Laser spectrum data processing method for analyzing uranium polymetallic ore elements

Abstract

The embodiment of the invention provides a laser spectrum data processing method for analyzing uranium polymetallic ore elements. According to the method, original spectrum data obtained by performing spectrum detection on a uranium polymetallic ore sample through spectrum detection equipment can be obtained. And inputting the original spectral data into the detection model to obtain detection result data output by the detection model. And obtaining experimental result data, and verifying the detection model by comparing the detection result data with the experimental result data. Wherein the detection model is a convolutional neural network model obtained by training based on sample spectral data, and the detection model can execute model training based on a dynamic uncertainty weighted loss function. According to the method, the multi-task convolutional neural network and the dynamic uncertainty weighted loss function in deep learning are combined, and the original spectral data is taken as input, so that the dependence on spectral preprocessing can be reduced, and meanwhile, the quantitative analysis precision of the target element concentration is improved.

Description

Technical Field

[0001] The present application relates to the technical field of spectral component analysis, and in particular to a laser spectral data processing method for analyzing uranium polymetallic ore elements. Background Art

[0002] Uranium polymetallic ore, as a mineral resource, can be associated with rare earth elements (such as samarium Sm, neodymium Nd, dysprosium Gd, etc.) and other precious metals. These rare earth elements play an irreplaceable role in the fields of new energy, nuclear industry, electronic technology, etc. Accurate analysis of the rare earth element content in uranium polymetallic ore can not only provide a scientific basis for resource development, but also promote the comprehensive utilization of mineral resources and environmental protection.

[0003] In order to analyze the element content in uranium polymetallic ores, laser-induced breakdown spectroscopy (LIBS) can be used for element analysis. Laser-induced breakdown spectroscopy is a fast, non-destructive multi-element analysis method that is widely used in the detection of minerals, metals and environmental samples. The real-time and in-situ multi-element analysis characteristics of laser-induced breakdown spectroscopy can be used for mineral analysis.

[0004] However, in the analysis of uranium polymetallic ores and their associated rare earth elements, the spectral signals of rare earth elements are accompanied by serious spectral line overlap and matrix effects. Among them, spectral interference and spectral line overlap will reduce the accuracy of the analysis, especially when the characteristic peaks of rare earth elements and uranium spectra are in similar bands, the interference is particularly serious, resulting in a significant decrease in the accuracy and reliability of the analysis results. The matrix effect will affect the stability of the quantitative results of rare earth element concentrations, resulting in large error fluctuations. Therefore, the application of laser induced breakdown spectroscopy technology in the detection of elements in uranium polymetallic ores is still limited by spectral interference, matrix effects and analysis accuracy. Summary of the invention

[0005] In view of this, an embodiment of the present application provides a laser spectral data processing method for analyzing elements in uranium polymetallic ores, so as to solve the problem of low accuracy in the process of spectral analysis of element content.

[0006] According to one aspect of the present application, a laser spectrum data processing method for analyzing uranium polymetallic ore elements is provided, the method comprising: Acquiring original spectral data, wherein the original spectral data is a two-dimensional array obtained by performing spectral detection on a uranium polymetallic ore sample through a spectral detection device; The original spectral data is input into a detection model to obtain detection result data output by the detection model, wherein the detection model is a convolutional neural network model obtained by training using sample spectral data according to a dynamic uncertainty weighted loss function; the sample spectral data includes spectral data provided with concentration labels, and the concentration labels are provided according to experimental detection results; the detection result data includes concentration information of target elements; the target elements are one or more element combinations in the uranium polymetallic ore sample that will produce spectral line overlap and / or matrix effects; Acquiring experimental result data, wherein the experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data; The detection model is verified by comparing the detection result data with the experimental result data.

[0007] Optionally, the method further includes: Acquiring the sample spectrum data; Inputting the sample spectrum data into the detection model to obtain training result data output by the detection model; Based on a dynamic uncertainty weighted loss function, calculating a training loss according to the training result data and the concentration label; Model parameters of the detection model are adjusted according to the training loss.

[0008] Optionally, adjusting the model parameters of the detection model according to the training loss includes: Extracting concentration information of the target element from the training result data; Reading the concentration tag corresponding to the target element; Calculating an uncertainty parameter according to the concentration information of the target element and the concentration label, wherein the uncertainty parameter includes at least one of a spectral feature signal-to-noise ratio, a spectral line interference degree, and a detection error; A weighting factor of the dynamic uncertainty weighted loss function is set according to the uncertainty parameter.

[0009] Optionally, setting a weighting factor of the dynamic uncertainty weighted loss function according to the uncertainty parameter includes: Compare the uncertainty parameters corresponding to the plurality of target elements to determine a key element and a non-key element, wherein the key element is the target element with the largest uncertainty parameter; Obtaining a preset parameter threshold and the uncertainty parameter corresponding to the key element; If the uncertainty parameter corresponding to the key element is higher than a preset parameter threshold, a weight value of the key element is set according to the uncertainty parameter, and the weight value has an inverse proportional relationship with the uncertainty parameter.

[0010] Optionally, obtain raw spectral data, including: Setting acquisition parameters of the spectrum detection device, wherein the acquisition parameters include laser energy, wavelength range and delay time; the delay time is the time difference between the moment when the spectrum detection device emits the detection light and the moment when the spectrum is collected; the delay time is set according to the experimental detection results; Controlling the spectrum detection device to emit detection light to the uranium polymetallic ore sample according to the laser energy and the wavelength range; After the delay time, the original spectrum data collected by the spectrum detection device is obtained.

[0011] Optionally, the detection model includes multiple convolution modules; the multiple convolution modules form a multi-level convolution processing network structure; in the direction from the input node to the output node, the convolution kernel size of the multiple convolution modules decreases successively; the convolution module includes a convolution layer, an activation layer and a pooling layer; the original spectral data is input into the detection model to obtain the detection result data output by the detection model, including: Calculating the dot product of the convolution kernel and the local area of ​​the original spectral data through the convolution layer of the convolution module to extract spectral feature data from the original spectral data; Using the activation layer of the convolution module, performing a nonlinear transformation function mapping on the spectral feature data to obtain nonlinear transformation feature data; The nonlinear transformation feature data is downsampled through the pooling layer of the convolution module to reduce the dimension of the nonlinear transformation feature data and generate feature information.

