high weight is significantly amplified, while the value of the feature dimension with low weight is relatively suppressed. Consequently, the features strongly correlated with rare earth mineral identification in the weighted bimodal feature vector are highlighted, while redundant noise features are weakened, thus achieving focus on rare earth mineral information. The optimized bimodal feature vector retains the same number of dimensions, but the information density and discriminative power of each dimension are improved. Other implementation methods can also be used in other embodiments, which are not limited here.

[0099] In specific implementation, multi-task parsing and grade estimation are performed across all feature dimensions to obtain multiple grade estimates and mineral categories of rare earth minerals in the screened rare earth ore stream. This can be achieved as follows: The attention-optimized feature vector is input into a parsing network using a multi-task learning architecture. This network shares the bottom feature extraction layer and then connects two specific task heads in parallel at the top: a mineral classification head, typically a softmax classifier responsible for mapping features to probabilities of different mineral categories, such as "fluorite," "monazite," "yttrium phosphate," or "gangue"; and a grade regression head, typically a linear regression layer connected to a network (such as DBP-ANN), responsible for predicting the total rare earth oxide (TREO) content or grade estimate of a specific key element for the ore particle. During training, the network uses labeled samples in an end-to-end manner, and the total loss function is a weighted sum of the classification loss and regression loss. During deployment, the network performs a forward propagation of the features of each ore particle once, simultaneously outputting its most likely mineral category and predicted grade estimate. Other implementation methods can also be used in other embodiments, which are not limited here.

[0100] In specific implementation, the sorting decision signal generated based on the mineral category and all grade estimates can be achieved in the following way: a series of sorting thresholds based on production process requirements are preset, such as "lower limit of concentrate grade" and "upper limit of tailings grade"; the decision logic is as follows: first, it is determined whether the mineral category belongs to the target rare earth mineral. If so, its predicted grade estimate is further compared with the threshold; if the grade is higher than the lower limit of concentrate, a sorting instruction is generated; if the grade is between the upper limit of tailings and the lower limit of concentrate, a middlings reprocessing instruction is generated; if the mineral category is determined to be gangue, or its grade is lower than the upper limit of tailings, a waste disposal instruction is directly generated; at the same time, to ensure reliability, the decision will refer to the classification confidence level (such as the maximum probability) output by the model; when the confidence level is too low, a conservative instruction of inspection or default waste disposal can be triggered; finally, this action instruction, the real-time spatial coordinates of the rare earth ore on the conveyor belt, and the predicted grade value are encapsulated into a structured data packet, i.e., the sorting decision signal; other embodiments can also be implemented in other ways, which are not limited here.

[0101] It should be noted that the bias weights in this application are importance coefficients dynamically allocated to different dimensions of the bimodal feature vector by the attention mechanism to amplify key signals and suppress noise during the identification process; the feature dimensions are feature vectors that carry mineral identification information after focusing and are optimized and weighted by the bias weights; the grade estimate and mineral category are quantitative grade prediction values ​​and qualitative mineral classification results that represent the quality and type of a single ore particle and are output in parallel by a multi-task analytical model; the sorting decision signal refers to the structured control command that drives the final actuator and is generated by comparing the grade estimate and mineral category with the process threshold, and includes action instructions, target coordinates and predicted grade.

[0102] The re-sorting module 500 is used to adjust the parameters of the pneumatic separator according to the sorting decision signal, thereby re-sorting the rare earth ore stream.

[0103] In some embodiments, the parameters of the pneumatic separator are adjusted according to the sorting decision signal to further sort the rare earth ore stream, which can be achieved by the following steps:

[0104] The sorting decision signal is analyzed to obtain the spatial coordinates, action commands, and predicted grade values ​​of the target particles;

[0105] Based on the spatial coordinates and the conveyor belt speed, calculate the precise delay for triggering the corresponding sorting nozzles;

[0106] Based on the action command and the predicted grade value, the blowing air pressure and pulse width are dynamically set;

[0107] When the precise delay arrives, the corresponding nozzles are controlled to perform blowing with the set parameters to separate the particles into the corresponding concentrate, middlings or tailings collection tanks.

[0108] In specific implementation, the precise delay for triggering the corresponding sorting nozzles based on the spatial coordinates and conveyor belt speed can be achieved in the following way: Based on the longitudinal position of the target particle in the spatial coordinates obtained from the analysis, and the current actual running linear speed of the secondary conveyor belt, a precise time delay is calculated in real time. This delay represents the time required from the current moment until the target particle moves with the conveyor belt to the predetermined blowing point directly below the pneumatic nozzle array. The dynamic setting of the blowing air pressure and pulse width based on the action command and predicted grade value can be achieved in the following way: Based on the analyzed action command and predicted grade value, the blowing parameters are dynamically set by querying a preset parameter mapping table. This mapping table defines process rules: for example, for particles with a "selection" command and high grade, a lower air pressure and a shorter pulse width are set to achieve precise and gentle separation and prevent particle splashing; for particles with a "discard" command, a higher air pressure and a standard pulse width are set to ensure they are completely blown away from the main track; other methods can also be used in other embodiments, which are not limited here.

