A method and system for detecting tungsten content based on a gradient algorithm

Abstract

The application provides a tungsten content detection method and system based on a gradient algorithm, which comprises the following steps: obtaining various different types of tungsten ore samples, and detecting the spectral characteristic data, microscopic structure image data and element composition detection data corresponding to the tungsten ore samples in real time through a preset device; integrating the data to generate sample characteristic data, and creating a corresponding dynamic gradient correction model in real time according to the sample characteristic data; dynamically dividing a plurality of content gradient intervals according to the distribution density of the sample characteristic data through the dynamic gradient correction model and a preset gradient adjustment algorithm; obtaining actual sample characteristic data of an actual tungsten ore, and outputting a corresponding target content gradient interval through the dynamic gradient correction model according to the actual sample characteristic data, so as to extract the target content in the target content gradient interval. The application can accurately detect the tungsten content in the ore, and correspondingly improves the detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of ore detection technology, and in particular to a method and system for detecting tungsten content based on a gradient algorithm. Background Technology

[0002] With the advancement of technology and the rapid development of the times, people have been able to mine all kinds of ores. In order to facilitate subsequent purification, it is also necessary to detect the content of effective minerals in the ores.

[0003] In order to objectively detect the mineral content in an ore, people first prepare a corresponding ore solution based on the ore and prepare multiple standard color cards in advance. Specifically, the color gradient of the standard color cards prepared by existing technology is usually a preset fixed range. For example, each gradient can correspond to a content difference of 0.1% or 0.5%, and then a subsequent comparison is carried out.

[0004] Furthermore, since each standard color chart can only correspond to a fixed range, in practical applications, for tungsten ore with low or high content, the gradient interval may be too large, resulting in the inability to accurately reflect the actual content of the ore. For example, the actual content may be 0.3%, but the color chart may only indicate 0.2% and 0.4%, thereby reducing the accuracy of tungsten ore detection. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a tungsten content detection method and system based on gradient algorithm, so as to solve the problem that the existing technology may fail to accurately reflect the actual content of ore due to excessively large gradient intervals.

[0006] The first aspect of the present invention proposes: A tungsten content detection method based on a gradient algorithm, wherein the method includes: Various types of tungsten ore samples are obtained, and the spectral feature data, microstructure image data, and elemental composition detection data corresponding to the tungsten ore samples are detected in real time using a preset device. The spectral feature data, the microstructure image data, and the elemental composition detection data are integrated and processed in real time to generate corresponding sample feature data in real time, and a corresponding dynamic gradient correction model is created in real time based on the sample feature data. The dynamic gradient correction model and the preset gradient adjustment algorithm dynamically divide the sample feature data into several content gradient intervals based on the distribution density of the sample feature data. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced; when the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. The actual sample feature data of the actual tungsten ore is obtained, and the target content gradient interval is output according to the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval.

[0007] The beneficial effects of this invention are as follows: By acquiring different types of tungsten ore samples in real time, it is possible to obtain the correlation between the spectral characteristic data, microstructure data, elemental composition detection data, and tungsten content of the tungsten ore. Based on this, the invention can integrate the corresponding sample characteristic data in real time and create a dynamic gradient correction model for dividing the content gradient interval in real time. Based on this, it can detect the target content gradient interval corresponding to the actual tungsten ore in real time and extract the corresponding target content, thereby avoiding the problem of not being able to detect the true content and improving the detection efficiency of tungsten content.

[0008] Furthermore, the step of real-time integration and processing of the spectral feature data, the microstructure image data, and the elemental composition detection data to generate corresponding sample feature data in real time includes: The spectral feature data is standardized and preprocessed to remove background noise and extract the feature peak positions, intensities, and half-widths (WHMs) contained in the spectral feature data. The feature peak positions, intensities, and WHMs are then converted into spectral feature vectors with fixed dimensions in real time. The microstructure image data is segmented and feature extracted. The distribution size, distribution density and porosity of the particles are obtained through edge detection and morphological analysis to generate an image feature matrix. The elemental composition detection data is normalized to convert the content ratio of each element into an elemental feature array. The spectral feature vector, the image feature matrix, and the elemental feature array are then matched and fused to form sample feature data with a unified dimension.

[0009] Furthermore, the step of performing dimensionality matching and fusion on the spectral feature vector, the image feature matrix, and the element feature array to form the sample feature data with a unified dimension includes: The spectral feature vector is reduced in dimensionality by using principal component analysis algorithm to retain principal component components whose cumulative contribution rate exceeds a preset contribution rate threshold, and the dimensionality-reduced spectral feature sub-vectors are obtained. A feature selection algorithm is used to filter out a preset number of feature parameters that are most correlated with tungsten content from the image feature matrix, and the feature parameters are converted into image feature sub-vectors with the same dimension as the spectral feature sub-vectors; The element feature array is subjected to dimensional expansion processing to supplement the feature dimension to the same dimension as the spectral feature sub-vector through an interpolation algorithm, so as to form a corresponding element feature sub-vector. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are then weighted and fused to generate the sample feature data with a unified dimension.