[0012] Optionally, the detection model further includes a fully connected layer, and the fully connected layer is connected to the convolution module; inputting the original spectral data into the detection model to obtain the detection result data output by the detection model includes: According to the number of elements of the target element, the number of output units of the fully connected layer is set; Obtaining a weight matrix and a bias vector of the fully connected layer; According to the number of output units, multiplying the feature information and the weight matrix, and adding the bias vector to obtain a one-dimensional vector; Regression analysis is performed based on the one-dimensional vector to obtain concentration information of the target element.

[0013] Optionally, verifying the detection model by comparing the detection result data with the experimental result data includes: Calculate verification parameters according to the test result data and the experimental result data, wherein the verification parameters include correlation coefficient and / or root mean square error; Get verification condition information; If the verification parameter satisfies the verification condition information, first verification result information is generated, where the first verification result information is used to indicate that the detection accuracy and detection stability of the detection model meet preset requirements; If the verification parameter does not satisfy the verification condition information, second verification result information is generated, and the second verification result information is used to characterize that the detection accuracy and detection stability of the detection model do not meet preset requirements.

[0014] Optionally, the method further includes: Obtain the number and type of elements of the target element; Setting the number of output nodes of the detection model according to the number of elements; According to the element type, determining the spectral line overlap information and matrix effect information between the plurality of target elements in the original spectral data; The loss function of the detection model is set according to the spectral line overlap information and the matrix effect information.

[0015] According to another aspect of the present application, a laser spectroscopy data processing system for analyzing uranium polymetallic ore elements is provided, the system comprising: A spectrum acquisition module, used to acquire original spectrum data, wherein the original spectrum data is a two-dimensional array obtained by performing spectrum detection on a uranium polymetallic ore sample through a spectrum detection device; A detection module, used for inputting the original spectral data into a detection model to obtain detection result data output by the detection model, wherein the detection model is a convolutional neural network model obtained by training using sample spectral data according to a dynamic uncertainty weighted loss function; the sample spectral data includes spectral data provided with concentration labels, and the concentration labels are provided according to experimental detection results; the detection result data includes concentration information of target elements; the target elements are one or more element combinations in the uranium polymetallic ore sample that will produce spectral line overlap and / or matrix effects; An experimental data acquisition module, used to acquire experimental result data, wherein the experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data; A verification module is used to verify the detection model by comparing the detection result data with the experimental result data.

[0016] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned laser spectral data processing method for analyzing uranium polymetallic ore elements when executing the program.

[0017] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the laser spectrum data processing method for analyzing uranium polymetallic ore elements is implemented.

[0018] By means of the above technical scheme, the embodiment of the present application provides a laser spectral data processing method for analyzing uranium polymetallic ore elements, and the method can obtain the original spectral data obtained by the spectral detection equipment for uranium polymetallic ore samples. And the original spectral data is input into the detection model to obtain the detection result data output by the detection model. Then the experimental result data is obtained, and the detection model is verified by comparing the detection result data and the experimental result data. Among them, the detection model is a convolutional neural network model obtained by training based on sample spectral data, and the detection model can perform model training based on a dynamic uncertainty weighted loss function. The method combines the multi-task convolutional neural network and the dynamic uncertainty weighted loss function in deep learning, takes the original spectral data as input, can reduce the dependence on spectral preprocessing, and improve the quantitative analysis accuracy of the target element concentration. The method can improve the adaptability to spectral line interference and matrix effects, and realize high-precision analysis of rare earth elements. And realize end-to-end multi-element synchronous analysis, can simplify the analysis process, and does not require complex preprocessing steps, reducing operating costs. In addition, it can also support the rapid detection of larger-scale mineral samples and improve the adaptability of the analysis process.

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic diagram of a spectral detection scenario provided in an embodiment of the present application; Figure 2 A schematic flow chart of a laser spectrum data processing method for analyzing uranium polymetallic ore elements provided in an embodiment of the present application; Figure 3 A schematic diagram of the detection model training process provided in an embodiment of the present application; Figure 4 A schematic diagram of a weighting factor setting process provided in an embodiment of the present application; Figure 5 A schematic diagram of a process for setting the weight value of key elements provided in an embodiment of the present application; Figure 6 A schematic diagram of the detection model structure provided in the embodiment of the present application; Figure 7 A schematic diagram of the process flow of the verification and detection model provided in the embodiment of the present application; Figure 8 A schematic diagram of the structure of a detection system for analyzing uranium and polymetallic ores using induced breakdown spectroscopy technology provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0022] In the embodiments of the present application, spectral data refers to the data obtained by measuring the optical characteristics of the sample such as reflection, absorption, and radiation under the action of detection light signals of different wavelengths or frequencies through spectral detection equipment. By analyzing spectral data, sample composition analysis, quality detection, anomaly detection, etc. can be achieved.

[0023] The sample refers to the object to be tested for component analysis, quality inspection and anomaly detection. Different samples can be prepared according to different application scenarios. For example, the sample can be a uranium polymetallic ore. Uranium polymetallic ore refers to a mineral resource containing uranium and other metal elements. As a mineral resource, uranium polymetallic ore may be accompanied by rare earth elements (such as samarium Sm, neodymium Nd, dysprosium Gd, etc.) and other precious metals.

[0024] The spectrum detection equipment can be a spectrophotometer, a spectrometer, a laser spectrometer, etc. The spectrum detection equipment can disperse the incident light into spectra of different wavelengths through optical elements, and detect the spectra at different wavelengths through photodetectors to form original spectrum data. Since problems such as ambient light interference, instrument errors, and uneven sample preparation may be encountered during the spectrum data acquisition process, the spectrum detection equipment also needs to include a shading component, a calibration component, and a sample holding component, etc. These components can be used to implement shading measures, instrument calibration, and sample preparation standardization during the spectrum detection process to improve the accuracy and stability of data acquisition.

