Physical and data hybrid driven LIBS trace element detection method and system

Abstract

The application discloses a LIBS trace element detection method and system driven by physical and data hybrid, and belongs to the technical field of spectral analysis. The method comprises the following steps: acquiring plasma electron temperature and electron density of a target element in LIBS, identifying, reconstructing and interval spectrum optimizing a specific spectral line, fusing multi-modal information such as optimized LIBS data, plasma physical parameters, experimental setting parameters and observation conditions, constructing a hybrid quantitative model of physical mechanism and data driving by using a physical information neural network, embedding Saha-Boltzmann graphic information, combining ResNet-Transformer feature extraction and Bootstrap-Dropout uncertainty quantification, effectively overcoming interference such as LIBS matrix effect, spectral line aliasing and self-absorption, and significantly improving the precision, stability and generalization ability of trace element detection.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of spectral analysis, and particularly relates to a LIBS microelement detection method and system driven by physical and data mixing. BACKGROUND

[0002] Laser-induced breakdown spectroscopy (LIBS) is a rapid analysis technique based on atomic emission spectroscopy. It generates plasma by high-energy laser pulses and analyzes the characteristic spectrum emitted by elements during the cooling process of plasma to achieve qualitative and quantitative analysis of sample composition. This technique has wide application potential in environmental monitoring, geological exploration, industrial material analysis and biomedical fields due to its advantages such as no need for complex sample pretreatment, ability to realize simultaneous detection of multiple elements, fast analysis speed and small sample damage.

[0003] However, LIBS technology still faces significant challenges in the quantitative detection of trace elements, especially trace elements. Among them, the matrix effect is the core problem affecting the accuracy of analysis, which is manifested in physical matrix effect (such as differences in sample density, hardness, thermal conductivity affecting plasma formation and evolution) and chemical matrix effect (interaction between elements changes excitation and ionization behavior). In addition, factors such as laser energy fluctuation, environmental condition fluctuation and instrument noise also lead to poor spectral signal stability, low signal-to-noise ratio, and difficulty for traditional quantitative analysis methods to achieve accurate measurement.

[0004] Current mainstream LIBS quantitative methods mainly include three categories: first, free calibration method, which does not require standards but relies on strict local thermal equilibrium conditions, which is difficult to guarantee in practical applications; second, methods based on single-variable physical models, which are intuitive in principle but are easily affected by spectral interference and noise, and have poor adaptability to complex matrices; third, data-driven models based on multivariate statistics, which can extract multi-dimensional spectral information but are prone to overfitting, ill-conditioned matrix and other problems, have insufficient generalization ability, and require high computational resources.

[0005] Especially for rare earth elements and other trace elements, their characteristic spectral line intensity is weak, easily submerged or overlapped by major element spectral lines, and there is a self-absorption effect, which further increases the difficulty of accurate identification and quantification. Existing hybrid models attempt to combine physical mechanisms and data algorithms, but still have deficiencies in the depth of physical parameter coupling, intelligent processing of spectral lines, and multi-modal information fusion, which limit their precision and reliability in actual field detection.

[0006] Therefore, there is an urgent need to develop a high-precision, strong generalization LIBS quantitative analysis method that can deeply integrate plasma physical parameters, realize intelligent analysis of characteristic spectral lines, and effectively fuse multi-modal data to meet the urgent needs of trace element detection in complex actual samples. SUMMARY

[0007] To solve the above technical problems, the present application provides a physical and data hybrid driven LIBS trace element detection method and system, aiming to improve the quantitative analysis accuracy and reliability, improve the model generalization ability, and reduce the calculation cost, so as to make it suitable for practical application scenarios.

[0008] To achieve the above object, the technical scheme adopted by the present application is as follows:

[0009] A physical and data hybrid driven LIBS trace element detection method, the method comprising:

[0010] Step 1, obtaining the plasma physical state parameters of the characteristic spectral lines of the elements to be analyzed in laser-induced breakdown spectroscopy (LIBS);

[0011] Step 2, identifying the characteristic spectral lines of the trace elements, and reconstructing the spectral lines with self-absorption or aliasing, and optimizing the interval LIBS spectrum containing the target elements;

[0012] Step 3, fusing the optimized LIBS spectral data, plasma physical state parameters, experimental parameters and observation conditions to construct a hybrid quantitative model of physical mechanism and data driving, and realizing the quantitative detection of trace elements, wherein the hybrid quantitative model adopts a physical information neural network regression model.