[0010] Furthermore, the step of weightedly fusing the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector to generate the sample feature data with a unified dimension includes: The global weight values ​​corresponding to the vector parameters in the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated using the random forest algorithm. Based on the global weight value, the weighted dimensions corresponding to the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are determined in real time. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated element-wise based on the weighted dimension to generate sample feature data with a unified dimension.

[0011] Furthermore, the step of dynamically dividing several content gradient intervals based on the distribution density of the sample feature data using the dynamic gradient correction model and the preset gradient adjustment algorithm includes: The distribution density of the sample feature data on the preset tungsten content numerical axis is calculated by a preset kernel density estimation algorithm to generate a continuous density distribution curve, and the density peak point and density valley point are marked in real time on the density distribution curve. Using the density valley point as the initial interval division point, the tungsten content numerical axis is initially divided into several candidate intervals, and each candidate interval contains at least one density peak point; The candidate intervals are dynamically adjusted to generate the corresponding content gradient intervals.

[0012] Furthermore, the step of dynamically adjusting the candidate intervals to generate the corresponding content gradient intervals includes: The mean distribution density of sample feature data within each candidate interval is calculated, and the mean distribution density of each candidate interval is compared with a preset high density threshold and a low density threshold to determine the density level corresponding to each candidate interval in real time. Based on the density level, the interval level corresponding to each candidate interval is determined in real time, and each interval level is unique; The corresponding extreme values ​​of content are set in real time according to the interval level, and each extreme value of content is divided to generate several content gradient intervals.

[0013] Furthermore, the step of comparing the mean distribution density of each candidate interval with preset high-density thresholds and low-density thresholds to determine the density level corresponding to each candidate interval in real time includes: Three-level density determination criteria are set, wherein the first determination criterion is that the average distribution density is greater than or equal to the preset high density threshold, the second determination criterion is that the average distribution density is between the preset low density threshold and the preset high density threshold, and the third determination criterion is that the average distribution density is less than or equal to the preset low density threshold. The mean distribution density of each candidate interval is substituted into the three-level judgment criteria for matching. If it meets the first judgment criterion, it is marked as high density level; if it meets the second judgment criterion, it is marked as medium density level; and if it meets the third judgment criterion, it is marked as low density level.

[0014] The second aspect of the present invention proposes: A tungsten content detection system based on a gradient algorithm, wherein the system comprises: The detection module is used to acquire various types of tungsten ore samples and to detect in real time the spectral feature data, microstructure image data and elemental composition detection data corresponding to the tungsten ore samples through a preset device. A creation module is used to integrate and process the spectral feature data, the microstructure image data, and the elemental composition detection data in real time to generate corresponding sample feature data in real time, and to create a corresponding dynamic gradient correction model in real time based on the sample feature data. The adjustment module is used to dynamically divide several content gradient intervals according to the distribution density of the sample feature data through the dynamic gradient correction model and the preset gradient adjustment algorithm. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced; when the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. The output module is used to acquire actual sample feature data of actual tungsten ore, and output the corresponding target content gradient interval based on the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval.

[0015] Furthermore, the creation module is specifically used for: The spectral feature data is standardized and preprocessed to remove background noise and extract the feature peak positions, intensities, and half-widths (WHMs) contained in the spectral feature data. The feature peak positions, intensities, and WHMs are then converted into spectral feature vectors with fixed dimensions in real time. The microstructure image data is segmented and feature extracted. The distribution size, distribution density and porosity of the particles are obtained through edge detection and morphological analysis to generate an image feature matrix. The elemental composition detection data is normalized to convert the content ratio of each element into an elemental feature array. The spectral feature vector, the image feature matrix, and the elemental feature array are then matched and fused to form sample feature data with a unified dimension.

[0016] Furthermore, the creation module is specifically used for: The spectral feature vector is reduced in dimensionality by using principal component analysis algorithm to retain principal component components whose cumulative contribution rate exceeds a preset contribution rate threshold, and the dimensionality-reduced spectral feature sub-vectors are obtained. A feature selection algorithm is used to filter out a preset number of feature parameters that are most correlated with tungsten content from the image feature matrix, and the feature parameters are converted into image feature sub-vectors with the same dimension as the spectral feature sub-vectors; The element feature array is subjected to dimensional expansion processing to supplement the feature dimension to the same dimension as the spectral feature sub-vector through an interpolation algorithm, so as to form a corresponding element feature sub-vector. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are then weighted and fused to generate the sample feature data with a unified dimension.