[0025] Different spectral detection equipment can collect different spectral data based on different principles. For example, some spectral detection equipment can perform component detection based on Laser-induced Breakdown Spectroscopy (LIBS). Laser-induced breakdown spectroscopy is an atomic emission spectroscopy analysis technology. When performing spectral detection, spectral detection equipment such as spectrometers can use high-energy laser pulses to irradiate samples so that the laser beam is quickly focused on the sample surface. After the energy of the laser beam is absorbed by the atoms and molecules of the sample, a small part of the material on the surface of the sample is evaporated to form a vapor plume, that is, the sample is evaporated. In this process, the laser beam interacts with the plume to generate a bright plasma. Each element in the plasma emits light of a specific wavelength after excitation and ionization. This light can be collected by the sensor of the spectrometer and used to analyze the composition of the sample.

[0026] like Figure 1 As shown, after acquiring the spectral data, the spectral detection device can send the acquired raw spectral data to the data processing device. The data processing device can have a built-in spectral detection algorithm or application to process the raw spectral data. The data processing device refers to a device with data processing capabilities, including but not limited to computers, mobile terminals, servers, workstations, smart wearable devices, industrial hosts, etc. The types of processing of the raw spectral data by the data processing device can include preprocessing and detection processing.

[0027] Spectral data preprocessing refers to the specific screening, transformation, and modification of the original spectral data before spectral detection and processing, so as to improve the data quality of the original spectral data and improve the accuracy and reliability of subsequent detection and processing. For example, spectral data preprocessing can include baseline correction, noise reduction, normalization, standard normal variable transformation, multivariate scattering correction, derivative spectroscopy, deconvolution, etc.

[0028] It should be noted that the content of the preprocessing of the spectral data and whether to perform the preprocessing need to be determined according to the specific detection requirements. That is, in some embodiments, it is possible to directly use the original spectral data for detection processing without performing preprocessing. For example, since the light emitted by the plasma is dispersed by the spectrometer, a spectrum diagram containing multiple spectral lines can be formed, and these spectral lines correspond to specific energy level transitions of the elements present in the sample. Therefore, the spectral analysis process based on the LIBS technology may require little or no sample preprocessing, and directly analyze the original spectral data. By comparing the characteristic wavelengths in the spectrum with the pre-built spectral database, the elements in the sample can be identified, and the content of the elements can be determined based on the spectral intensity. Quantitative analysis can also be performed by establishing a calibration curve or using a calibration-free LIBS (CF-LIBS) method. Therefore, LIBS technology can achieve rapid analysis, no or little sample preparation, and non-destructive analysis, so that LIBS technology can be applied to the analysis of samples in different states such as solids, liquids, gases and aerosols, and thus applied to multiple fields.

[0029] After preprocessing, the data processing equipment can detect and process the preprocessed spectral data. The principle of spectral data detection and processing is based on measuring the wavelength and intensity of the emitted, absorbed or scattered radiation generated by the transition between quantized energy levels inside the material when the material interacts with the radiation energy, and analyzing it to detect the elemental composition, structure and content of the material.

[0030] In order to predict and analyze the elemental composition, structure, content and other information in the sample, the data processing device can perform feature extraction on the spectral data when performing spectral detection processing. Feature extraction is to extract parameters that can represent the characteristics of the sample from the spectral data. Then, based on the extracted feature quantity, a mathematical model is established, that is, a relationship is established between the feature quantity and the sample property for prediction and analysis. For example, the data processing device can perform detection processing based on one or more combinations of analytical methods such as atomic absorption spectroscopy (AAS), inductively coupled plasma mass spectrometry (ICP-MS) and X-ray fluorescence spectroscopy (XRFS).

[0031] Through the above analysis method, characteristic information can be extracted from the spectral data according to a preset analysis method, so as to perform component detection based on the characteristic information. However, these analysis methods have high equipment costs, complex sample pretreatment, slow analysis speed, and are difficult to achieve in-situ detection. Therefore, in some embodiments, the spectral data can also be analyzed based on a machine learning algorithm. After acquiring the spectral data, the data processing device can use a machine learning model to extract features from the spectral data and extract characteristic information such as peak value, wavelength, and intensity from the spectral data. Then, through function mapping, downsampling, and regression analysis, the analysis results of the spectral data on the concentration information of a specific substance are determined.

[0032] For example, the interval partial least squares method (interval-PLS) can be used to quantitatively analyze rock and soil samples containing rare earth elements. Or the spectra of rare earth elements in rare earth ores can be directly quantitatively analyzed to improve the analysis accuracy. However, when using machine learning algorithms for spectral analysis, there are still high requirements for the spectral preprocessing of samples, and it is easy to be interfered by substances, which reduces the detection accuracy and stability.

[0033] Among them, interference between substances refers to the mutual interference between the characteristic quantities of different substances in the spectral data when the sample being tested contains two or more substances or elements, that is, the generation of spectral line overlap and matrix effect. Spectral line overlap refers to the phenomenon that in spectral analysis, due to the similar energy of the spectral lines of different elements, multiple signals overlap each other at adjacent positions. This overlapping interference will cause abnormal measurement intensity of the overlapped elements, thereby affecting the accuracy of the analysis results. For example, spectral line overlap can occur between the same-order lines of adjacent or next-nearest adjacent elements in the periodic table, the overlap of first-order lines of different line series, the overlap of K-series lines of light elements and higher-order lines of heavy elements, and the overlap of first-order lines of L-series and higher-order lines of K-series of lighter elements.

[0034] Matrix effect refers to the influence of other elements (collectively referred to as matrix) in the sample other than the element to be measured on the characteristic spectrum measurement of the element to be measured. Matrix effect can include spectral matrix effect and non-spectral matrix effect. Spectral matrix effect refers to the continuous background emission in the inductively coupled plasma (ICP) discharge, and non-spectral matrix effect is related to many factors such as the nature of the analyte and interferent, operating parameters, instrument type and injection method.

[0035] For different samples, the degree of spectral line overlap and matrix effect during spectral analysis is different. For example, the element of iron ore that shows the most complex characteristic peak in the LIBS spectrum is Fe itself. Other interfering elements such as calcium Ca, aluminum Al, silicon Si and other elements have limited interference due to their own fewer characteristic peaks. In uranium polymetallic ores rich in rare earth elements, each rare earth element has complex and numerous characteristic spectral lines due to its physical properties. In addition, matrix elements and target elements, and target elements interfere with each other, which makes the quantitative analysis of rare earth elements in uranium polymetallic ores much more complicated than the quantitative analysis of elements in iron ore when performing LIBS analysis. In addition, the characteristic spectral lines of each rare earth element are highly dense due to their complex energy level transitions, resulting in strong overlap of spectral lines, which will increase the difficulty of identifying the target elements. At the same time, the matrix effect will significantly affect the emission intensity of rare earth elements, and the plasma interaction between matrix elements (such as uranium and iron) and rare earth elements may change the excitation efficiency of rare earth elements. In addition, due to the similar physical and chemical properties between different rare earth elements, the similarity and proximity of their spectral lines will also increase the difficulty of quantitative analysis.

