Method for sorting low-sulfur gold ore and related equipment
By combining spectral testing and sorting models, the problem of low identification rate of low-sulfur gold ore was solved, and a highly efficient and automated sorting method was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF MINERAL RESOURCES CHINA METALLURGICAL GEOLOGY ADMINISTRATION
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-09
Smart Images
Figure CN122173984A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer intelligent sorting technology, and in particular to a sorting method and related equipment for low-sulfur gold ore. Background Technology
[0002] Low-sulfur gold deposits (such as Carlin-type gold deposits and some altered rock-type gold deposits) are an important type of gold resource in my country. They are characterized by extremely fine gold mineral grains, often occurring as "invisible gold" within pyrite, arsenopyrite, or directly within gangue minerals (such as quartz, sericite, and kaolinite), and low sulfide mineral content. For low-sulfur gold deposits, the color and texture changes caused by mineralization and alteration (such as silicification, sericitization, and argillization) are often very subtle, with little difference from the surrounding rock, resulting in extremely low recognition rates for traditional photoelectric sorting machines, making them unsuitable for industrial applications. Summary of the Invention
[0003] In view of this, the purpose of this application is to propose a sorting method and related equipment for low-sulfur gold ore, so as to solve the problem of low identification rate of low-sulfur gold ore in related technologies.
[0004] To achieve the above objectives, this application provides a method for separating low-sulfur gold ore, comprising:
[0005] Obtain the ore block to be sorted, and perform spectral testing on the ore block to obtain spectral data; Extract the target band data from the spectral data, and generate a spectral feature vector based on the target band data; The spectral feature vector is input into a pre-trained sorting model, and the sorting model outputs the predicted gold grade of the ore block to be sorted. In response to the predicted gold grade being greater than a preset gold grade threshold, the ore block to be sorted is determined to be a gold ore.
[0006] Optionally, the target band data includes the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity calculated based on the absorption depth of multiple target bands; The step of generating a spectral feature vector based on the target band data includes: The spectral feature vector is constructed based on the position of the maximum absorption band, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands.
[0007] Optionally, before constructing the spectral feature vector based on the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands, the following steps are included: Obtain the globally optimal feature parameters of the sorting model; The absorption width, the position of the maximum absorption band, the absorption depth, and the illite crystallinity are corrected using the globally optimal characteristic parameters.
[0008] Optionally, the training method for the sorting model includes: Build the training dataset; Construct the nonlinear regression function and its constraints; The nonlinear regression function and constraints are converted into an initial sorting model using the Lagrange multiplier method. Based on the training dataset, the initial sorting model is iteratively trained to obtain the sorting model.
[0009] Optionally, constructing the training dataset includes: Collect samples from multiple mineral blocks; Spectroscopic detection was performed on the multiple ore block samples to obtain spectral data samples, and the multiple ore block samples were tested using the fire assay method to determine the gold grade label corresponding to each ore block sample. Extract the target band data from the spectral data sample, and generate a spectral feature vector sample based on the target band data; An initial training dataset is generated based on the spectral feature vector samples and the gold grade labels; The initial training dataset is filtered to obtain the training dataset.
[0010] Optionally, the step of filtering the initial training dataset to obtain the training dataset includes: Calculate the Euclidean distance between each pair of spectral feature vector samples in the initial training dataset, and select the two spectral feature vector samples corresponding to the maximum Euclidean distance to put into a pre-constructed set; Perform the following multi-round iterative operation on the remaining spectral feature vector samples in the initial training dataset: Calculate the minimum Euclidean distance between each remaining spectral feature vector sample in the initial training dataset and all spectral feature vector samples in the set; The spectral feature vector samples corresponding to the maximum value of all minimum Euclidean distances are placed into the set. In response to determining that the number of spectral feature vector samples in the set is less than a preset threshold, the next round of iteration is initiated. In response to determining that the number of spectral feature vector samples in the set is equal to the preset threshold, the multi-round iteration is terminated, and the training dataset is obtained.
[0011] Optionally, the step of iteratively training the initial sorting model based on the training dataset to obtain the sorting model includes: Initialize the population, where each individual corresponds to a set of feature parameters, and perform the following multi-round iterative operation: Based on the population, a new sterile line population is generated using a hybridization breeding algorithm; For each new sterile line individual in the new sterile line population, the training dataset is corrected according to the characteristic parameters of the new sterile line individual, and the initial sorting model is configured according to the characteristic parameters of the new sterile line individual; The configured initial sorting model is trained based on the calibrated training dataset. The fitness of the new sterile line individuals is calculated based on the trained initial sorting model on the calibrated training dataset. The population is iteratively updated based on the fitness of the new sterile line individuals. If the preset iteration stop condition is not met, the next iteration operation is entered. If the preset iteration stop condition is met, the multi-round iteration operation is exited, and the characteristic parameters of the individual with the minimum fitness in this round are determined as the global optimal characteristic parameters. The global optimal feature parameters are substituted into the initial sorting model to obtain the sorting model.