[0017] Furthermore, the creation module is specifically used for: The global weight values ​​corresponding to the vector parameters in the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated using the random forest algorithm. Based on the global weight value, the weighted dimensions corresponding to the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are determined in real time. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated element-wise based on the weighted dimension to generate sample feature data with a unified dimension.

[0018] Furthermore, the adjustment module is specifically used for: The distribution density of the sample feature data on the preset tungsten content numerical axis is calculated by a preset kernel density estimation algorithm to generate a continuous density distribution curve, and the density peak point and density valley point are marked in real time on the density distribution curve. Using the density valley point as the initial interval division point, the tungsten content numerical axis is initially divided into several candidate intervals, and each candidate interval contains at least one density peak point; The candidate intervals are dynamically adjusted to generate the corresponding content gradient intervals.

[0019] Furthermore, the adjustment module is specifically used for: The mean distribution density of sample feature data within each candidate interval is calculated, and the mean distribution density of each candidate interval is compared with a preset high density threshold and a low density threshold to determine the density level corresponding to each candidate interval in real time. Based on the density level, the interval level corresponding to each candidate interval is determined in real time, and each interval level is unique; The corresponding extreme values ​​of content are set in real time according to the interval level, and each extreme value of content is divided to generate several content gradient intervals.

[0020] Furthermore, the adjustment module is specifically used for: Three-level density determination criteria are set, wherein the first determination criterion is that the average distribution density is greater than or equal to the preset high density threshold, the second determination criterion is that the average distribution density is between the preset low density threshold and the preset high density threshold, and the third determination criterion is that the average distribution density is less than or equal to the preset low density threshold. The mean distribution density of each candidate interval is substituted into the three-level judgment criteria for matching. If it meets the first judgment criterion, it is marked as high density level; if it meets the second judgment criterion, it is marked as medium density level; and if it meets the third judgment criterion, it is marked as low density level.

[0021] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the tungsten content detection method based on the gradient algorithm as described above.

[0022] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the tungsten content detection method based on the gradient algorithm as described above.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1 A flowchart of the tungsten content detection method based on gradient algorithm provided in the first embodiment of the present invention; Figure 2 The diagram shows the structure of the tungsten content detection system based on the gradient algorithm provided in the third embodiment of the present invention.

[0025] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0026] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0027] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Please see Figure 1 The figure shows a tungsten content detection method based on gradient algorithm provided in the first embodiment of the present invention. The tungsten content detection method based on gradient algorithm provided in this embodiment can accurately detect the tungsten content inside each tungsten ore, thereby improving the detection efficiency.

[0030] Specifically, this embodiment provides: A tungsten content detection method based on a gradient algorithm specifically includes the following steps: Step S10: Obtain various types of tungsten ore samples, and use a preset device to detect in real time the spectral feature data, microstructure image data and elemental composition detection data corresponding to the tungsten ore samples; It should be noted that, in order to accurately detect the tungsten content in various types of tungsten ore, a large number of ore samples must first be obtained. Based on this, subsequent model training is carried out. Specifically, for ease of implementation, this invention will first obtain ore samples covering different origins, compositions, and grades to ensure sample diversity. Based on this, the spectral responses (such as characteristic absorption peaks and emission peaks) reflecting tungsten and associated elements in the ore are obtained using spectrometers and other equipment, thereby generating the required spectral feature data. Similarly, the microscopic features of the ore, such as crystal structure and particle morphology, are obtained using microscopes and electron microscopes, thereby generating the required microscopic structure image data. Correspondingly, the content ratio of tungsten and other elements is obtained using X-ray fluorescence spectroscopy (XRF) and inductively coupled plasma mass spectrometry (ICP-MS), thereby generating the required elemental composition detection data for subsequent processing.

[0031] Step S20: The spectral feature data, the microstructure image data, and the elemental composition detection data are integrated and processed in real time to generate corresponding sample feature data in real time, and a corresponding dynamic gradient correction model is created in real time based on the sample feature data. It should be noted that after obtaining the required spectral feature data, microstructure image data, and elemental composition detection data through the above steps, in order to facilitate subsequent model training and shorten data processing time, this invention will, specifically, standardize and fuse the three types of data sequentially according to pre-set rules, thereby generating sample feature data with the same format. Based on this, to facilitate subsequent model training, this invention will also invoke an existing CNN neural network. Specifically, the current sample feature data will be divided into corresponding validation, training, and test sets. At the same time, a loss function adapted to the current sample feature data will be invoked, thereby ultimately completing the model training of the current sample feature data and generating the required dynamic gradient correction model for subsequent processing.