[0036] In the quantitative analysis of rare earth elements, due to the nonlinear response characteristics of spectral data and background interference, the plasma spectral signal generated by LIBS shows a nonlinear relationship with the composition content of the sample, and the complex spectrum of rare earth elements will further complicate this nonlinearity. In addition, the weak spectral lines of rare earth elements are easily masked or interfered by other strong spectral lines, and some spectral lines may drift under the influence of matrix effects, resulting in a significant increase in the error of quantitative analysis.

[0037] In order to improve the accuracy of spectral analysis of element content and reduce the impact of spectral line overlap, matrix effect, background interference and other problems, some embodiments of this application provide a laser spectral data processing method for analyzing uranium polymetallic ore elements. The method can directly use the original spectral data as input by combining the multi-task convolutional neural network (CNN) and dynamic uncertainty weighted loss function in deep learning, reduce the dependence on spectral preprocessing, and improve the quantitative analysis accuracy of rare earth element concentration. Figure 2 As shown, the method includes: S101, obtaining original spectral data.

[0038] When performing element detection on uranium polymetallic ore, the data processing device can first collect spectrum of the uranium polymetallic ore sample through the spectrum detection device to obtain original spectrum data. The original spectrum data is a two-dimensional array obtained by performing spectrum detection on the uranium polymetallic ore sample through the spectrum detection device.

[0039] According to the different detection principles of the spectrum detection equipment, the specific content contained in the original spectrum data is different, that is, the type of values ​​contained in the two-dimensional array is different. For example, the two-dimensional array includes wavelength and intensity, which is used to characterize the signal intensity of the sample under test for a specific wavelength. The signal intensity can be absorbance, transmittance, reflectivity, count rate, etc., which is used to characterize the sample's absorption or reflection ability for a specific wavelength of light.

[0040] When the data processing device acquires the original spectrum data, it can first set the acquisition parameters of the spectrum detection device, wherein the acquisition parameters include laser energy and wavelength range.

[0041] Laser energy refers to the light energy emitted per unit area by the laser of the spectral detection equipment. The laser energy is measured in millijoules (mJ). In the laser induced breakdown spectroscopy (LIBS) system, the laser energy has a direct impact on the ablation effect of the sample surface, which in turn affects the formation of plasma and the intensity of the spectral signal. For example, the laser energy of the spectrometer can be set to 200mJ and the pulse time to 5ns.

[0042] The wavelength range refers to the wavelength range of the optical signal that the spectral detection device can detect. The setting of the wavelength range can be based on the transmission or reflection spectral characteristics of the optical elements (such as gratings and prisms) used in the spectral instrument, and the spectral response range of the detector. For example, the wavelength range of the spectrometer can be set to 190nm-1100nm.

[0043] Since the spectral data of uranium polymetallic ore is easily affected by background interference, which reduces the accuracy of subsequent analysis, in some embodiments, background signal interference can be reduced based on a delayed acquisition method. That is, the acquisition parameters also include a delay time.

[0044] The delay time refers to the time difference between the moment when the spectrum detection device emits the detection light and the moment when the spectrum is collected. The delay time can be set according to the experimental detection results. That is, before performing the uranium polymetallic ore element detection, multiple uranium polymetallic ore samples from a specific mining area can be collected, and the uranium polymetallic ore samples can be spectrally detected at a preset frequency to obtain multiple spectrum data. Then, the obtained multiple spectrum data are analyzed to determine the interference degree of the background signal in each spectrum data.

[0045] Since the background signals that can interfere with the analysis results are mostly generated in the early stage of spectral detection, that is, in the early stage of detection, the initial state of the spectral detection equipment, operating vibration, ambient light, etc. will interfere with the detection process. These interferences will tend to stabilize over time, so by comparing multiple spectral data, the earliest time when the spectral data stabilizes can be determined, that is, the minimum delay time can be determined.

[0046] After setting the acquisition parameters, the set acquisition parameters can be input into the spectrum detection device to control the spectrum detection device to collect spectrum data according to the set acquisition parameters. That is, the spectrum detection device is controlled to emit detection light to the uranium polymetallic ore sample according to the laser energy and the wavelength range, and after a delay time, the original spectrum data collected by the spectrum detection device is obtained.

[0047] For example, when detecting rare earth elements such as samarium Sm, neodymium Nd, and dysprosium Gd in uranium polymetallic ores, the following acquisition parameters can be set according to the detection requirements, that is, the laser energy is 40mJ, the pulse time is 5ns, and the wavelength range is 250nm-750nm. When performing spectral data acquisition, the spectrometer can be set to emit detection light in the range of 250nm-750nm to the uranium polymetallic ore sample according to the laser energy of 40mJ. The delay time is 3μs, and the original spectral data collected by the spectrometer is obtained 3μs after the detection light is emitted.

[0048] S102: Input the original spectrum data into a detection model to obtain detection result data output by the detection model.

[0049] After acquiring the raw spectral data, the data processing device can call the detection model and input the acquired raw spectral data into the detection model. The detection model is a convolutional neural network model obtained by training based on sample spectral data. The detection model performs convolution calculation on the raw spectral data to realize feature extraction, feature analysis and result output, and the detection result data can be obtained. Among them, the detection result data may include concentration information of the target element. The target element is the element of the concentration to be detected, that is, the target element is the rare earth metal or other precious metal element associated with the uranium polymetallic ore. In addition, the target element is a combination of one or more elements in the uranium polymetallic ore sample that will produce spectral line overlap and / or matrix effect. For example, the target element may include three rare earth elements, namely samarium Sm, neodymium Nd, and dysprosium Gd.

