A method and device for decomposing a LIBS spectral signal

Abstract

The application discloses a LIBS spectrum signal decomposition method, relates to the technical field of spectrum analysis and chemical detection, and reduces the calculation complexity by intelligently dividing a full spectrum into a plurality of sub-regions with independent spectral line characteristics. Then, in each region, an improved pseudo-Vogit model with multiple physical constraints is used for decomposition and optimization to ensure the physical reliability of the result. Finally, the result is self-improved through a residual analysis mechanism, and significant improvement is realized in calculation efficiency, decomposition precision, result reliability and automation degree.

Description

Technical Field

[0001] This application relates to the fields of spectral analysis and chemical detection technology, specifically to a LIBS spectral signal decomposition method and apparatus. Background Technology

[0002] Laser-induced breakdown spectroscopy (LIBS), as a highly efficient method for material composition analysis, has been widely applied in metallurgy, environmental monitoring, space exploration, and other fields. Its basic principle is to excite a sample with a high-energy laser pulse to generate plasma, and then analyze the characteristic spectra of atoms or ions emitted by the plasma, thereby achieving qualitative and quantitative detection of elements.

[0003] However, LIBS technology faces a long-standing core challenge in practical applications: spectral overlap. In the high-temperature, high-density plasma environment, multiple elements in a sample simultaneously emit a large number of spectral lines. These lines exhibit a certain width due to physical broadening mechanisms, typically appearing as Gaussian, Lorentz, or a convolution of both—the Vogit line. When the center wavelengths of the emission lines from different elements are close, and their spacing is less than the broadening width of the lines themselves or the resolution limit of the spectrometer, the signals from these lines will overlap, forming complex overlapping peaks. Spectral overlap severely interferes with the accurate identification and intensity extraction of characteristic spectral lines, and is one of the fundamental reasons for increased errors in LIBS quantitative analysis and reduced reliability of plasma physical parameter (such as temperature and electron density) inversion.

[0004] To overcome the problem of spectral overlap, existing technologies mainly seek solutions in two directions: one is to improve the hardware resolution of the spectrometer, but this significantly increases system costs and has limited effectiveness in dealing with inherent physical broadening; the other is to use spectral decomposition algorithms to mathematically analyze the acquired overlapping spectral signals and separate the individual spectral line components from the mixed signal. Among the many decomposition algorithms, the fitting method based on the Vogit line shape model is a relatively common one.

[0005] However, those skilled in the art recognize that existing spectral decomposition methods, especially traditional global Vogit fitting, have the following significant drawbacks:

[0006] 1. High computational complexity and poor stability: Traditional methods typically perform global modeling over the entire spectral range or a large fixed window, requiring simultaneous optimization of numerous parameters (peak position, intensity, broadening, etc. for each spectral component). This results in a heavy computational burden, and the large parameter space makes the optimization algorithm prone to getting trapped in local optima. The fitting results are extremely sensitive to initial value guesses, leading to insufficient stability.

[0007] 2. Lack of physical constraints and weak interpretability: Existing methods often simply pursue mathematical goodness of fit (such as least squares error) without incorporating the physical principles of spectral generation as constraints into the modeling process. This may lead to physically unreasonable fitted spectral parameters (such as negative intensity and non-physical broadening), reducing the reliability and interpretability of the analytical results.

[0008] Therefore, to overcome the limitations of existing technologies, there is an urgent need to develop a novel spectral decomposition method that is computationally efficient, robust, reliable, and possesses rigorous physical interpretability. This method not only needs to achieve rapid convergence and low computational complexity at the algorithmic level to meet the online and real-time detection requirements of LIBS, but also must establish a strong constraint relationship between the mathematical model and the physical mechanism. This ensures that each fitting parameter (such as peak position, intensity, and broadening) has a clear physical meaning, thereby avoiding mathematical overfitting and physically unsolvable outputs. Only in this way can the analytical capability of LIBS for complex overlapping spectra be fundamentally improved, providing a more accurate and reliable data foundation for elemental quantitative analysis, plasma diagnostics, and material characterization. This will further propel LIBS technology from semi-quantitative to high-precision quantitative analysis, expanding its practical application depth and breadth in key fields such as high-end manufacturing, environmental monitoring, and space exploration. Summary of the Invention

[0009] The purpose of this application is to provide a LIBS spectral signal decomposition method and apparatus, which solves the problems of existing laser-induced breakdown spectroscopy (LIBS) decomposition technology, especially the traditional global Vogit fitting method, such as high computational complexity, heavy dependence on initial parameters, poor interpretability and low reliability due to lack of physical constraints.