[0012] Optionally, generating a new sterile line population based on the population using a hybridization breeding algorithm includes: The new sterile line population is generated using the following formula: in, Represents each individual in the population Corresponding feature parameters Includes penalty factors Kernel function parameters Absorption width parameter and absorption feature offset parameters ; express The first individual of the new sterile line generated by hybridization during the round breeding process Vigene, express Maintain the first in the line during rotational breeding The first individual Vigene, express The first sterile line in the rotational breeding The first individual Vigene, , The two numbers are random numbers in the range [0, 1] and are not equal; during random hybridization, During enantiomeric hybridization, ;like Its fitness value is better than Then replace To update the sterile line, or to retain the original individual.
[0013] Based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described above when executing the computer program.
[0014] Based on the same inventive concept, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the method described above.
[0015] As described above, the low-sulfur gold ore sorting method and related equipment provided in this application include the following steps: acquiring a block of ore to be sorted; performing spectral testing on the block to obtain spectral data; extracting target band data from the spectral data; and generating a spectral feature vector based on the target band data. The spectral feature vector accurately reflects the actual mineral composition of the block of ore to be sorted. The spectral feature vector is input into a pre-trained sorting model, which outputs the predicted gold grade of the block of ore to be sorted. The sorting model is pre-trained and can accurately determine the gold grade corresponding to the spectral feature vector. In response to the predicted gold grade being greater than a preset gold grade threshold, the block of ore to be sorted is determined to be gold ore. This method improves the accuracy and efficiency of gold ore identification. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic flowchart of a sorting method for low-sulfur gold ore according to an embodiment of this application. Figure 2 This is a flowchart illustrating the sorting model training method according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a sorting device for low-sulfur gold ore according to an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.
[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0020] As described in the background section, low-sulfur gold deposits (such as Carlin-type gold deposits and some altered rock-type gold deposits) are an important type of gold resource in my country. They are characterized by extremely fine gold mineral grains, often occurring as "invisible gold" in pyrite, arsenopyrite, or directly within gangue minerals (such as quartz, sericite, kaolinite, etc.), and low sulfide mineral content.
[0021] Traditional manual sorting or photoelectric sorting is inefficient. Manual sorting relies on worker experience, is labor-intensive, inefficient, has inconsistent sorting accuracy, and cannot identify mineralization without visible features. Traditional photoelectric sorting mainly relies on the differences in color and visible light reflectance between ore and waste rock. For low-sulfur gold deposits, the color and texture changes caused by mineralization alteration (such as silicification, sericitization, and argillization) are often very subtle, with little difference from the surrounding rock, resulting in extremely low recognition rates for traditional photoelectric sorting machines, making them unsuitable for industrial applications.
[0022] In view of this, this application proposes a sorting method for low-sulfur gold ore, which combines microscopic spectral information and macroscopic sorting decisions to provide a sorting method for low-sulfur gold ore that can capture the spectral fingerprint of its key alteration minerals and automatically determine the gold grade.
[0023] The embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0024] This application provides a method for sorting low-sulfur gold ore, referring to... Figure 1 This includes the following steps: Step 101: Obtain the ore block to be sorted, and perform spectral testing on the ore block to be sorted to obtain spectral data.
[0025] Specifically, the ore blocks to be sorted are collected rock samples. Spectral testing is performed on the ore blocks to identify spectral characteristics related to them, obtaining spectral data. In practice, detailed spectral data of the ore blocks in the visible-near-infrared-shortwave infrared band (400nm-2500nm) is acquired. The average of multiple measurements can be used as the final spectral data to eliminate noise interference. For example, the spectral test can be a rock diffuse reflectance spectral test, measured using a ground-based spectrometer in the 400-2500nm band.
[0026] Step 102: Extract the target band data from the spectral data, and generate a spectral feature vector based on the target band data.
[0027] Specifically, the spectral data includes information such as the position (P) of the maximum absorption band, absorption depth (D), absorption width (W), and symmetry (S) at 560nm, 2175nm, 2205nm, and 2330nm. It also includes the absorption depth at 1910nm. The ratio of the absorption depths at 2205nm and 1910nm (D2205 / D1910) is calculated as the illite crystallinity (IC) data.