[0050] In order to detect the concentration information of the target element, the input node of the detection model is set to receive the original spectral data, and the output node of the detection model is set to output the concentration information of the target element. The number of output nodes is equal to the number of target elements. For example, when analyzing the concentration of samarium Sm, neodymium Nd, and dysprosium Gd in uranium polymetallic ore through the detection model, the number of output nodes of the detection model can be set to 3, which are used to output the concentrations of samarium Sm, neodymium Nd, and dysprosium Gd respectively.

[0051] In some embodiments, the detection model can be trained in advance using sample spectral data. The sample spectral data includes spectral data set with a concentration label, and the concentration label is set according to the experimental detection results. Before performing element detection, sample spectral data needs to be generated based on the experiment. During the experiment, the uranium polymetallic ore sample can be first spectrally collected to obtain spectral data. Then, based on chemical reagent detection and other methods, the concentration information of the target element in the uranium polymetallic ore sample is determined. The concentration information obtained by chemical detection is then set as a concentration label to the spectral data, thereby obtaining a data pair containing a spectrum and a concentration label, and generating sample spectral data.

[0052] In order to improve the detection accuracy of the detection model, the sample spectral data generated by the experimental detection results needs to reach a predetermined sample capacity. Obviously, the larger the sample capacity, the higher the accuracy of the detection model obtained by training and the stronger the generalization ability. In order to increase the sample capacity of the sample spectral data, it is also possible to combine atomic absorption spectroscopy, inductively coupled plasma mass spectrometry, X-ray fluorescence spectroscopy and other analytical methods on the basis of chemical reagent inspection to perform concentration analysis on uranium polymetallic ore samples to obtain the concentration information of the target element. The concentration information obtained from the concentration analysis is set as a concentration label to the spectral data to increase the sample capacity.

[0053] After generating sample spectral data, you can also divide the sample spectral data into data sets, that is, divide the sample spectral data into training sets, test sets, and validation sets. Among them, the training set is used to execute the training model, that is, adjust the model parameters through the preset training method and loss function, so that the model can learn the patterns and laws in the data. The test set is used to evaluate the generalization ability of the model, that is, the performance of the model on unseen data. The validation set is used for model selection and hyperparameter adjustment, so as to select the best model and adjust the model parameters. The training set, test set, and validation set can meet the preset data volume ratio. For example, the data volume ratio of the training set, test set, and validation set is 7:2:1.

[0054] In some embodiments, before executing model training, relevant parameters of the model may be set according to specific analysis needs. That is, the input nodes, output nodes, and loss function of the trained model (the detection model in training) are set. Among them, the output nodes may be set according to the number of elements of the target element, so the data processing device may obtain the number of elements of the target element and set the number of output nodes of the detection model according to the number of elements.

[0055] The loss function can be set according to the element type of the target element, that is, different loss functions can be set for different element types. Therefore, the data processing device can obtain the element type of the target element. Then, according to the element type, the spectral line overlap information and matrix effect information between the multiple target elements in the original spectrum data are determined. Then, the loss function of the detection model is set according to the spectral line overlap information and matrix effect information.

[0056] For example, when the detection model is used to detect the concentration of Fe in an iron ore sample, the interference elements such as calcium Ca, aluminum Al, and silicon Si in the iron ore sample have fewer characteristic peaks, and the interference effect is limited, that is, the spectral line overlap information and matrix effect information between multiple target elements indicate that the spectral line overlap and matrix effect problems between the current sample elements are not significant. Therefore, the loss function of the detection model can be set to the mean square error loss function, the mean absolute error loss function, etc., to improve the iterative efficiency of the model training process.

[0057] When the detection model is used to detect the concentration of rare earth elements such as samarium Sm, neodymium Nd, and dysprosium Gd in uranium polymetallic ores, the spectral line overlap information and matrix effect information between multiple target elements indicate that the spectral line overlap and matrix effect problems between the current sample elements are significant. Therefore, the loss function of the monitoring model can be set as a dynamic uncertainty weighted loss function, so that the detection model performs model training based on the dynamic uncertainty weighted loss function.

[0058] Among them, the Dynamically Weighted Balanced Loss (DWBLoss) is a loss function used to deal with class imbalance problems in machine learning. It can improve the model confidence calibration by dynamically adjusting class weights and introducing regularization components. DWB Loss can make the training process pay more attention to difficult-to-train instances through dynamic weighting, thereby improving the performance of the model in terms of class imbalance and difficult and easy sample instances. Based on the weight of DWB Loss in the model training process, it can be allocated according to the class frequency and the predicted probability of the true class, allowing the model to adjust instances with different difficulties.

[0059] like Figure 3As shown, after acquiring the sample spectral data and setting the relevant information of the trained model (the detection model in training), the data processing device can input the sample spectral data into the detection model in training to obtain the training result data output by the detection model. The training result data is the target element concentration information obtained by the trained model after feature extraction, analysis, and calculation based on the sample spectral data. Since the trained model is still in training at this time, that is, the detection model has not converged, there is a certain deviation between the target element concentration information in the training result data and the actual concentration information. The data processing device can calculate the training loss based on the training result data and the concentration label based on the dynamic uncertainty weighted loss function.

[0060] After calculating the training loss, the data processing device can adjust the model parameters of the trained model according to the training loss. In some embodiments, during training, the data processing device can judge the calculated training loss, that is, compare the training loss with the preset loss threshold. When the training loss is greater than the preset loss threshold, it means that the current trained model has not yet been trained and converged, so the model parameters of the trained model can be modified based on the training loss, that is, the model weights, dynamic weighting factors, trainable variance parameters, regularization parameters, learning rates, hyperparameters, etc. are modified. The data processing device can perform iterative training based on the trained model after the model parameters are modified until the training loss is less than or equal to the preset loss threshold. When the training loss is less than or equal to the preset loss threshold, it means that the current model has been trained and converged, so the model parameters of the trained model can be recorded and output to obtain a detection model.

[0061] Since the detection model performs model training based on a dynamic uncertainty weighted loss function, when adjusting the model parameters of the detection model according to the training loss, the training weights can also be set according to the training state parameters during the training process. Figure 4 As shown, in some embodiments, when executing model training, the data processing device can also extract the concentration information of the target element from the training result data and read the concentration label corresponding to the target element, and then calculate the uncertainty parameter based on the concentration information of the target element and the concentration label.