[0013] On the other hand, the present application provides a physical and data hybrid driven LIBS trace element detection system, comprising:

[0014] A physical parameter calculation module for obtaining the plasma physical state parameters of the characteristic spectral lines of the elements to be analyzed in laser-induced breakdown spectroscopy (LIBS);

[0015] A spectral line identification and reconstruction module for identifying the characteristic spectral lines of the trace elements, and reconstructing the spectral lines with self-absorption or aliasing, and optimizing the interval LIBS spectrum containing the target elements;

[0016] A data fusion and quantitative analysis module for fusing the optimized LIBS spectral data, plasma physical state parameters, experimental parameters and observation conditions to construct a hybrid quantitative model of physical mechanism and data driving, and realizing the quantitative detection of trace elements, wherein the hybrid quantitative model adopts a physical information neural network regression model.

[0017] In a third aspect, the present application provides an electronic device comprising: one or more processors; a memory for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned physical and data hybrid driven LIBS trace element detection method.

[0018] In a fourth aspect, the present application provides a computer readable storage medium, which stores executable instructions, and the instructions enable a processor to implement the aforementioned LIBS trace element detection method driven by a combination of physics and data when executed by the processor.

[0019] The present application has the following beneficial effects:

[0020] Improving quantitative analysis accuracy: By introducing plasma electron temperature and electron density as physical weight factors, the contribution of trace element characteristic spectral lines is effectively highlighted, and the matrix effect and background interference are significantly suppressed, making the detection results closer to the true content.

[0021] Enhancing spectral line recognition and anti-interference ability: Using intelligent spectral line recognition and reconstruction technology, combined with overlapping peak analysis and interval spectrum optimization, the problem of difficult recognition caused by trace element characteristic spectral lines being submerged, aliasing and self-absorption effect of major elements is effectively solved.

[0022] Optimizing model performance and practicality: Through multi-modal information fusion and physical information neural network (PINN) hybrid architecture, the model generalization ability and stability are greatly improved while maintaining the rationality of the physical mechanism, and the BootstrapDropout quantifies the uncertainty to realize high confidence prediction and model self-iterative optimization.

[0023] Promoting technology practicality: The method effectively controls the calculation cost while ensuring high precision, significantly improves the detection reliability of LIBS technology for trace elements in complex actual scenarios, and provides a feasible solution for on-site rapid analysis in environmental, geological and industrial fields. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of the LIBS trace element detection method driven by a combination of physics and data according to the present application;

[0025] Figure 2 The spectral line of the cerium element at a wavelength of 413.4 nm in the LIBS spectrum of the soil sample according to the present application. DETAILED DESCRIPTION

[0026] The present application will be further described below in conjunction with the drawings and examples.

[0027] As Figure 1As shown, the present application provides a physical and data hybrid driven LIBS trace element detection method and system, which solves the problems of low signal-to-noise ratio, spectral line interference, and insufficient physical parameter coupling in the existing LIBS technology in trace element detection, and proposes a technical solution integrating deep coupling of physical parameters, intelligent processing of spectral lines, and multi-modal data fusion. The method specifically comprises:

[0028] Step 1, obtaining the plasma physical state parameters of the characteristic spectral line of the element to be analyzed, including plasma electron temperature and electron density.

[0029] The Boltzmann and Saha-Boltzmann are used to describe the plasma temperature, the Stark broadening and the Saha equation are used to calculate the electron density, the average spectrum of LIBS is continuously measured at each position, and each spectrum is measured and described, the physical state parameters of the characteristic spectral line of the element to be analyzed are obtained, the electron temperature T e and the electron density n e of the plasma are accurately calculated, and they are embedded as weight factors in the quantitative model to strengthen the signal contribution of trace elements and lay a physical foundation for quantitative analysis.