[0028] 560 nm, 1910 nm, 2175 nm, 2205 nm, and 2330 nm are diagnostic absorption bands for altered minerals in hydrothermal gold deposits, corresponding to iron staining alteration, hydroxyl background absorption, sericitization / illite alteration, argillization, and carbonatization, respectively. These bands can comprehensively characterize alteration information directly related to the intensity of gold mineralization. During gold deposit formation, hydrothermal fluids oxidize iron in the host rock, resulting in iron staining alteration. 560 nm represents Fe... 3+ Strong absorption peaks generated by electron transitions can quickly identify the iron staining intensity of mineralization zones, serving as the most direct surface / core indicator of gold mineralization. 1910nm is a universal reference peak for short-wave infrared radiation, corresponding to all minerals containing -OH hydroxyl groups and water of crystallization. Clay, mica, and carbonates all exhibit strong absorption at 1910nm, which can be used as background absorption peaks to calculate illite crystallinity, eliminating differences in water content and retaining only true mineralization information. 2205nm corresponds to illite and sericite; sericite is the most characteristic and core alteration in hydrothermal gold deposits. 2205nm is a diagnostic absorption peak for Al-OH stretching and bending coupled vibrations; stronger absorption and more regular morphology indicate stronger alteration and a higher probability of gold mineralization, positively correlated with the gold grade of the ore block. 2175nm corresponds to montmorillonite, kaolinite, and some altered mica, and can be used to eliminate interference from non-mineralized clays, improving model accuracy. The minerals corresponding to 2330nm are calcite, dolomite, and ferrodolomite. Gold mineralization is often accompanied by carbonatization and alteration of the host rock; 2330nm represents Mg-OH / CO3. 2-Vibration absorption is used to determine the hydrothermal environment of ore formation and to assist in ore delineation.
[0029] Furthermore, generating a spectral feature vector based on the target band data includes: The spectral feature vector is constructed based on the position of the maximum absorption band, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands. The feature parameters of all target bands are integrated with the illite crystallinity to construct a multidimensional spectral feature vector containing wavelength position, absorption depth, absorption width, symmetry, and crystallinity, which serves as input data for the subsequent sorting model.
[0030] Furthermore, before constructing the spectral feature vector based on the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands, the following steps are included: Obtain the global optimal feature parameters of the sorting model; use the global optimal feature parameters to correct the absorption width, the position of the maximum absorption band, the absorption depth and the illite crystallinity.
[0031] Specifically, the globally optimal feature parameters are the optimal parameters determined by the sorting model after training, including the optimal absorption width parameter. and optimal absorption feature offset parameters By using the optimal absorption width parameter and optimal absorption feature offset parameters Corrections were made to the absorption width, maximum absorption band location, absorption depth, and illite crystallinity in the target band data.
[0032] Corrected absorption width , Indicates the absorption width in the target band data; the location of the corrected maximum absorption band. , Indicates the location of the maximum absorption band in the target band data; corrected absorption depth. , Indicates the absorption depth in the target band data, and the corrected illite crystallinity. .
[0033] The corrected absorption width, maximum absorption band position, absorption depth, and illite crystallinity are encoded into spectral feature vectors. These spectral feature vectors can be further normalized to eliminate dimensional effects, and the normalized spectral feature vectors serve as input data for the final sorting model.
[0034] Step 103: Input the spectral feature vector into the pre-trained sorting model, and output the predicted gold grade of the ore block to be sorted through the sorting model.
[0035] Specifically, load the trained sorting model. Support vectors Support vectors coefficient bias Optimal kernel function parameters The sorting model is shown in the following formula: (1), (2), in, Represents the spectral eigenvector. Represents the kernel function. and It is a Lagrange multiplier. This represents the i-th training sample in the training dataset of the sorting model, where each training sample corresponds to a pair of Lagrange multipliers. and , This represents the total number of samples in the training dataset. Support vectors Support vectors coefficient bias Optimal kernel function parameters These are all fixed parameters obtained after the sorting model has been trained.
[0036] The spectral feature vectors are input into formula (1) for forward calculation to obtain the predicted gold grade of the ore block to be sorted. For each support vector... Calculate kernel function value Multiply by the support vector coefficient Sum the results of all support vectors, then calculate the sum with... The sum can be used to obtain the final value. The numerical value, i.e., the predicted gold grade. For example, 0.5, by predicting the gold grade, it can be determined whether the ore block is gold ore or waste rock.
[0037] Step 104: In response to the predicted gold grade being greater than the preset gold grade threshold, determine that the ore block to be sorted is a gold ore.
[0038] Specifically, if the predicted gold grade is greater than a preset gold grade threshold, the ore block to be sorted is determined to be gold ore; if the predicted gold grade is less than or equal to the preset gold grade threshold, the ore block to be sorted is determined to be waste rock. For example, the preset gold grade threshold is 0.2.
[0039] Based on steps 101 to 104 above, the sorting method for low-sulfur gold ore provided in this embodiment includes: acquiring a block of ore to be sorted; performing spectral testing on the block to obtain spectral data; extracting target band data from the spectral data; and generating a spectral feature vector based on the target band data. The spectral feature vector accurately reflects the actual mineral composition of the block of ore to be sorted. The spectral feature vector is input into a pre-trained sorting model, and the model outputs the predicted gold grade of the block of ore to be sorted. The sorting model is pre-trained and can accurately determine the gold grade corresponding to the spectral feature vector. In response to the predicted gold grade being greater than a preset gold grade threshold, the block of ore to be sorted is determined to be gold ore. The method of this embodiment improves the accuracy and efficiency of gold ore identification.