[0010] This application is achieved through the following technical solution:

[0011] The first aspect of this application provides a LIBS spectral signal decomposition method, comprising:

[0012] The LIBS spectral signal to be decomposed is acquired, and the region of interest is divided into multiple regions of interest.

[0013] Sub-peak detection is performed on the spectral signal in the region of interest, and at the same time, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest.

[0014] The region of interest after preprocessing is subjected to fitting residual analysis to obtain the analysis results. Based on the analysis results, the region of interest after preprocessing is iteratively optimized using a residual completion mechanism to obtain the LIBS spectral signal decomposition results.

[0015] In one possible implementation, the LIBS spectral signal to be decomposed is acquired, and the LIBS spectral signal to be decomposed is divided into regions of interest to obtain multiple regions of interest, including:

[0016] The LIBS spectral signal to be decomposed is acquired, and Gaussian smoothing is applied to the acquired LIBS spectral signal to obtain the denoised LIBS spectral signal.

[0017] Identify all local maxima of the denoised LIBS spectral signal to obtain peak candidate points;

[0018] Based on preset absolute intensity thresholds and relative significance thresholds, the candidate peak points are screened to obtain significant peak anchor points;

[0019] Based on the wavelength distance between significant peak anchor points, the significant peak anchor points are clustered to obtain at least one peak cluster;

[0020] The initial region is determined by the adjacent local minimum or energy integral boundary of each peak cluster, and the initial region is merged or segmented and optimized according to the maximum region width and minimum region spacing criteria to output multiple regions of interest.

[0021] In one possible implementation, the significant peak anchors are clustered based on the wavelength distance between them to obtain at least one peak cluster, including:

[0022] Determine the peak spacing threshold. When the wavelength distance between significant peak anchor points is less than the peak spacing threshold, the significant peak anchor points are clustered into the same peak cluster, thus obtaining at least one peak cluster.

[0023] In one possible implementation, an initial region is determined by the adjacent local minimums or energy integral boundaries of each peak cluster, and the initial region is optimized by merging or segmenting according to the maximum region width and minimum region spacing criteria, outputting multiple regions of interest, including:

[0024] The left and right boundaries of each peak cluster are determined by the adjacent local minimum points, thus obtaining the initial boundary corresponding to each peak cluster;

[0025] If the spectral energies at both ends of the initial boundary are higher than the preset energy threshold, the window formed by the initial boundary is expanded using the energy integration criterion to ensure complete coverage of spectral line information and obtain the initial region.

[0026] If the width of the initial region corresponding to a single peak cluster exceeds the maximum allowable value, it is divided into multiple sub-regions according to the valley position. If the distance between adjacent initial regions is less than a small threshold, they are merged into one region, and multiple regions of interest are output.

[0027] In one possible implementation, sub-peak detection is performed on the spectral signal in the region of interest, and simultaneously, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in a preprocessed region of interest, including:

[0028] Identify the main peak, right shoulder peak, and left shoulder peak of the spectral signal in the region of interest, and use the main peak, right shoulder peak, and left shoulder peak together as the seed peak of the region of interest;

[0029] Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a physically constrained pseudo-Vogit line shape, thus obtaining the preprocessed region of interest.

[0030] In one possible implementation, the seed peak is used as the decomposition starting point, and a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in a preprocessed region of interest, including:

[0031] Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a pseudo-Vogit line shape.

[0032] Non-negative least squares constraints are applied to each component to ensure physical non-negativity and suppress overfitting, while minimizing the residuals to obtain the constrained components.

[0033] For the constrained components, determine the fitting optimization change corresponding to the component to be removed, and eliminate the components whose goodness of fit change is less than the preset change threshold to obtain the preprocessed region of interest.

[0034] In one possible implementation, a fitting residual analysis is performed on the preprocessed region of interest to obtain the analysis results, including:

[0035] For the region of interest after preprocessing, a null hypothesis and an alternative hypothesis are established; wherein, the null hypothesis is the hypothesis that the region of interest after preprocessing has no missing Vogit spectral lines, and the alternative hypothesis is the hypothesis that the region of interest after preprocessing contains Vogit spectral lines.

[0036] Obtain the residual sequence corresponding to the region of interest after preprocessing. If the residual data in the residual sequence does not satisfy the null hypothesis but satisfies the alternative hypothesis, the analysis result is determined to be that the residual contains Vogit spectral lines; otherwise, the analysis result is determined to be that the residual has no missing Vogit spectral lines.