[0030] Plasma electron temperature calculation: assuming that the laser-induced plasma meets the local thermodynamic equilibrium condition, the particle number distribution of each energy level satisfies the Boltzmann law, and the electron temperature is approximately equal to the excitation temperature. When the electron energy level p→q transitions, the radiation intensity I can be represented as:

[0031] (1)

[0032] In the formula, I represents the intensity of the electron radiation spectrum line, N0 represents the ground state particle number density, g0 and represent the statistical weights of the ground state and the p energy level, E p represents the excitation energy of the upper energy level (p energy level), represents the Boltzmann constant, T represents the electron temperature, A represents the proportion of the number of atoms that transition from the p energy level to the q energy level and radiate photons to the total number of atoms in the p energy level, h represents the Planck constant, and v represents the frequency of the radiation light. Replace the frequency with wavelength, and take the natural logarithm of the above formula (1):

[0033] (2)

[0034] In the formula, , , and respectively represent the intensity of the spectral line, the statistical weight of the upper energy level, the wavelength, and the transition probability; and respectively represent the excitation energy and the electron temperature, All the constants in equation (1) are included in the constant C. The oscillator strength is replaced by .

[0035] (3)

[0036] The spectral line of a specific element radiates different wavelengths, and is linearly related to with a slope of , so that the electron temperature .

[0037] Plasma electron density calculation: Electron density reflects the characteristics of the plasma environment and is an important basis for verifying the thermodynamic equilibrium condition. When the plasma state generated by laser incidence on the target satisfies the local thermodynamic equilibrium condition:

[0038] (4)

[0039] In the formula, is the full width at half maximum of the spectral line, is the electron collision parameter, is the electron density. Thus, according to the formula, the electron number density can be measured using the spectral line width. For LIBS plasma with high plasma density, preferentially select the spectral line dominated by Stark broadening and other broadening (such as Doppler broadening, instrument broadening) has little effect. The measured spectral line width of the target spectrum can be fitted by a Gaussian function, a Lorentz function, or a Voigt function.

[0040] When the plasma broadening is treated as a Gaussian function, the actual spectral line broadening is represented as:

[0041] (5)

[0042] When the instrument broadening is approximated as a Lorentz function, the actual spectral line broadening is:

[0043] (6)

[0044] In the formula, is the instrument broadening, which can be obtained by fitting the narrow spectrum of a standard light source before the experiment; is the Stark broadening, which can be fitted according to the spectral line profile shape using a Gaussian function or a Lorentz function; when the Stark broadening (Lorentz type) and the Doppler broadening (Gaussian type) exist simultaneously, the spectral line profile can be fitted using a Voigt function (convolution of Gaussian and Lorentz functions) to separate the Gaussian type (Doppler broadening) and Lorentz type (Stark broadening) broadening components.

[0045] The calculated electron temperature Te and electron density n e As a weight factor embedded in the mixed quantitative model, it enhances the physical signal contribution of trace element characteristic spectral lines, reduces background interference, and improves the accuracy of quantitative analysis.

[0046] Step 2, identify the characteristic spectral lines of trace elements, and reconstruct the spectral lines with self-absorption or overlap to optimize the interval LIBS spectrum.

[0047] Separately analyze the characteristic spectral lines with low content / weak signal in low signal-to-noise ratio LIBS spectrum to ensure the accuracy of element physical characteristics and provide reliable information for quantitative analysis. Gaussian, Lorentzian, and Voigt functions are used for low signal-to-noise ratio spectral modeling, and Gaussian function is used to describe Doppler broadening in low-pressure environment.

[0048] Spectral resolution (i.e. overlapping / self-absorption spectral peak reconstruction): refer to the NIST atomic spectrum database or self-built reference data of characteristic spectral lines of typical elements to pre-judge the wavelength position of the characteristic spectral lines of the elements to be analyzed. For characteristic spectral lines with self-absorption and overlap, targeted identification and reconstruction optimization are adopted. The physical characteristics of the spectral lines can be described by spectral peak reconstruction, or the physical characteristics of the spectral lines can be inferred by referring to the physical characteristics of the internal standard elements. By mathematical analysis method, the spectral overlap is decomposed, the initial value (number of overlapping peaks and relative intensity) of curve fitting is assumed by means of fractional differentiation theory, Gaussian function or Lorentzian model and Levenberg-Marquart method are applied to curve fitting, and LIBS signal overlapping peak analysis is realized.