[0040] In some embodiments, reference Figure 2 The training method for the sorting model includes the following steps: Step 201: Construct the training dataset.
[0041] Furthermore, step 201 includes: Step a: Collect multiple mineral block samples.
[0042] Specifically, multiple ore block samples were collected, covering rock samples collected from different areas of the mine, including high-grade ore blocks, low-grade ore blocks, and waste rock, covering various alteration types such as silicification, sericitization, carbonatization, and pyritization.
[0043] Step b involves performing spectral analysis on the multiple ore block samples to obtain spectral data samples, and then using the fire assay method to test the multiple ore block samples to determine the gold grade label corresponding to each ore block sample.
[0044] Specifically, for each ore block sample, its detailed spectral data in the visible-near-infrared-shortwave infrared band (400-2500 nm) is extracted to obtain a spectral data sample. To eliminate noise interference, the spectral data of each ore block sample is measured three times, and the average of the three measurements is taken as the final spectral data sample. The gold grade of each ore block sample is tested using the fire assay method, and this result serves as the gold grade label for each ore block sample, used in the construction of the subsequent training dataset.
[0045] Step c: Extract the target band data from the spectral data sample, and generate a spectral feature vector sample based on the target band data.
[0046] Specifically, the target wavelengths include 560 nm, 1910 nm, 2175 nm, 2205 nm, and 2330 nm. The target wavelength data includes information such as the maximum absorption position (P), absorption depth (D), absorption width (W), and symmetry (S) at 560 nm, 2175 nm, 2205 nm, and 2330 nm, as well as the absorption depth at 1910 nm. The ratio of the absorption depths at 2205 nm to 1910 nm (D2205 / D1910) is calculated as illite crystallinity (IC) data. The above target wavelength data is encoded to generate spectral feature vector samples.
[0047] Step d: Generate an initial training dataset based on the spectral feature vector samples and the gold grade label. The initial training dataset is constructed based on the spectral feature vector samples and the gold grade label.
[0048] Step e: Filter the initial training dataset to obtain the training dataset.
[0049] Furthermore, step e includes: Calculate the Euclidean distance between each pair of spectral feature vector samples in the initial training dataset, and select the two spectral feature vector samples corresponding to the maximum Euclidean distance to be placed into a pre-constructed set. Perform the following multi-round iterative operation on the remaining spectral feature vector samples in the initial training dataset: calculate the minimum Euclidean distance between each remaining spectral feature vector sample in the initial training dataset and all spectral feature vector samples in the set; place the spectral feature vector sample corresponding to the maximum value among all minimum Euclidean distances into the set; in response to determining that the number of spectral feature vector samples in the set is less than a preset threshold, proceed to the next round of iterative operation; in response to determining that the number of spectral feature vector samples in the set is equal to the preset threshold, exit the multi-round iterative operation to obtain the training dataset.
[0050] Specifically, the Euclidean distance between each pair of spectral feature vector samples in the initial training dataset is calculated, and the two spectral feature vector samples corresponding to the largest Euclidean distance are selected and added to a pre-constructed set. The remaining spectral feature vector samples in the initial training dataset are iterated through, and the Euclidean distance from each sample to all samples in the set is calculated. The minimum Euclidean distance for each remaining spectral feature vector sample is obtained, and the sample with the largest minimum distance is added to the set. This process is repeated until the number of samples in the set reaches 70% of the total number of samples (i.e., a preset threshold), at which point the iteration process stops, and the resulting set is used as the training dataset. Spectral feature vector samples from the initial training dataset that are not added to the training dataset are combined to form the test dataset. 10% of the samples in the training dataset are randomly selected and moved to the test dataset, and 10% of the samples in the test dataset are simultaneously moved to the training dataset. The final training and test datasets are then determined. This method of selecting training datasets ensures the representativeness and balance of the training dataset, avoids overfitting and insufficient generalization caused by random partitioning, and guarantees that the training dataset covers the complete feature space.
[0051] Step 202: Construct the nonlinear regression function and constraints.
[0052] Specifically, the nonlinear regression function is shown in the following equation: (3), The constraints are shown in the following equation: (4), in, Indicates the penalty factor. Indicates the insensitive loss function. Represents slack variables. Represents the model weight vector. Indicates bias. This represents a sample of spectral feature vectors in the training dataset. This represents the gold grade label in the training dataset. This represents the total number of training samples in the training dataset.
[0053] Step 203: Convert the nonlinear regression function and constraints into an initial sorting model using the Lagrange multiplier method.