[0037] In one possible implementation, based on the analysis results, a residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain LIBS spectral signal decomposition results, including:

[0038] If the analysis result shows that the residual contains Vogit spectral lines, the NIST atomic emission spectrometry database is queried to confirm whether there are real element spectral lines at the wavelength positions corresponding to the residuals. If so, the residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition result. Otherwise, it is determined to be a false positive signal caused by noise and is removed.

[0039] In one possible implementation, a residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition results, including:

[0040] New pseudo-Vogit components are added at the wavelength positions corresponding to the residuals, and then substituted back into the preprocessed region of interest for full fitting. The parameters of all components are optimized, and the decomposition results are updated until the residuals have no significant Vogit structure, thus obtaining the LIBS spectral signal decomposition results.

[0041] A second aspect of this application provides a LIBS spectral signal decomposition apparatus, comprising:

[0042] The ROI extraction module is used to acquire the LIBS spectral signal to be decomposed and to divide the LIBS spectral signal to be decomposed into regions of interest to obtain multiple regions of interest.

[0043] The signal decomposition module is used to perform sub-peak detection on the spectral signal in the region of interest, and simultaneously use a physically constrained pseudo-Vogit line shape to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest.

[0044] The residual completion module is used to perform fitting residual analysis on the preprocessed region of interest, obtain the analysis results, and iteratively optimize the preprocessed region of interest using a residual completion mechanism based on the analysis results to obtain the LIBS spectral signal decomposition results.

[0045] Compared with the prior art, this application has the following advantages and beneficial effects:

[0046] This application provides a LIBS spectral signal decomposition method, which reduces computational complexity by intelligently segmenting the entire spectrum into multiple sub-regions with independent spectral features. Subsequently, within each region, an improved pseudo-Vogit model incorporating multiple physical constraints is used for decomposition and optimization to ensure the physical reliability of the results. Finally, the results are self-improved through a residual analysis mechanism, achieving significant improvements in computational efficiency, decomposition accuracy, result reliability, and automation. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0048] Figure 1 A flowchart illustrating a LIBS spectral signal decomposition method provided in an embodiment of this application;

[0049] Figure 2 This is a schematic diagram of the structure of a LIBS spectral signal decomposition device provided in an embodiment of this application;

[0050] The attached diagram shows the markings and corresponding component names:

[0051] 201 - ROI extraction module, 202 - Signal decomposition module, 203 - Residual compensation module. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0053] To address the problems of high computational complexity, heavy reliance on initial parameters, poor interpretability, and low reliability in existing laser-induced breakdown spectroscopy (LIBS) decomposition techniques, especially the traditional global Vogit fitting method, this application aims to provide a computationally efficient, robust, reliable, and physically interpretable automatic spectral decomposition method and system. This application can intelligently handle severely overlapping peaks in complex matrix LIBS spectra, fundamentally improving the accuracy and efficiency of spectral analysis and laying a reliable foundation for subsequent precise quantitative analysis and plasma physics parameter inversion.

[0054] like Figure 1 As shown, this application provides a LIBS spectral signal decomposition method, including:

[0055] S101. Obtain the LIBS spectral signal to be decomposed, and divide the LIBS spectral signal to be decomposed into regions of interest to obtain multiple regions of interest;

[0056] S102. Sub-peak detection is performed on the spectral signal in the region of interest, and at the same time, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest.

[0057] S103. Perform fitting residual analysis on the preprocessed region of interest to obtain the analysis results. Based on the analysis results, use a residual completion mechanism to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition results.

[0058] Compared with existing spectral decomposition techniques such as traditional global Vogit fitting, this application significantly improves computational efficiency, decomposition accuracy, result reliability, and automation by introducing an adaptive ROI partitioning and physical constraint mechanism. The effectiveness of the spectral decomposition method was verified by decomposing and analyzing the spectral data of steel alloy samples with different concentrations. Experimental results show that the decomposition point finding mechanism can locate most emission positions, achieving a goodness of fit of over 0.95 in the concentrated interval of the simulated data. The residual completion mechanism can find more data components, achieving a goodness of fit of over 0.9 for fitting a single component. In actual spectral decomposition, the goodness of fit for most spectra is above 0.95 through the action of both mechanisms. Simultaneously, the decomposition can also identify highly correlated components. The decomposed spectra not only highly match the true values ​​in spectral position but also show significant improvement in the quantitative analysis of plasma temperature and elemental concentration, with the goodness of fit increasing from a minimum of 0.8875 to a minimum of 0.9065 while maintaining consistent temperature changes. Finally, in the quantitative analysis, the Ri of Mn and Cr... 2 The values ​​also improved from 0.9397 and 0.9728 to 0.9978 and 0.9929, respectively, while the RMSE decreased to 0.0334 and 0.0171, respectively. This demonstrates the effectiveness of the decomposition scheme.