[0062] Wherein, the uncertainty parameter includes at least one of the signal-to-noise ratio of the spectral feature, the degree of spectral line interference, and the detection error. Based on the dynamic uncertainty weighted loss function, the uncertainty estimation can be introduced in the training process, and the uncertainty of the spectral signal due to spectral line overlap, matrix effect and other factors can be used as a weighting factor to dynamically adjust the training target of the model. Thus, the weighting factor of the dynamic uncertainty weighted loss function is set according to the uncertainty parameter.

[0063] For example, different rare earth elements can be assigned different weights based on the signal-to-noise ratio of the spectral features, the degree of spectral line interference, and the predicted error level. For target variables with higher uncertainty, such as rare earth elements with low signal intensity or severe overlap, their proportion in the loss function can be reduced, thereby reducing the model's overfitting of these high-uncertainty target variables.

[0064] For this reason, Figure 5 As shown, in some embodiments, the data processing device can compare the uncertainty parameters corresponding to the multiple target elements to determine the key elements and non-key elements. Among them, the key element is the target element with the largest uncertainty parameter. Then obtain the preset parameter threshold and the uncertainty parameter corresponding to the key element. If the uncertainty parameter corresponding to the key element is higher than the preset parameter threshold, the weight value of the key element is set according to the uncertainty parameter. In order to reduce the overfitting of the model to the high-uncertainty target elements, the weight value has an inverse proportional relationship with the uncertainty parameter.

[0065] For example, when using the dynamic uncertainty weighted loss function (DWB Loss) to calculate the training loss, the dynamic weighted cross entropy loss term and the regularization component can be calculated separately. The dynamic weighted cross entropy loss term can be calculated according to the following formula: ; in, n is the number of training samples; c is the number of categories; w j Yes Category j The weight of p ij It is a sample i Belongs to category j The predicted probability of y ij It is a sample i The label can be encoded using one-hot encoding.

[0066] And, the regularization component is calculated as follows: ; in, n is the number of training samples; c is the number of categories; p ij It is a sample i Belongs to category j The predicted probability of .

[0067] After calculating the dynamically weighted cross entropy loss term and regularization component respectively in the above manner, the sum of the cross entropy loss term and the regularization component can be calculated, that is, the total DWB Loss is obtained, which can be expressed as: L DWB = L CE + L BS ; in, L DWB is the dynamic uncertainty weighted loss, L CE is the dynamically weighted cross entropy loss term; L BS is the regularization component.

[0068] It can be seen that by introducing a dynamic uncertainty weighted loss function into the detection model, the uncertainty introduced in the spectral signal due to spectral line overlap, matrix effect, etc. can be used as a weighting factor to dynamically adjust the training target of the model, thereby balancing the difficulty of task training between rare earth elements (such as samarium Sm, neodymium Nd, and dysprosium Gd) and improving the robustness of the detection model in simultaneous analysis of multiple elements. Therefore, the detection model obtained through training can focus more on features with clear signals and less interference, effectively alleviating the problem of error accumulation in complex spectral data.

[0069] After the detection model is obtained through training, the concentration of the target element can be analyzed based on the original spectral data. Figure 6 As shown, in order to perform concentration analysis, in some embodiments, the detection model may include multiple convolution modules, and the multiple convolution modules form a multi-level convolution processing network structure. And in the direction from the input node to the output node, the convolution kernel size of the multiple convolution modules decreases successively. For example, the detection model may include 5 convolution modules, and the 5 convolution modules are connected in sequence to form a 5-level convolution processing network structure. Each module is responsible for extracting feature information at different levels. After multi-level convolution processing, the feature inventory can be gradually reduced. For example, the convolution kernel size of the first-level convolution module is 9×9, then after the convolution module performs preliminary feature extraction processing, more than 50,000 feature points in the spectral data can be preliminarily extracted. Therefore, through the multi-level convolution processing network structure, features that have a significant impact on concentration analysis can be gradually extracted from the spectral data.

[0070] The convolution module includes a convolution layer, an activation layer, and a pooling layer. Among them, the convolutional layer (ConvolutionalLayer, Conv) as the core of CNN can slide on the input data through the convolution kernel (or filter), calculate the dot product of the filter and the local area of ​​the input data, and thus extract features. The parameters of the convolution layer include the number and scale of the convolution kernel, such as 3×3, 5×5, 7×7, 9×9, etc., stride, padding, etc. The convolution layer can capture the local features of the input data, such as edges, textures, etc., and generate feature maps.

[0071] The activation layer can introduce nonlinearity to the feature map output by the convolution layer, so that the CNN network can learn complex function mappings. For example, the activation layer can have built-in activation functions such as ReLU (Rectified Linear Unit), Sigmoid, Tanh, etc., so that the network can learn nonlinear relationships.

[0072] The pooling layer is used to reduce the spatial size of the feature map, reduce the number of parameters and the amount of calculation, and extract important features. The pooling layer can perform downsampling based on pooling operations such as max pooling and average pooling, that is, by selecting the maximum value or average value in the area to achieve downsampling of the feature map.

[0073] Convolutional layers, activation layers, and pooling layers can appear alternately in CNN. The convolutional layer extracts features, the activation layer introduces nonlinearity, and the pooling layer reduces the dimension of the feature map, so that the detection model can predict the concentration information of the target element based on the input raw spectral data.

[0074] In some embodiments, when the detection model detects the concentration information of the target element, the convolution layer of the convolution module can be used to calculate the local area dot product between the convolution kernel and the original spectral data to extract spectral feature data from the original spectral data. Then, the activation layer of the convolution module is used to perform a nonlinear transformation function mapping on the spectral feature data to obtain nonlinear transformation feature data. Then, the pooling layer of the convolution module is used to downsample the nonlinear transformation feature data to reduce the dimension of the nonlinear transformation feature data and generate feature information.

[0075] In addition, the detection model further includes a fully connected layer, and the fully connected layer is connected to the convolution module. After the data processing device inputs the original spectral data into the detection model and performs preliminary calculations through the convolution module, the number of output units of the fully connected layer can be set according to the number of elements of the target element. Then, the weight matrix and bias vector of the fully connected layer are obtained, and according to the number of output units, the feature information and the weight matrix are multiplied, and the bias vector is added to obtain a one-dimensional vector. Regression analysis is performed based on the one-dimensional vector to obtain the concentration information of the target element.