[0049] Interval LIBS spectrum optimization: refer to the prior information of typical trace elements, extract the interval LIBS spectrum covering the elements to be analyzed from the complete LIBS spectrum, and analyze and reconstruct the spectral lines with overlap or self-absorption, and optimize the interval LIBS spectrum.

[0050] Step 3, fuse the optimized LIBS spectral data, plasma physical state parameters, experimental parameters and observation conditions to construct a hybrid quantitative model of physical mechanism and data driving, and realize the quantitative detection of trace elements.

[0051] Input layer construction: use the optimized interval LIBS spectrum, introduce experimental setting parameters, environmental parameters, characteristic spectral line physical parameters and other additional information to form a generalized spectrum, including: fusion of optimized interval LIBS spectrum, experimental setting parameters (laser energy, focusing distance), environmental parameters (temperature, humidity, pressure), physical parameters (T e , n e ) four-dimensional data.

[0052] PINN hybrid architecture: In the hybrid quantitative model, a physical information neural network (PINN) regression model is used, Saha-Boltzmann graphical information is embedded in the physical mechanism, the physical rationality of parameters such as electron temperature and particle density is constrained, ResNet-Transformer feature extraction is adopted in mathematical driving, and the collaborative optimization of multi-modal LIBS and element concentration is realized through feature fusion.

[0053] Uncertainty quantification: Bootstrap-Dropout is used to dynamically evaluate the prediction confidence, and only high-confidence results are retained for model iteration update, so that the residual error of the objective function is minimized. The Bootstrap-Dropout method is an uncertainty quantification technique combining Bootstrap resampling and Dropout regularization. In model training, part of the neurons are randomly discarded to prevent overfitting; in the prediction stage, multiple prediction results are obtained by sampling different neuron subsets multiple times, and the variance or confidence interval is calculated to evaluate the prediction uncertainty, and high-confidence results are selected for model iteration optimization.

[0054] Embodiment 1: Quantitative detection of rare earth element neodymium (Nd).

[0055] Experimental setup: A pulsed laser with a wavelength of 1064 nm and a pulse width of 10 ns is used, the laser energy is 100 mJ, focused on the surface of a rock sample containing neodymium element, the environmental pressure is 1 standard atmosphere, and the temperature is 25℃. A spectrometer is used to collect LIBS spectrum, the spectral range is 200-800 nm, and the resolution is 0.1 nm.

[0056] Step 1, plasma physical parameter calculation.

[0057] Electron temperature calculation: Select multiple characteristic spectral lines of neodymium element, such as wavelength 492.5 nm, 535.0 nm, 585.2 nm, obtain the intensity I, wavelength λ, upper energy level statistical weight g, transition probability A of each spectral line. According to formula (3), the value of each spectral line and the corresponding excitation energy E p are calculated, linear fitting is performed, the slope is−0.625 / T e , and the electron temperature T e is calculated to be about 8000K.

[0058] Electron density calculation: Select the spectral line of neodymium element with wavelength 585.2 nm, measure its spectral line broadening, use Gaussian function, according to formula (5), the instrument broadening =0.05nm is known, the measured =0.2nm, the Stark broadening = 0.3 nm. According to the relationship between the Stark broadening and the electron density, the electron density was calculated to be about 1.0 x 1017cm-3. −3 .

[0059] Step 2, characteristic spectral line identification, reconstruction and optimization.

[0060] According to the NIST atomic spectral database, the wavelength position of the characteristic spectral line of neodymium element was determined, and it was found that the spectral line at a wavelength of 492.5 nm was overlapped with the spectral line of iron element.

[0061] The fractional differential theory was used to assume the initial value of curve fitting, assuming that the number of overlapping peaks was 2 and the relative intensity was 3:1. The Levenberg-Marquart method combined with Gaussian function was used to fit the overlapped spectral line, to realize the decomposition and reconstruction of the spectral line, and to obtain the accurate shape and intensity of the characteristic spectral line of neodymium element.