[0054] Specifically, the optimization variable of the nonlinear regression function in step 202 is , as well as ,in The dimension of the spectral feature vector is exactly equal to the dimension of the spectral feature vector. When the dimension of the spectral feature vector is high, the number of optimization variables will increase dramatically with the feature dimension, and the computational load and memory usage will increase exponentially. The constraints in formula (4) are complex, including inequality constraints and non-negativity constraints. The number of constraints increases linearly with the number of samples, making it extremely difficult and unstable to solve directly using conventional optimization algorithms. Therefore, it is necessary to reduce the optimization complexity. In this embodiment, the original optimization problem is solved by the Lagrange multiplier method to obtain the dual objective function: (5), The corresponding constraints are as follows: (6), The final initial sorting model is as follows: (7), The kernel function is as follows: (8), in, and Represents the Lagrange multipliers. Represents the kernel function. Represents support vectors, Represents the spectral eigenvector. Indicates bias. Represents kernel function parameters. The penalty factor is represented by the kernel function. The nonlinear mapping from low-dimensional spectral features to high-dimensional space is achieved through the kernel function, transforming the nonlinear problem into a linearly solvable problem in the high-dimensional space. Furthermore, explicit computation of the high-dimensional vector is unnecessary; the kernel function value can be directly used to replace the inner product. The constraints in equation (6) are also greatly simplified.
[0055] Step 204: Based on the training dataset, iteratively train the initial sorting model to obtain the sorting model.
[0056] Furthermore, step 204 includes: Initialize the population, where each individual corresponds to a set of feature parameters, and perform the following multi-round iterative operation: Based on the population, a new sterile line population is generated using a hybridization breeding algorithm. For each new sterile line individual in the new sterile line population, the training dataset is corrected according to the characteristic parameters of the new sterile line individual, and the initial sorting model is configured according to the characteristic parameters of the new sterile line individual. The configured initial sorting model is trained based on the corrected training dataset, and predictions are made on the corrected training dataset based on the trained initial sorting model to calculate the fitness of the new sterile line individual. The population is iteratively updated according to the fitness of the new sterile line individual. In response to the failure to reach a preset iteration stop condition, the next round of iteration operation is entered. In response to the achievement of the preset iteration stop condition, the multi-round iteration operation is exited, and the characteristic parameters of the individual with the minimum fitness in this round are determined as the global optimal characteristic parameters. The global optimal characteristic parameters are substituted into the initial sorting model to obtain the sorting model.
[0057] Specifically, the goal of iterative training in this step is to minimize the prediction error of the sorting model for gold grade, that is, to minimize the difference between the predicted value of the sorting model and the gold grade label obtained by fire assay. A commonly used evaluation metric (i.e., fitness) is the root mean square error (RMSE). The smaller the RMSE, the better the parameter combination. The RMSE calculation method is as follows: (9), in, This represents the predicted value of the i-th spectral feature vector sample. Let N represent the gold grade label of the i-th spectral feature vector sample, and N represent the total number of training samples.
[0058] Each individual in the population corresponds to a set of feature parameters. The feature parameters X include four-dimensional parameters { , , , The population includes maintainer lines and sterile lines. Maintainer lines are the core population storing "superior genes," and individuals within them are parameter combinations with better fitness selected in previous iterations. They are not actively updated during iterations but only serve as "fathers" for hybridization, providing superior parameters to ensure the population's superior genes are not lost. Sterile lines are the core population for iterative optimization. Individuals within them are parameter combinations to be selected and evolved. New individuals are generated through hybridization with maintainer lines, continuously weeding out inferior individuals until convergence to the optimal parameters. The initial population size (the number of individuals in each type) can be set to 20-50. Too small a size can easily lead to local optima, while too large a size increases computational complexity. Preset iteration stopping conditions can include two types: one is the maximum number of iterations, which can be set to 50-100; the other is a convergence threshold, where the minimum fitness difference between two adjacent iterations is less than a preset threshold (e.g., 10). -6 This indicates that the parameters have stabilized and there is no need to continue iterating; the multi-round iteration operation can be exited.
[0059] Before the iteration begins, the population is initialized, and an initial sterile line is randomly generated. Within a preset parameter value range, N individuals (population size) with four-dimensional parameters are randomly generated to form the initial sterile line. Each individual is X{ , , , To ensure that parameters differ between individuals (which is beneficial for maintaining diversity), all individuals from the initial sterile line are directly copied to the maintainer line as the "paternal parent" for the first round of hybridization. Subsequent maintainer lines are updated only by "retaining the best individuals," and are not randomly generated. Individuals with parameters that are out of range or meaningless (such as...) are removed. , The population is then randomly regenerated to ensure that all individuals in the initial population are valid.