[0109] In summary, the technical solution adopted in this application can accurately identify the characteristics of rare earth minerals and perform intelligent screening and conveying for efficient rare earth ore beneficiation.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0112] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. An integrated system of intelligent screening and conveying for rare earth ore beneficiation process, wherein, The rare earth ore stream is pre-crushed and fed into a conveyor belt. The intelligent integrated screening and conveying system comprises: Data acquisition module: Performs spectral detection and image acquisition on the rare earth ore flow on the conveyor belt to obtain image features, near-infrared spectral features and X-ray fluorescence spectral data of the rare earth ore flow; Preliminary screening module: Based on the image features, the near-infrared spectral features, and the spectral data, determine the mineral regions and particle size distribution of rare earth minerals in the rare earth ore stream, and then determine the combination of screening parameters and adjust the screening equipment to perform preliminary screening of mineral particles with different surface properties; Scanning excitation module: The selected rare earth ore flow is scanned and excited by a coaxially integrated laser-induced breakdown spectroscopy probe, and atomic emission spectra are obtained; Signal generation module: Based on the atomic emission spectrum and the near-infrared spectral features, the module splices and fuses them to obtain a dual-modal feature vector of the ore components in the rare earth ore stream. The module then performs attention optimization identification on the modal feature vector focusing on rare earth minerals to obtain a sorting decision signal. Secondary sorting module: Adjusts the parameters of the pneumatic separator according to the sorting decision signal, thereby re-sorting the rare earth ore stream; Specifically, determining the mineral regions and grain size distribution of rare earth minerals in the rare earth ore flow based on the image features, near-infrared spectral features, and spectral data includes: performing spectral matching on the near-infrared spectral features to obtain potential regions of rare earth minerals; using the potential regions as spatial guides, performing deformable convolution semantic segmentation on the image features to obtain segmentation masks; segmenting the regions of rare earth minerals based on all segmentation masks, and enhancing the edges of all segmented regions to obtain multiple segmented ore regions; determining a grain size distribution histogram based on all segmented ore regions; and cross-correlating the grain size distribution histogram with the potential regions to obtain the mineral regions and grain size distribution of rare earth minerals in the rare earth ore flow. Specifically, the process of splicing and fusing the atomic emission spectrum and the near-infrared spectral features to obtain the dual-modal feature vector of the ore composition in the rare earth ore stream includes: extracting the characteristic spectral lines of rare earth elements from the atomic emission spectrum; determining the absorption band features related to rare earth minerals in the near-infrared spectral features; splicing the characteristic spectral lines and the absorption band features across the entire spectrum to obtain specific mineral information of rare earth minerals; and performing feature optimization and fusion on the specific mineral information to obtain the dual-modal feature vector of the ore composition in the rare earth ore stream. Specifically, the process of focusing attention optimization on rare earth minerals in the modal feature vector to obtain a sorting decision signal includes: assigning multiple bias weights to the rare earth element spectral lines and characteristic functional groups in the dual-modal feature vector; performing attention optimization on the dual-modal feature vector based on all bias weights to obtain multiple feature dimensions focusing on rare earth minerals; performing multi-task analysis and grade estimation through all feature dimensions to obtain multiple grade estimates and mineral categories of rare earth minerals in the screened rare earth ore stream; and generating a sorting decision signal based on the mineral categories and all grade estimates.

2. The intelligent screening and conveying integrated system for rare earth ore dressing process according to claim 1, characterized in that, Spectral detection and image acquisition were performed on the rare earth ore flow on the conveyor belt to obtain image features, near-infrared spectral features, and X-ray fluorescence spectral data of the rare earth ore flow, specifically including: An integrated detection unit consisting of an industrial camera, a near-infrared spectrometer, and a micro-area X-ray fluorescence spectrometer is fixedly installed above the conveyor belt; Set up a central timing controller; When the rare earth ore flow passes through the detection area at a constant speed with the conveyor belt, the detection unit is synchronously triggered by the central timing controller to perform data acquisition, thereby obtaining the image features, near-infrared spectral features, and X-ray fluorescence spectral data of the rare earth ore flow.

3. The intelligent screening and conveying integrated system for rare earth ore dressing process according to claim 1, characterized in that, Determining the combination of screening parameters and adjusting the screening equipment for preliminary screening of mineral particles with different surface properties specifically includes: Multiple surface properties of minerals in the rare earth ore flow were extracted based on the near-infrared spectral characteristics. The parameter combinations for multiple screenings are determined based on all surface properties and the particle size distribution. All parameter combinations are sent to the screening equipment, and the servo mechanism of the screening equipment is adjusted to perform preliminary screening of mineral particles with different surface characteristics.

4. The intelligent screening and conveying integrated system for rare earth ore dressing process according to claim 1, characterized in that, The selected rare earth ore flow was scanned and excited using a coaxially integrated laser-induced breakdown spectroscopy probe, and atomic emission spectra were obtained. Specifically, this included: The preliminary screened rare earth ore stream is then introduced into the secondary conveying unit; Above the secondary conveying unit, a coaxially integrated laser-induced breakdown spectroscopy probe is used to scan and excite the pre-screened rare earth ore stream. Obtain the atomic emission spectra obtained from scanning excitation.

5. The integrated system of screening and conveying for rare earth ore beneficiation process according to claim 1, characterized in that, The parameters of the pneumatic separator are adjusted according to the sorting decision signal to further sort the rare earth ore stream. Specifically, this includes: The sorting decision signal is analyzed to obtain the spatial coordinates, action commands, and predicted grade values ​​of the target particles; Based on the spatial coordinates and the conveyor belt speed, calculate the precise delay for triggering the corresponding sorting nozzles; Based on the action command and the predicted grade value, the blowing air pressure and pulse width are dynamically set; When the precise delay arrives, the corresponding nozzles are controlled to perform blowing with the set parameters to separate the particles into the corresponding concentrate, middlings or tailings collection tanks.

6. The integrated system of screening and conveying for rare earth ore beneficiation process according to claim 1, characterized in that, Image features of rare earth ore flows were acquired using industrial cameras.

7. The integrated system of screening and conveying for rare earth ore beneficiation process according to claim 1, characterized in that, X-ray fluorescence spectrometry data of rare earth ore flows were collected using an X-ray fluorescence spectrometer.

Images