[0032] Step S30: The dynamic gradient correction model and the preset gradient adjustment algorithm are used to dynamically divide several content gradient intervals according to the distribution density of the sample feature data. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced. When the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. It should be noted that after obtaining the required dynamic gradient correction model in real time through the above steps, the relationship between the current sample feature data and tungsten content can be determined in real time through this dynamic gradient correction model. Specifically, this invention will detect the distribution area of ​​each sample in the current sample feature data in real time. When a densely distributed area is detected, it indicates that there are many samples within a certain tungsten content range and the feature differences are small. Therefore, the gradient interval will be reduced accordingly to improve detection accuracy. For example, 1%-2% can be subdivided into 1%-1.5% and 1.5%-2%. Similarly, when a sparsely distributed area is detected, it indicates that there are few samples with a certain tungsten content and the feature differences are large. In this case, the gradient interval needs to be expanded accordingly to reduce invalid subdivisions. For example, 5%-10% can be merged into one interval for subsequent processing.

[0033] Step S40: Obtain actual sample feature data of actual tungsten ore, and output the corresponding target content gradient interval based on the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval.

[0034] It should be noted that after obtaining the required dynamic gradient correction model and several content gradient intervals in real time through the above steps, actual ore detection can then be performed. It should be pointed out that since the dynamic gradient correction model provided by this invention is trained based on generated feature data, it is necessary to obtain the actual sample feature data of the current tungsten ore as input to the model. Specifically, the data type contained in the actual sample feature data is consistent with the data type contained in the aforementioned sample feature data. Based on this, this invention can use the dynamic gradient correction model to match the target content gradient interval corresponding to the current actual tungsten ore in real time, and finally output the tungsten content inside the current actual tungsten ore according to the target content gradient interval. Specifically, for ease of understanding, for example, if the target interval is 1.2%-1.8%, the average feature output of the samples within the interval is 1.5%, thereby objectively and accurately identifying the tungsten content inside each tungsten ore, thus improving detection efficiency.

[0035] Second Embodiment Furthermore, the step of real-time integration and processing of the spectral feature data, the microstructure image data, and the elemental composition detection data to generate corresponding sample feature data in real time includes: The spectral feature data is standardized and preprocessed to remove background noise and extract the feature peak positions, intensities, and half-widths (WHMs) contained in the spectral feature data. The feature peak positions, intensities, and WHMs are then converted into spectral feature vectors with fixed dimensions in real time. The microstructure image data is segmented and feature extracted. The distribution size, distribution density and porosity of the particles are obtained through edge detection and morphological analysis to generate an image feature matrix. The elemental composition detection data is normalized to convert the content ratio of each element into an elemental feature array. The spectral feature vector, the image feature matrix, and the elemental feature array are then matched and fused to form sample feature data with a unified dimension.

[0036] It should be noted that after obtaining the required spectral feature data, microstructure image data, and elemental composition detection data through the above steps, in order to objectively and effectively standardize these three types of data, this invention first purifies the spectral feature data through smoothing and baseline correction. Simultaneously, it extracts the characteristic peak parameters strongly correlated with tungsten content from the current spectral feature data. These characteristic peak parameters specifically include position (reflecting element type, such as the characteristic peak wavelength of tungsten), intensity (positively correlated with content), and full width at half maximum (FWHM) (reflecting crystal structure integrity). Based on this, to facilitate the quantification of these three parameters, this invention can also convert the current characteristic peak parameters into a spectral feature vector with a fixed dimension in real time using the existing DTW algorithm. Similarly, after obtaining the microstructure image data, threshold segmentation and edge detection (such as the Canny algorithm) are used to separate the effective region (such as tungsten mineral particles) from the background in the ore. Based on this, morphological analysis (such as corrosion and dilation calculations) is used to obtain the particle size distribution (maximum / mineral density). The minimum particle size, distribution density (number of particles per unit area), and porosity (percentage of voids) are used to generate image features in matrix form (e.g., a 10×10 feature matrix). Correspondingly, after obtaining the elemental composition detection data, the content percentage of each element (such as tungsten, iron, and silicon) is converted into a value between 0 and 1 (eliminating absolute differences), forming an elemental feature array (e.g., [0.2, 0.15, 0.08,...]). Based on this, the present invention can ultimately perform dimensional matching and fusion processing on the spectral feature vector, image feature matrix, and elemental feature array, thereby forming sample feature data with a unified dimension, and then completing the subsequent model construction for subsequent processing.