[0060] Individuals are randomly selected from the maintainer line (the a-th individual) and from the sterile line (the b-th individual) as parents for hybridization. Hybridization rules are followed: random hybridization, a ≠ b; enantiomeric hybridization, a = b. In each round of hybridization, the two rules can be mixed in a 7:3 ratio. Two distinct random numbers within the interval [0,1] are generated. , As the hybridization weight, the new sterile line population is generated using the following formula: (10) in, This represents the feature parameters corresponding to each individual in the population, including the penalty factor. Kernel function parameters Absorption width parameter and absorption feature offset parameters ; express The first individual of the new sterile line generated by hybridization during the round breeding process Vigene, express Maintain the first in the line during rotational breeding The first individual Vigene, express The first sterile line in the rotational breeding The first individual Vigene, , The two numbers are random numbers in the range [0, 1] and are not equal; during random hybridization, During enantiomeric hybridization, ;like Its fitness value is better than Then replace To update the sterile line, or to retain the original individual.
[0061] For each individual X, the first... A dimensional gene refers to the d-th parameter in the four-dimensional parameters of an individual, where d ranges from 1 to 4. For example, the first parameter is... The second parameter is The third parameter is The fourth parameter is For each individual in the new sterile line population, extract its corresponding four-dimensional parameters { , , , },pass and The training dataset is then calibrated using the same method as in the previous embodiments, and will not be repeated here. and Configure the initial sorting model, which is about to and Substituting into formulas (5) to (8), the initial sorting model for this round is determined. The initial sorting model for this round is trained using the calibrated training dataset. The core is to solve the objective function in formula (5) using a quadratic programming algorithm (such as Sequential Minimal Optimization, SMO) to find the optimal Lagrange multiplier. and Most of them and Zero, non-zero and The corresponding support vectors are calculated from the support vectors that satisfy formula (6). The support vectors, support vector coefficients, biases, and kernel function parameters are solidified to form the initial sorting model completed in this round of training.
[0062] Then, based on the trained initial sorting model, predictions are made on the calibrated training dataset, and the fitness shown in formula (9) is calculated as the fitness of each individual in the new sterile line population. According to the principle of "optimal fitness", the sterile line and the maintainer line are updated to gradually improve the overall quality of the population: For each individual in the new sterile line population, the fitness is compared with that of the corresponding individual in the original sterile line population. If the fitness of the new sterile line individual is smaller, the new sterile line individual replaces the original sterile line individual. If the fitness of the new sterile line individual is larger, the original sterile line individual is retained and the new sterile line individual is eliminated. Finally, the updated sterile line is obtained, and the overall fitness is better than the previous round. All individuals in the current round of maintainer line and updated sterile line are merged, the fitness of each individual is extracted, sorted according to fitness from smallest to largest, and the top N best individuals are selected to form a new generation of maintainer line, which is used as the maintainer line for the next round of iteration. The minimum fitness of all individuals in this round of iteration and the corresponding four-dimensional parameter X{ are recorded. , , , If the preset iteration stopping condition is not met, proceed to the next iteration round; if the preset iteration stopping condition is met, exit the multi-round iteration operation. Set the four-dimensional parameters X{ of the minimum fitness individual in this round. , , , } is used as the globally optimal feature parameter. By fixing the globally optimal feature parameter, the sorting model is obtained.
[0063] The training method of the sorting model in this embodiment establishes a nonlinear mapping relationship between spectral features and gold grade, solving the problem of traditional geological analysis relying on experience and being difficult to quantify. A hybridization breeding mechanism is nested during training to simultaneously optimize four-dimensional parameters, avoiding the subjectivity of manual parameter tuning and making the model more adaptable to the characteristics of gold ore spectral data. After model training, only support vectors retain non-zero Lagrange multipliers, becoming the core basis for the sorting model's predictions. This process automatically eliminates redundant samples that do not contribute to the prediction, reducing model complexity and improving inference efficiency, making it suitable for rapid on-site detection scenarios in mines. After training, the sorting model can directly input the spectral features of the ore block and output continuous gold grade prediction values, replacing the traditional fire assay method (which takes several days and is costly), shortening the detection cycle to minutes and significantly improving exploration efficiency.
[0064] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.
[0065] It should be noted that some embodiments of this application have been described above. In some cases, the actions or steps described in the above embodiments can be performed in a different order than that shown in the above embodiments and the desired result can still be achieved. In addition, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0066] Based on the same inventive concept, and corresponding to any of the above embodiments, this application also provides a sorting device for low-sulfur gold ore.
[0067] refer to Figure 3 The sorting device for the low-sulfur gold ore includes: The acquisition module 301 is configured to acquire the ore block to be sorted, perform spectral testing on the ore block to be sorted, and obtain spectral data. Extraction module 302 is configured to extract target band data from the spectral data and generate a spectral feature vector based on the target band data; The prediction module 303 is configured to input the spectral feature vector into a pre-trained sorting model and output the predicted gold grade of the ore block to be sorted through the sorting model. The determination module 304 is configured to determine the ore block to be sorted as gold ore in response to the predicted gold grade being greater than a preset gold grade threshold.