[0062] Interval LIBS spectrum optimization: from the complete LIBS spectrum, the interval LIBS spectrum covering the characteristic spectral line (480-520 nm) of neodymium element was intercepted, and the reconstructed spectral line was analyzed to optimize the interval LIBS spectrum.

[0063] Step 3, multi-modal information fusion and mixed model quantification.

[0064] Additional information such as experimental setup parameters (laser energy, wavelength, pulse width), environmental parameters (temperature, air pressure), characteristic spectral line physical parameters (electron temperature, electron density) was introduced, and the optimized interval LIBS spectrum formed a generalized spectrum.

[0065] The electron temperature and electron density were used as weighted factors in multivariate analysis, and were input into the physical information neural network (PINN) regression model. The model embedded Saha-Boltzmann graphical information in the physical mechanism, and constrained the physical rationality of electron temperature and particle number density; in the mathematical driving, ResNet-Transformer feature extraction was used, and through feature fusion, the multi-modal LIBS and the concentration of neodymium element were cooperatively optimized.

[0066] The BootstrapDropout method was introduced to quantify the uncertainty of the model, and the high confidence inversion results were dynamically selected for model self-iterative update to minimize the residual error of the objective function. After the model was trained, the R² of the validation set was 0.96, and the RSD was 5.2%.

[0067] Example 2: Quantitative detection of cerium (Ce) element in soil.

[0068] Experimental setup: The pulsed laser with wavelength 1064 nm, pulse width 4 ns, frequency 3 Hz, laser energy 21 mJ, was focused on the surface of the soil sample containing the element cerium, the ambient pressure was 0.8 standard atmosphere (low pressure environment), and the temperature was 20℃. The LIBS spectrum was collected by a spectrometer, the spectral range was 240-840 nm, and the resolution was 0.08 nm.

[0069] Step 1, calculation of plasma physical parameters:

[0070] Calculation of electron temperature: The characteristic spectral lines of the element cerium at wavelengths 369.7 nm, 413.4 nm and 456.2 nm were selected, and the electron temperature was calculated according to the method of Example 1, and T e was about 7500 K.

[0071] Calculation of electron density: In a low pressure environment, the instrument broadening was approximately treated as a Lorentz function, according to formula (6), the measured = 0.22 nm, the instrument broadening = 0.08 nm, the calculated Stark broadening = 0.14 nm, and the electron density was about 8.0 x 1016 cm −3 .

[0072] Step 2, identification and reconstruction and optimization of characteristic spectral lines: The spectral line at wavelength 413.4 nm of the element cerium had a mixed spectral peak phenomenon (as shown in Figure 2 ), and the spectral peak reconstruction method was used, combined with the Lorentzian model and the Levenberg-Marquart method for curve fitting, to solve the mixing problem.

[0073] Interval LIBS spectrum optimization: The interval LIBS spectrum covering the characteristic spectral line (400-430 nm) of the element cerium was intercepted, and the reconstructed spectral line was optimized.

[0074] Step 3, multi-modal information fusion and mixed model quantification: According to the method of Example 1, multi-modal information fusion was carried out, and the electron temperature and electron density were input as weighted factors into the PINN regression model for quantitative analysis. The content of the element cerium in the soil sample was 18.3 ppm, and the relative error with the actual value 18.8 ppm was 2.7%.

[0075] On the other hand, the present application provides a LIBS trace element quantitative analysis system driven by physical and data mixing, which comprises various modules capable of realizing each step of the aforementioned method, specifically comprising:

[0076] A physical parameter calculation module for obtaining the plasma physical state parameters of the characteristic spectral line of the element to be analyzed;

[0077] A spectral line recognition and reconstruction module is configured to recognize characteristic spectral lines of trace elements and reconstruct spectral lines with self-absorption or aliasing, and optimize the LIBS spectrum in a range;

[0078] A data fusion and quantitative analysis module is configured to fuse the optimized LIBS data, plasma physical state parameters, experimental parameters and observation conditions, construct a hybrid quantitative model of physical mechanism and data driving, and realize quantitative detection of trace elements, wherein the hybrid quantitative model adopts a physical information neural network regression model.

[0079] In a third aspect, the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned physical and data hybrid driving LIBS trace element quantitative analysis method.