[0037] Furthermore, the step of performing dimensional matching and fusion of the spectral feature vector, the image feature matrix, and the element feature array to form the sample feature data with a unified dimension includes: The spectral feature vector is reduced in dimensionality by using principal component analysis algorithm to retain principal component components whose cumulative contribution rate exceeds a preset contribution rate threshold, and the dimensionality-reduced spectral feature sub-vectors are obtained. A feature selection algorithm is used to filter out a preset number of feature parameters that are most correlated with tungsten content from the image feature matrix, and the feature parameters are converted into image feature sub-vectors with the same dimension as the spectral feature sub-vectors; The element feature array is subjected to dimensional expansion processing to supplement the feature dimension to the same dimension as the spectral feature sub-vector through an interpolation algorithm, so as to form a corresponding element feature sub-vector. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are then weighted and fused to generate the sample feature data with a unified dimension.

[0038] It should be noted that after obtaining the required spectral feature vector, image feature matrix, and element feature array through the above steps, the final matching and fusion will then be performed. Specifically, this invention will convert the high-dimensional spectral vector (e.g., 30-dimensional) into low-dimensional principal components (e.g., 10-dimensional) through linear transformation, retaining principal components with a cumulative contribution rate ≥ a preset threshold (e.g., 85%) (i.e., retaining most of the original information), thereby removing redundant information (e.g., highly correlated feature peaks) in the spectral features and reducing the amount of computation. Similarly, this invention will select the N features with the highest correlation to tungsten content from the image matrix (e.g., 100-dimensional) (e.g., the top 10 selected based on Pearson coefficient calculation). Based on this, the selected features will be converted to the same dimension as the spectral sub-vector (e.g., 10-dimensional) to form the image feature sub-vector (ensuring compatibility with spectral features), thus forming the aforementioned image feature sub-vector. Correspondingly, if the element array dimension is low (e.g., 10-dimensional), features will be supplemented to the target dimension (e.g., 10-dimensional) through an interpolation algorithm (e.g., linear interpolation). (The dimension is consistent with the spectral sub-vector), forming an element feature sub-vector. Based on this, the above spectral feature sub-vector, image feature sub-vector, and element feature sub-vector are immediately weighted and fused to generate the above sample feature data for subsequent processing.

[0039] Furthermore, the step of weightedly fusing the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector to generate the sample feature data with a unified dimension includes: The global weight values ​​corresponding to the vector parameters in the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated using the random forest algorithm. Based on the global weight value, the weighted dimensions corresponding to the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are determined in real time. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated element-wise based on the weighted dimension to generate sample feature data with a unified dimension.

[0040] It should be noted that in the actual weighted fusion process, this invention constructs multiple decision trees to calculate the importance of each feature parameter (such as the characteristic peak intensity of the spectrum, the particle density of the image) to the tungsten content prediction result (i.e., global weight, ranging from 0 to 1). This allows for the quantification of the contribution of different features. Based on this, according to the global weight value, dimensional weights are assigned to the three types of sub-vectors: spectrum, image, and element (for example, the dimension corresponding to a feature with a higher weight has a higher proportion during fusion). Specifically, for ease of understanding, if the weight of a certain dimension in the spectrum sub-vector is 0.8 and that of the image is 0.5, then the influence of that dimension of the spectrum is greater during fusion. Based on this, each dimension of the three types of sub-vectors (such as the 1st to 10th dimensions in a 10-dimensional array) is finally superimposed according to the weighted dimension (e.g., spectrum value × 0.4 + image value × 0.3 + element value × 0.3) to generate the final sample feature data (ensuring that the fused data has both unified dimensions and retains the differences in the importance of each feature) for subsequent processing.

[0041] Furthermore, the step of dynamically dividing several content gradient intervals based on the distribution density of the sample feature data using the dynamic gradient correction model and the preset gradient adjustment algorithm includes: The distribution density of the sample feature data on the preset tungsten content numerical axis is calculated by a preset kernel density estimation algorithm to generate a continuous density distribution curve, and the density peak point and density valley point are marked in real time on the density distribution curve. Using the density valley point as the initial interval division point, the tungsten content numerical axis is initially divided into several candidate intervals, and each candidate interval contains at least one density peak point; The candidate intervals are dynamically adjusted to generate the corresponding content gradient intervals.

[0042] It should be noted that after creating the required dynamic gradient correction model in real time through the above steps, a suitable gradient adjustment algorithm will be retrieved from the preset database. Specifically, to facilitate interval division, a numerical axis corresponding to tungsten content will be pre-created. Based on this, the distribution density corresponding to each tungsten content will be calculated in real time on this numerical axis, and each current distribution density will be mapped to the pre-created two-dimensional coordinate system in real time, thereby forming a continuous density distribution curve in the current two-dimensional coordinate system. Based on this, it is also necessary to detect the density peak points and density valley points corresponding to the current density distribution curve in real time. In the actual division process, each current valley point is used as a dividing point, thereby dividing the current complete tungsten content numerical axis into several candidate intervals. Based on this, it is also necessary to make adaptive adjustments according to the sample data distribution of each tungsten content to facilitate subsequent processing.