[0068] In some embodiments, the target band data includes the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity calculated based on the absorption depth of multiple target bands; the extraction module 302 is configured to construct the spectral feature vector based on the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands.
[0069] In some embodiments, before constructing the spectral feature vector based on the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands, the extraction module 302 is configured to obtain the globally optimal feature parameters of the sorting model; and to correct the absorption width, the maximum absorption band position, the absorption depth, and the illite crystallinity using the globally optimal feature parameters.
[0070] In some embodiments, a training module is further included, configured to construct a training dataset; construct a nonlinear regression function and constraints; convert the nonlinear regression function and constraints into an initial sorting model using the Lagrange multiplier method; and iteratively train the initial sorting model based on the training dataset to obtain the sorting model.
[0071] In some embodiments, a training module is further included, configured to: collect multiple ore block samples; perform spectral detection on the multiple ore block samples to obtain spectral data samples; and test the multiple ore block samples using the fire assay method to determine the gold grade label corresponding to each ore block sample; extract target band data from the spectral data samples and generate spectral feature vector samples based on the target band data; generate an initial training dataset based on the spectral feature vector samples and the gold grade label; and perform data filtering on the initial training dataset to obtain the training dataset.
[0072] In some embodiments, the training module is configured to calculate the Euclidean distance between each pair of spectral feature vector samples in the initial training dataset, select the two spectral feature vector samples corresponding to the maximum Euclidean distance and place them into a pre-constructed set; perform the following multi-round iterative operation on the remaining spectral feature vector samples in the initial training dataset: calculate the minimum Euclidean distance between each remaining spectral feature vector sample in the initial training dataset and all spectral feature vector samples in the set; place the spectral feature vector sample corresponding to the maximum value among all minimum Euclidean distances into the set; in response to determining that the number of spectral feature vector samples in the set is less than a preset threshold, proceed to the next round of iterative operation; in response to determining that the number of spectral feature vector samples in the set is equal to the preset threshold, exit the multi-round iterative operation to obtain the training dataset.
[0073] In some embodiments, the training module is configured to initialize a population, where each individual in the population corresponds to a set of feature parameters, and perform the following multi-round iterative operations: Based on the population, a new sterile line population is generated using a hybridization breeding algorithm; for each new sterile line individual in the new sterile line population, the training dataset is corrected according to the feature parameters of the new sterile line individual, and the initial sorting model is configured according to the feature parameters of the new sterile line individual; the configured initial sorting model is trained based on the corrected training dataset, and predictions are made on the corrected training dataset based on the trained initial sorting model to calculate the fitness of the new sterile line individual; the population is iteratively updated according to the fitness of the new sterile line individual, and in response to the failure to reach a preset iteration stop condition, the next round of iterative operation is entered; in response to the reaching of the preset iteration stop condition, the multi-round iterative operation is exited, and the feature parameters of the individual corresponding to the minimum fitness in this round are determined as the globally optimal feature parameters; the globally optimal feature parameters are substituted into the initial sorting model to obtain the sorting model.
[0074] In some embodiments, the training module is configured to generate a new population of sterile lines using the following formula: , in, Represents each individual in the population Corresponding feature parameters Includes penalty factors Kernel function parameters Absorption width parameter and absorption feature offset parameters ; express The first individual of the new sterile line generated by hybridization during the round breeding process Vigene, express Maintain the first in the line during rotational breeding The first individual Vigene, express The first sterile line in the rotational breeding The first individual Vigene, , The two numbers are random numbers in the range [0, 1] and are not equal; during random hybridization, During enantiomeric hybridization, ;like Its fitness value is better than Then replace To update the sterile line, or to retain the original individual.
[0075] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.
[0076] The apparatus described above is used to implement the sorting method for low-sulfur gold ore in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0077] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the sorting method for low-sulfur gold ore described in any of the above embodiments.
[0078] Figure 4 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.
[0079] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0080] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0081] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.
[0082] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0083] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0084] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.
[0085] The electronic devices described in the above embodiments are used to implement the sorting method for low-sulfur gold ore in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0086] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the sorting method for low-sulfur gold ore as described in any of the above embodiments.
[0087] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0088] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the sorting method for low-sulfur gold ore as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0089] Based on the same concept, corresponding to any of the above embodiments, this application also provides a computer program product, including computer program instructions, which, when run on a computer, cause the computer to perform the method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0090] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.
[0091] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to choose, based on the prompt message, whether to provide personal information to the software or hardware such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution.
[0092] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose whether to "agree" or "disagree" to the electronic device providing personal information.