[0080] In a fourth aspect, the present application provides a computer readable storage medium having stored executable instructions, which when executed by a processor, enable the processor to implement the aforementioned physical and data hybrid driving LIBS trace element quantitative analysis method.

[0081] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A physical and data hybrid driven LIBS trace element detection method, characterized in that, The method comprises: Step 1, obtaining plasma physical state parameters of characteristic spectral lines of elements to be analyzed in laser-induced breakdown spectroscopy (LIBS), the parameters comprising plasma electron temperature and electron density; Step 2, identifying characteristic spectral lines of trace elements, and reconstructing spectral lines with self-absorption or overlap to optimize interval LIBS spectrum containing target elements; wherein: Gaussian function, Lorentz function or Voigt function are used to describe characteristic spectral lines, and atomic spectrum database is referred to for pre-judgment of spectral line wavelength position, and for characteristic spectral lines with self-absorption and overlap, spectral peak reconstruction or reference to internal standard element physical characteristics is used to deduce physical characteristics, and a mathematical analysis method combining fractional differential theory is used for overlapping peak analysis; The method for optimizing interval LIBS spectrum containing target elements is: according to prior knowledge of elements to be analyzed, a specific wavelength interval covering the characteristic spectral lines of the elements is intercepted from the complete LIBS spectrum, and spectral lines with overlap or self-absorption in the interval are analyzed and reconstructed; Step 3, fusing optimized LIBS spectrum data, plasma physical state parameters, experimental parameters and observation conditions to construct a hybrid quantitative model of physical mechanism and data driving, and realizing quantitative detection of trace elements, wherein the hybrid quantitative model adopts a physical information neural network regression model. 2.The physical and data hybrid driven LIBS trace element detection method according to claim 1, wherein, The electron temperature is calculated by establishing a linear relationship between the spectral parameters of multiple characteristic spectral lines of a specific element based on the local thermodynamic equilibrium assumption and the Boltzmann law; and the electron density is calculated by selecting corresponding Gaussian function or Lorentz function according to the spectral line broadening curve shape based on the Stark broadening effect. 3.The physical and data hybrid driven LIBS trace element detection method according to claim 1, wherein, In step 3, the fusion of optimized LIBS spectrum data, plasma physical state parameters, experimental parameters and observation conditions comprises using optimized interval LIBS spectrum data to introduce characteristic spectral line physical parameters, experimental setting parameters and environmental parameters to form a generalized spectrum. 4.The physical and data hybrid driven LIBS trace element detection method according to claim 1, wherein, In step 3, the physical information neural network regression model embeds Saha-Boltzmann graphic information at the physical mechanism level to constrain the physical rationality of plasma parameters, and adopts ResNet-Transformer architecture for feature extraction and fusion at the data driving level; The Bootstrap-Dropout method is introduced to quantify the uncertainty of model prediction and dynamically select high-confidence inversion results for self-iterative updating of the model, wherein the Bootstrap-Dropout method represents an uncertainty quantification technology combining Bootstrap resampling and Dropout regularization.

5. A physical and data hybrid driven LIBS trace element detection system, applied to the method of any one of claims 1-4, characterized in that, It comprises: A physical parameter calculation module for obtaining plasma physical state parameters of characteristic spectral lines of elements to be analyzed in laser-induced breakdown spectroscopy (LIBS); A spectral line identification and reconstruction module for identifying characteristic spectral lines of trace elements, and reconstructing spectral lines with self-absorption or overlap to optimize interval LIBS spectrum containing target elements; A data fusion and quantitative analysis module is configured to fuse the optimized LIBS spectral data, plasma physical state parameters, experimental parameters and observation conditions, construct a hybrid quantitative model combining physical mechanism and data driving, and realize quantitative detection of trace elements, wherein the hybrid quantitative model adopts a physical information neural network regression model.

6. An electronic device, comprising: Comprise: one or more processors; a memory for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the physical and data hybrid driven LIBS trace element detection method of any one of claims 1-4.

7. A computer readable storage medium characterized in that, A computer readable storage medium having stored thereon executable instructions that, when executed by a processor, enable the processor to implement the physical and data hybrid driven LIBS trace element detection method of any one of claims 1-4.

Images