[0043] Furthermore, the step of dynamically adjusting the candidate intervals to generate the corresponding content gradient intervals includes: The mean distribution density of sample feature data within each candidate interval is calculated, and the mean distribution density of each candidate interval is compared with a preset high density threshold and a low density threshold to determine the density level corresponding to each candidate interval in real time. Based on the density level, the interval level corresponding to each candidate interval is determined in real time, and each interval level is unique; The corresponding extreme values ​​of content are set in real time according to the interval level, and each extreme value of content is divided to generate several content gradient intervals.

[0044] It should be noted that after obtaining several candidate intervals in real time through the above steps, since each candidate interval corresponds to a series of sample feature data, this invention first calculates the mean distribution density of the sample feature data in each candidate interval. Specifically, this mean distribution density reflects the overall density of each interval in real time. Therefore, to facilitate subsequent division, this invention compares the mean distribution density of each candidate interval with preset high-density thresholds (e.g., 0.8) and low-density thresholds (e.g., 0.3) to determine whether the interval is high-density (mean ≥ 0.8), medium-density (0.3-0.8), or low-density (≤ 0.3). Specifically, a unique interval level is assigned to each density level (e.g., high density corresponds to level 1, medium density to level 2, and low density to level 3). The level determines the interval spacing (level 1 has the smallest spacing, level 3 has the largest spacing). Based on this, the appropriate extreme content value is set in real time according to the interval level of each candidate interval. Specifically, for ease of understanding, for example, upper and lower limits are set for each interval based on the interval level (e.g., the upper limit of level 1 = lower limit + ...). 0.5%, Level 3 interval upper limit = lower limit + 2%), based on this, the subsequent segmentation is completed, so that the above-mentioned content gradient intervals can be generated for subsequent processing.

[0045] Furthermore, the step of comparing the mean distribution density of each candidate interval with preset high-density thresholds and low-density thresholds to determine the density level corresponding to each candidate interval in real time includes: Three-level density determination criteria are set, wherein the first determination criterion is that the average distribution density is greater than or equal to the preset high density threshold, the second determination criterion is that the average distribution density is between the preset low density threshold and the preset high density threshold, and the third determination criterion is that the average distribution density is less than or equal to the preset low density threshold. The mean distribution density of each candidate interval is substituted into the three-level judgment criteria for matching. If it meets the first judgment criterion, it is marked as high density level; if it meets the second judgment criterion, it is marked as medium density level; and if it meets the third judgment criterion, it is marked as low density level.

[0046] It should be noted that, in order to objectively and accurately determine the density level of each candidate interval, this invention introduces a three-level density judgment standard. Specifically, the first standard is: if the mean distribution density is ≥ a preset high density threshold (e.g., 0.8), it corresponds to a high density level (the samples are highly concentrated, and intervals need to be subdivided). The second standard is: if the mean is between the low density threshold (e.g., 0.3) and the high density threshold, it corresponds to a medium density level (the samples are moderately concentrated, and the interval intervals are moderate). The third standard is: if the mean is ≤ a low density threshold, it corresponds to a low density level (the samples are dispersed, and the interval intervals can be expanded). Finally, the mean of each candidate interval is substituted into the above standards to match the corresponding level (e.g., mean 0.9 → high density, 0.5 → medium density, 0.2 → low density), providing a clear basis for subsequent interval adjustments (ensuring the objectivity and consistency of the level judgment), thereby objectively and accurately determining the density level of each candidate interval for subsequent processing.

[0047] Please see Figure 2 The third embodiment of the present invention provides; A tungsten content detection system based on a gradient algorithm, wherein the system comprises: The detection module is used to acquire various types of tungsten ore samples and to detect in real time the spectral feature data, microstructure image data and elemental composition detection data corresponding to the tungsten ore samples through a preset device. A creation module is used to integrate and process the spectral feature data, the microstructure image data, and the elemental composition detection data in real time to generate corresponding sample feature data in real time, and to create a corresponding dynamic gradient correction model in real time based on the sample feature data. The adjustment module is used to dynamically divide several content gradient intervals according to the distribution density of the sample feature data through the dynamic gradient correction model and the preset gradient adjustment algorithm. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced; when the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. The output module is used to acquire actual sample feature data of actual tungsten ore, and output the corresponding target content gradient interval based on the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval.

[0048] Furthermore, the creation module is specifically used for: The spectral feature data is standardized and preprocessed to remove background noise and extract the feature peak positions, intensities, and half-widths (WHMs) contained in the spectral feature data. The feature peak positions, intensities, and WHMs are then converted into spectral feature vectors with fixed dimensions in real time. The microstructure image data is segmented and feature extracted. The distribution size, distribution density and porosity of the particles are obtained through edge detection and morphological analysis to generate an image feature matrix. The elemental composition detection data is normalized to convert the content ratio of each element into an elemental feature array. The spectral feature vector, the image feature matrix, and the elemental feature array are then matched and fused to form sample feature data with a unified dimension.