[0076] After obtaining the test result data output by the detection model, the data processing device can visualize the output test result data. For example, in the user interface of the target element concentration information detection application, a test result display control can be included. After obtaining the test result data output by the detection model, the data processing device can extract the concentration information of the target element from the test result data and add the concentration information to the result display control to present the concentration information of rare earth elements such as samarium Sm, neodymium Nd, and dysprosium Gd in the uranium polymetallic ore in the user interface.

[0077] S103. Obtain experimental result data.

[0078] The experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data. After obtaining the detection result data output by the detection model, the experimental result data can also be obtained. The experimental result data is data obtained by performing target element concentration detection on the uranium polymetallic ore sample according to other analysis methods. The other analysis methods can be one or more combinations of analysis methods such as chemical reagent detection method, atomic absorption spectrometry, inductively coupled plasma mass spectrometry, X-ray fluorescence spectrometry, etc. described in the above embodiments.

[0079] It should be understood that, since different analytical methods have different requirements for uranium polymetallic ore samples, in order to ensure data consistency, when conducting concentration detection experiments, the concentration analysis method of non-destructive testing should be performed first, and then the concentration analysis method of destructive testing should be performed. For example, the experimental result data can be obtained by performing analytical methods such as X-ray fluorescence spectroscopy first, and then the chemical reagent detection method can be performed to obtain the experimental result data, so as to improve the accuracy of the subsequent verification process.

[0080] S104: Verify the detection model by comparing the detection result data with the experimental result data.

[0081] After obtaining the test result data and the experimental result data, the data processing device can verify the detection model based on the test result data and the experimental result data, that is, determine whether the test result data output by the detection model is accurate, and which one is more accurate between the test result data output by the detection model and the experimental result data obtained by other experimental methods.

[0082] Therefore, the data processing device can verify the detection model by comparing the detection result data with the experimental result data, such as Figure 7 As shown, in some embodiments, in order to verify the detection model, the data processing device can calculate the verification parameters according to the detection result data and the experimental result data. The verification parameters include the correlation coefficient and / or the root mean square error.

[0083] For example, the correlation coefficient can be calculated as follows: ; Among them, r is the correlation coefficient; xi and yi are the observed values ​​of two variables, and is the mean of the observations.

[0084] The root mean square error can be calculated as follows: ; Among them, RMSE is the root mean square error; yi is the actual observed value, is the model prediction value, n is the number of observation points.

[0085] Then obtain the verification condition information, and compare the verification parameters with the verification condition information. If the verification parameters meet the verification condition information, generate first verification result information, which is used to characterize that the detection accuracy and detection stability of the detection model meet the preset requirements. If the verification parameters do not meet the verification condition information, generate second verification result information, which is used to characterize that the detection accuracy and detection stability of the detection model do not meet the preset requirements.

[0086] For example, in the process of detecting the concentration information of rare earth elements such as samarium Sm, neodymium Nd, and dysprosium Gd, the cross-validation method can be used to optimize the model parameters, and by comparing with the experimental results data, the quantitative analysis accuracy and stability of the model can be verified. According to the concentration prediction effect of the detection model on samarium Sm, neodymium Nd, and dysprosium Gd, it can be calculated that the correlation coefficients are all over 0.99.

[0087] By applying the technical solutions provided in the above embodiments, the method can combine the multi-task convolutional neural network and the dynamic uncertainty weighted loss function in deep learning to improve the accuracy and efficiency of rare earth element analysis in uranium polymetallic ores. The method can improve the adaptability to spectral line interference and matrix effects, and realize high-precision analysis of rare earth elements. It can also realize end-to-end multi-element synchronous analysis, simplify the analysis process, and do not require complex preprocessing steps, reducing operating costs. In addition, the method can support the rapid detection of larger-scale mineral samples, and has strong technical adaptability, which can be extended to the analysis field of other complex mineral samples.

[0088] Further, as a specific implementation of the laser spectrum data processing method for analyzing uranium polymetallic ore elements described in the above embodiment, the embodiment of the present application provides a laser spectrum data processing system for analyzing uranium polymetallic ore elements, such as Figure 8 As shown, the system includes: A spectrum acquisition module, used to acquire original spectrum data, wherein the original spectrum data is a two-dimensional array obtained by performing spectrum detection on a uranium polymetallic ore sample through a spectrum detection device; A detection module, used for inputting the raw spectral data into a detection model to obtain detection result data output by the detection model, wherein the detection model is a convolutional neural network model obtained by training the sample spectral data according to a dynamic uncertainty weighted loss function; the sample spectral data includes spectral data provided with a concentration label, and the concentration label is provided according to an experimental detection result; the detection result data includes concentration information of a target element; An experimental data acquisition module, used to acquire experimental result data, wherein the experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data; A verification module is used to verify the detection model by comparing the detection result data with the experimental result data.

[0089] It should be noted that, for other corresponding descriptions of the functional units involved in the detection system for analyzing uranium polymetallic ores using induced breakdown spectroscopy technology provided in the embodiment of the present application, reference can be made to the corresponding descriptions in the laser spectroscopy data processing method for analyzing uranium polymetallic ores elements provided in the above embodiment, and will not be repeated here.

[0090] By applying the technical solution provided in the above embodiment, the laser spectrum data processing system for analyzing uranium polymetallic ore elements can obtain the original spectrum data obtained by the spectrum detection equipment for uranium polymetallic ore samples. And input the original spectrum data into the detection model to obtain the detection result data output by the detection model. Then obtain the experimental result data, and verify the detection model by comparing the detection result data with the experimental result data. Among them, the detection model is a convolutional neural network model obtained by training based on sample spectrum data, and the detection model can perform model training based on a dynamic uncertainty weighted loss function.

[0091] The system combines the multi-task convolutional neural network and dynamic uncertainty weighted loss function in deep learning, takes the original spectral data as input, reduces the dependence on spectral preprocessing, and improves the quantitative analysis accuracy of the target element concentration. The system can improve the adaptability to spectral line interference and matrix effects, and achieve high-precision analysis of rare earth elements. It can also achieve end-to-end multi-element synchronous analysis, simplify the analysis process, and eliminate the need for complex preprocessing steps, reducing operating costs. In addition, it can also support the rapid detection of larger-scale mineral samples and improve the adaptability of the analysis process.