[0093] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.
[0094] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in detail for the sake of brevity.
[0095] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0096] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.
[0097] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.
Claims
1. A method for separating low-sulfur gold ore, characterized in that, include: Obtain the ore block to be sorted, and perform spectral testing on the ore block to obtain spectral data; Extract the target band data from the spectral data, and generate a spectral feature vector based on the target band data; The spectral feature vector is input into a pre-trained sorting model, and the sorting model outputs the predicted gold grade of the ore block to be sorted. In response to the predicted gold grade being greater than a preset gold grade threshold, the ore block to be sorted is determined to be a gold ore.
2. The method according to claim 1, characterized in that, The target band data includes the maximum absorption band position, absorption depth, absorption width, symmetry, and illite crystallinity calculated based on the absorption depth of multiple target bands. The step of generating a spectral feature vector based on the target band data includes: The spectral feature vector is constructed based on the position of the maximum absorption band, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands.
3. The method according to claim 2, characterized in that, Before constructing the spectral feature vector based on the position of the maximum absorption band, absorption depth, absorption width, symmetry, and illite crystallinity of multiple target bands, the following steps are included: Obtain the globally optimal feature parameters of the sorting model; The absorption width, the position of the maximum absorption band, the absorption depth, and the illite crystallinity are corrected using the globally optimal characteristic parameters.
4. The method according to claim 1, characterized in that, The training methods for the sorting model include: Build the training dataset; Construct the nonlinear regression function and its constraints; The nonlinear regression function and constraints are converted into an initial sorting model using the Lagrange multiplier method. Based on the training dataset, the initial sorting model is iteratively trained to obtain the sorting model.
5. The method according to claim 4, characterized in that, The construction of the training dataset includes: Collect samples from multiple mineral blocks; Spectroscopic detection was performed on the multiple ore block samples to obtain spectral data samples, and the multiple ore block samples were tested using the fire assay method to determine the gold grade label corresponding to each ore block sample. Extract the target band data from the spectral data sample, and generate a spectral feature vector sample based on the target band data; An initial training dataset is generated based on the spectral feature vector samples and the gold grade labels; The initial training dataset is filtered to obtain the training dataset.
6. The method according to claim 5, characterized in that, The step of filtering the initial training dataset to obtain the training dataset includes: Calculate the Euclidean distance between each pair of spectral feature vector samples in the initial training dataset, and select the two spectral feature vector samples corresponding to the maximum Euclidean distance to put into a pre-constructed set; Perform the following multi-round iterative operation on the remaining spectral feature vector samples in the initial training dataset: Calculate the minimum Euclidean distance between each remaining spectral feature vector sample in the initial training dataset and all spectral feature vector samples in the set; The spectral feature vector samples corresponding to the maximum value of all minimum Euclidean distances are placed into the set. In response to determining that the number of spectral feature vector samples in the set is less than a preset threshold, the next round of iteration is initiated. In response to determining that the number of spectral feature vector samples in the set is equal to the preset threshold, the multi-round iteration is terminated, and the training dataset is obtained.
7. The method according to claim 4, characterized in that, The step of iteratively training the initial sorting model based on the training dataset to obtain the sorting model includes: Initialize the population, where each individual corresponds to a set of feature parameters, and perform the following multi-round iterative operation: Based on the population, a new sterile line population is generated using a hybridization breeding algorithm; For each new sterile line individual in the new sterile line population, the training dataset is corrected according to the characteristic parameters of the new sterile line individual, and the initial sorting model is configured according to the characteristic parameters of the new sterile line individual; The configured initial sorting model is trained based on the calibrated training dataset. The fitness of the new sterile line individuals is calculated based on the trained initial sorting model on the calibrated training dataset. The population is iteratively updated based on the fitness of the new sterile line individuals. If the preset iteration stop condition is not met, the next iteration operation is entered. If the preset iteration stop condition is met, the multi-round iteration operation is exited, and the characteristic parameters of the individual with the minimum fitness in this round are determined as the global optimal characteristic parameters. The global optimal feature parameters are substituted into the initial sorting model to obtain the sorting model.
8. The method according to claim 7, characterized in that, The process of generating a new sterile line population based on the aforementioned population using a hybridization breeding algorithm includes: The new sterile line population is generated using the following formula: ， in, Represents each individual in the population Corresponding feature parameters Includes penalty factors Kernel function parameters Absorption width parameter and absorption feature offset parameters ; express The first individual of the new sterile line generated by hybridization during the round breeding process Vigene, express Maintain the first in the line during rotational breeding The first individual Vigene, express The first sterile line in the rotational breeding The first individual Vigene, , The two numbers are random numbers in the range [0, 1] and are not equal; during random hybridization, During enantiomeric hybridization, ;like Its fitness value is better than Then replace To update the sterile line, or to retain the original individual.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 8.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 8.