[0049] Furthermore, the creation module is specifically used for: The spectral feature vector is reduced in dimensionality by using principal component analysis algorithm to retain principal component components whose cumulative contribution rate exceeds a preset contribution rate threshold, and the dimensionality-reduced spectral feature sub-vectors are obtained. A feature selection algorithm is used to filter out a preset number of feature parameters that are most correlated with tungsten content from the image feature matrix, and the feature parameters are converted into image feature sub-vectors with the same dimension as the spectral feature sub-vectors; The element feature array is subjected to dimensional expansion processing to supplement the feature dimension to the same dimension as the spectral feature sub-vector through an interpolation algorithm, so as to form a corresponding element feature sub-vector. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are then weighted and fused to generate the sample feature data with a unified dimension.

[0050] Furthermore, the creation module is specifically used for: The global weight values ​​corresponding to the vector parameters in the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated using the random forest algorithm. Based on the global weight value, the weighted dimensions corresponding to the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are determined in real time. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated element-wise based on the weighted dimension to generate sample feature data with a unified dimension.

[0051] Furthermore, the adjustment module is specifically used for: The distribution density of the sample feature data on the preset tungsten content numerical axis is calculated by a preset kernel density estimation algorithm to generate a continuous density distribution curve, and the density peak point and density valley point are marked in real time on the density distribution curve. Using the density valley point as the initial interval division point, the tungsten content numerical axis is initially divided into several candidate intervals, and each candidate interval contains at least one density peak point; The candidate intervals are dynamically adjusted to generate the corresponding content gradient intervals.

[0052] Furthermore, the adjustment module is specifically used for: The mean distribution density of sample feature data within each candidate interval is calculated, and the mean distribution density of each candidate interval is compared with a preset high density threshold and a low density threshold to determine the density level corresponding to each candidate interval in real time. Based on the density level, the interval level corresponding to each candidate interval is determined in real time, and each interval level is unique; The corresponding extreme values ​​of content are set in real time according to the interval level, and each extreme value of content is divided to generate several content gradient intervals.

[0053] Furthermore, the adjustment module is specifically used for: Three-level density determination criteria are set, wherein the first determination criterion is that the average distribution density is greater than or equal to the preset high density threshold, the second determination criterion is that the average distribution density is between the preset low density threshold and the preset high density threshold, and the third determination criterion is that the average distribution density is less than or equal to the preset low density threshold. The mean distribution density of each candidate interval is substituted into the three-level judgment criteria for matching. If it meets the first judgment criterion, it is marked as high density level; if it meets the second judgment criterion, it is marked as medium density level; and if it meets the third judgment criterion, it is marked as low density level.

[0054] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tungsten content detection method based on the gradient algorithm as described above.

[0055] The fifth embodiment of the present invention provides a readable storage medium storing a computer program thereon, wherein the program, when executed by a processor, implements the tungsten content detection method based on the gradient algorithm as described above.

[0056] In summary, the tungsten content detection method and system based on gradient algorithm provided in the above embodiments of the present invention can accurately detect the tungsten content in tungsten ore, thereby improving the detection efficiency.

[0057] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0058] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0059] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0060] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0061] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0062] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A tungsten content detection method based on gradient algorithm, characterized in that, The method includes: Various types of tungsten ore samples are obtained, and the spectral feature data, microstructure image data, and elemental composition detection data corresponding to the tungsten ore samples are detected in real time using a preset device. The spectral feature data, the microstructure image data, and the elemental composition detection data are integrated and processed in real time to generate corresponding sample feature data in real time, and a corresponding dynamic gradient correction model is created in real time based on the sample feature data. The dynamic gradient correction model and the preset gradient adjustment algorithm dynamically divide the sample feature data into several content gradient intervals based on the distribution density of the sample feature data. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced; when the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. The actual sample feature data of actual tungsten ore is obtained, and the target content gradient interval is output according to the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval; The step of dynamically dividing several content gradient intervals based on the distribution density of the sample feature data using the dynamic gradient correction model and the preset gradient adjustment algorithm includes: The distribution density of the sample feature data on the preset tungsten content numerical axis is calculated by a preset kernel density estimation algorithm to generate a continuous density distribution curve, and the density peak point and density valley point are marked in real time on the density distribution curve. Using the density valley point as the initial interval division point, the tungsten content numerical axis is initially divided into several candidate intervals, and each candidate interval contains at least one density peak point; The candidate intervals are dynamically adjusted to generate the corresponding content gradient intervals. The step of dynamically adjusting the candidate intervals to generate the corresponding content gradient intervals includes: The mean distribution density of sample feature data within each candidate interval is calculated, and the mean distribution density of each candidate interval is compared with a preset high density threshold and a low density threshold to determine the density level corresponding to each candidate interval in real time. Based on the density level, the interval level corresponding to each candidate interval is determined in real time, and each interval level is unique; The corresponding extreme values ​​of content are set in real time according to the interval level, and each extreme value of content is divided to generate several content gradient intervals.