[0092] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0093] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0094] In one embodiment, a computer-readable storage medium is also provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0095] In one embodiment, a computer program product is also provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0096] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0097] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0098] Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0099] Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0100] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0101] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A laser spectrum data processing method for analyzing uranium polymetallic ore elements, characterized in that: The method comprises: Acquiring original spectral data, wherein the original spectral data is a two-dimensional array obtained by performing spectral detection on a uranium polymetallic ore sample through a spectral detection device; The original spectral data is input into a detection model to obtain detection result data output by the detection model, wherein the detection model is a convolutional neural network model obtained by training using sample spectral data according to a dynamic uncertainty weighted loss function; the sample spectral data includes spectral data provided with concentration labels, and the concentration labels are provided according to experimental detection results; the detection result data includes concentration information of target elements; the target elements are one or more element combinations in the uranium polymetallic ore sample that will produce spectral line overlap and / or matrix effects; Acquiring experimental result data, wherein the experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data; The detection model is verified by comparing the detection result data with the experimental result data.

2. The method according to claim 1, characterized in that: The method further comprises: Acquiring the sample spectrum data; Inputting the sample spectrum data into the detection model to obtain training result data output by the detection model; Based on a dynamic uncertainty weighted loss function, calculating a training loss according to the training result data and the concentration label; Model parameters of the detection model are adjusted according to the training loss.

3. The method according to claim 2, characterized in that Adjusting the model parameters of the detection model according to the training loss includes: Extracting concentration information of the target element from the training result data; Reading the concentration tag corresponding to the target element; Calculating an uncertainty parameter according to the concentration information of the target element and the concentration label, wherein the uncertainty parameter includes at least one of a spectral feature signal-to-noise ratio, a spectral line interference degree, and a detection error; A weighting factor of the dynamic uncertainty weighted loss function is set according to the uncertainty parameter.

4. The method according to claim 3, characterized in that: Setting a weighting factor of the dynamic uncertainty weighted loss function according to the uncertainty parameter includes: Compare the uncertainty parameters corresponding to the plurality of target elements to determine a key element and a non-key element, wherein the key element is the target element with the largest uncertainty parameter; Obtaining a preset parameter threshold and the uncertainty parameter corresponding to the key element; If the uncertainty parameter corresponding to the key element is higher than a preset parameter threshold, a weight value of the key element is set according to the uncertainty parameter, and the weight value has an inverse proportional relationship with the uncertainty parameter.

5. The method according to claim 1, characterized in that Get raw spectral data, including: Setting acquisition parameters of the spectrum detection device, wherein the acquisition parameters include laser energy, wavelength range and delay time; the delay time is the time difference between the moment when the spectrum detection device emits the detection light and the moment when the spectrum is collected; the delay time is set according to the experimental detection results; Controlling the spectrum detection device to emit detection light to the uranium polymetallic ore sample according to the laser energy and the wavelength range; After the delay time, the original spectrum data collected by the spectrum detection device is obtained.

6. The method according to claim 1, characterized in that The detection model includes a plurality of convolution modules; the plurality of convolution modules form a multi-level convolution processing network structure; in the direction from the input node to the output node, the convolution kernel sizes of the plurality of convolution modules decrease in sequence; the convolution module includes a convolution layer, an activation layer and a pooling layer; Inputting the original spectral data into a detection model to obtain detection result data output by the detection model includes: Calculating the dot product of the convolution kernel and the local area of ​​the original spectral data through the convolution layer of the convolution module to extract spectral feature data from the original spectral data; Using the activation layer of the convolution module, performing a nonlinear transformation function mapping on the spectral feature data to obtain nonlinear transformation feature data; The nonlinear transformation feature data is downsampled through the pooling layer of the convolution module to reduce the dimension of the nonlinear transformation feature data and generate feature information.

7. The method according to claim 6, characterized in that The detection model further includes a fully connected layer, and the fully connected layer is connected to the convolution module; the original spectral data is input into the detection model to obtain the detection result data output by the detection model, including: According to the number of elements of the target element, the number of output units of the fully connected layer is set; Obtaining a weight matrix and a bias vector of the fully connected layer; According to the number of output units, multiplying the feature information and the weight matrix, and adding the bias vector to obtain a one-dimensional vector; Regression analysis is performed based on the one-dimensional vector to obtain concentration information of the target element.

8. The method according to claim 1, characterized in that By comparing the detection result data with the experimental result data, verifying the detection model includes: Calculate verification parameters according to the test result data and the experimental result data, wherein the verification parameters include correlation coefficient and / or root mean square error; Get verification condition information; If the verification parameter satisfies the verification condition information, first verification result information is generated, where the first verification result information is used to indicate that the detection accuracy and detection stability of the detection model meet preset requirements; If the verification parameter does not satisfy the verification condition information, second verification result information is generated, and the second verification result information is used to characterize that the detection accuracy and detection stability of the detection model do not meet preset requirements.

9. The method according to claim 1, characterized in that: The method further comprises: Obtain the number and type of elements of the target element; Setting the number of output nodes of the detection model according to the number of elements; According to the element type, determining the spectral line overlap information and matrix effect information between the plurality of target elements in the original spectral data; The loss function of the detection model is set according to the spectral line overlap information and the matrix effect information.

10. A laser spectrum data processing system for analyzing uranium polymetallic ore elements, characterized in that: The system comprises: A spectrum acquisition module, used to acquire original spectrum data, wherein the original spectrum data is a two-dimensional array obtained by performing spectrum detection on a uranium polymetallic ore sample through a spectrum detection device; A detection module, used for inputting the raw spectral data into a detection model to obtain detection result data output by the detection model, wherein the detection model is a convolutional neural network model obtained by training using sample spectral data according to a dynamic uncertainty weighted loss function; the sample spectral data includes spectral data provided with concentration labels, and the concentration labels are provided according to experimental detection results; the detection result data includes concentration information of target elements; the target elements are one or more element combinations in the uranium polymetallic ore sample that will produce spectral line overlap and / or matrix effects; An experimental data acquisition module, used to acquire experimental result data, wherein the experimental result data is obtained by performing a concentration detection experiment on the uranium polymetallic ore sample corresponding to the original spectral data; A verification module is used to verify the detection model by comparing the detection result data with the experimental result data.

Images