2. The tungsten content detection method based on gradient algorithm according to claim 1, characterized in that: The step of real-time integration and processing of the spectral feature data, the microstructure image data, and the elemental composition detection data to generate corresponding sample feature data includes: The spectral feature data is standardized and preprocessed to remove background noise and extract the feature peak positions, intensities, and half-widths (WHMs) contained in the spectral feature data. The feature peak positions, intensities, and WHMs are then converted into spectral feature vectors with fixed dimensions in real time. The microstructure image data is segmented and feature extracted. The distribution size, distribution density and porosity of the particles are obtained through edge detection and morphological analysis to generate an image feature matrix. The elemental composition detection data is normalized to convert the content ratio of each element into an elemental feature array. The spectral feature vector, the image feature matrix, and the elemental feature array are then matched and fused to form sample feature data with a unified dimension.

3. The tungsten content detection method based on gradient algorithm according to claim 2, characterized in that: The step of performing dimensionality matching and fusion of the spectral feature vector, the image feature matrix, and the element feature array to form the sample feature data with a unified dimension includes: The spectral feature vector is reduced in dimensionality by using principal component analysis algorithm to retain principal component components whose cumulative contribution rate exceeds a preset contribution rate threshold, and the dimensionality-reduced spectral feature sub-vectors are obtained. A feature selection algorithm is used to filter out a preset number of feature parameters that are most correlated with tungsten content from the image feature matrix, and the feature parameters are converted into image feature sub-vectors with the same dimension as the spectral feature sub-vectors; The element feature array is subjected to dimensional expansion processing to supplement the feature dimension to the same dimension as the spectral feature sub-vector through an interpolation algorithm, so as to form a corresponding element feature sub-vector. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are then weighted and fused to generate the sample feature data with a unified dimension.

4. The tungsten content detection method based on gradient algorithm according to claim 3, characterized in that: The step of weightedly fusing the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector to generate the sample feature data with a unified dimension includes: The global weight values ​​corresponding to the vector parameters in the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated using the random forest algorithm. Based on the global weight value, the weighted dimensions corresponding to the spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are determined in real time. The spectral feature sub-vector, the image feature sub-vector, and the element feature sub-vector are calculated element-wise based on the weighted dimension to generate sample feature data with a unified dimension.

5. The tungsten content detection method based on gradient algorithm according to claim 1, characterized in that: The step of comparing the mean distribution density of each candidate interval with preset high-density thresholds and low-density thresholds to determine the density level corresponding to each candidate interval in real time includes: Three-level density determination criteria are set, wherein the first determination criterion is that the average distribution density is greater than or equal to the preset high density threshold, the second determination criterion is that the average distribution density is between the preset low density threshold and the preset high density threshold, and the third determination criterion is that the average distribution density is less than or equal to the preset low density threshold. The mean distribution density of each candidate interval is substituted into the three-level density judgment criteria for matching. If it meets the first judgment criterion, it is marked as high density level; if it meets the second judgment criterion, it is marked as medium density level; and if it meets the third judgment criterion, it is marked as low density level.

6. A tungsten content detection system based on a gradient algorithm, characterized in that, For implementing the gradient algorithm-based tungsten content detection method as described in any one of claims 1 to 5, the system comprises: The detection module is used to acquire various types of tungsten ore samples and to detect in real time the spectral feature data, microstructure image data and elemental composition detection data corresponding to the tungsten ore samples through a preset device. A creation module is used to integrate and process the spectral feature data, the microstructure image data, and the elemental composition detection data in real time to generate corresponding sample feature data in real time, and to create a corresponding dynamic gradient correction model in real time based on the sample feature data. The adjustment module is used to dynamically divide several content gradient intervals according to the distribution density of the sample feature data through the dynamic gradient correction model and the preset gradient adjustment algorithm. When the sample feature data is densely distributed within a certain content range, the gradient interval within that range is automatically reduced; when the sample feature data is sparsely distributed within a certain content range, the gradient interval within that range is automatically expanded. The output module is used to acquire actual sample feature data of actual tungsten ore, and output the corresponding target content gradient interval based on the actual sample feature data through the dynamic gradient correction model, so as to extract the target content within the target content gradient interval.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the tungsten content detection method based on the gradient algorithm as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the tungsten content detection method based on the gradient algorithm as described in any one of claims 1 to 5.

Images