[0059] This application relates to a Vogit spectral decomposition method based on adaptive ROI (Region of Interest) partitioning and physical constraints, used for high-precision and high-efficiency automatic analysis of overlapping peaks in LIBS spectra. This method is particularly suitable for solving spectral line interference problems in LIBS spectra of complex matrices such as steel, alloys, soil, and environmental samples, aiming to improve the detection limits, quantitative analysis accuracy, and reliability of plasma physics parameter inversion for constant elements. It can be widely applied in metallurgical process control, material composition identification, environmental monitoring, and scientific research analysis.

[0060] In one possible implementation, the LIBS spectral signal to be decomposed is acquired, and the LIBS spectral signal to be decomposed is divided into regions of interest to obtain multiple regions of interest, including:

[0061] The LIBS spectral signal to be decomposed is acquired, and Gaussian smoothing is applied to the acquired LIBS spectral signal to obtain the denoised LIBS spectral signal.

[0062] Identify all local maxima of the denoised LIBS spectral signal to obtain peak candidate points;

[0063] Based on preset absolute intensity thresholds and relative significance thresholds, the candidate peak points are screened to obtain significant peak anchor points;

[0064] Based on the wavelength distance between significant peak anchor points, the significant peak anchor points are clustered to obtain at least one peak cluster;

[0065] The initial region is determined by the adjacent local minimum or energy integral boundary of each peak cluster, and the initial region is merged or segmented and optimized according to the maximum region width and minimum region spacing criteria to output multiple regions of interest.

[0066] In one possible implementation, the significant peak anchors are clustered based on the wavelength distance between them to obtain at least one peak cluster, including:

[0067] Determine the peak spacing threshold. When the wavelength distance between significant peak anchor points is less than the peak spacing threshold, the significant peak anchor points are clustered into the same peak cluster, thus obtaining at least one peak cluster.

[0068] In one possible implementation, an initial region is determined by the adjacent local minimums or energy integral boundaries of each peak cluster, and the initial region is optimized by merging or segmenting according to the maximum region width and minimum region spacing criteria, outputting multiple regions of interest, including:

[0069] The left and right boundaries of each peak cluster are determined by the adjacent local minimum points, thus obtaining the initial boundary corresponding to each peak cluster;

[0070] If the spectral energies at both ends of the initial boundary are higher than the preset energy threshold, the window formed by the initial boundary is expanded using the energy integration criterion to ensure complete coverage of spectral line information and obtain the initial region.

[0071] If the width of the initial region corresponding to a single peak cluster exceeds the maximum allowable value, the region is divided into multiple sub-regions based on the valley bottom position. If the distance between adjacent initial regions is less than a small threshold, they are merged into one region, resulting in multiple regions of interest. The valley bottom position is the point of minimum intensity within that region, and this point is used as the dividing boundary to segment the original region into two independent sub-regions.

[0072] In one possible implementation, sub-peak detection is performed on the spectral signal in the region of interest, and simultaneously, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in a preprocessed region of interest, including:

[0073] Identify the main peak, right shoulder peak, and left shoulder peak of the spectral signal in the region of interest, and use the main peak, right shoulder peak, and left shoulder peak together as the seed peak of the region of interest;

[0074] Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a physically constrained pseudo-Vogit line shape, thus obtaining the preprocessed region of interest.

[0075] In one possible implementation, the seed peak is used as the decomposition starting point, and a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in a preprocessed region of interest, including:

[0076] Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a pseudo-Vogit line shape.

[0077] Non-negative least squares constraints are applied to each component to ensure physical non-negativity and suppress overfitting, while minimizing the residuals to obtain the constrained components. For example, a new pseudo-Vogit component is added at the wavelength position corresponding to the residual. This component is added as a new seed peak to the original component set and re-substituted into the preprocessed region of interest for full fitting. The parameters of all components are optimized using the non-negative least squares method, and the decomposition results are updated. The above residual analysis and supplementary fitting process is repeated until the maximum amplitude of the residual is lower than the preset threshold or there is no obvious Vogit linear structure in the residual sequence. At this point, it is determined that the residual has no significant Vogit structure, and the constrained components are obtained.

[0078] For the constrained components, determine the corresponding goodness-of-fit optimization change for the removed components, and discard components whose goodness-of-fit change is less than a preset threshold to obtain the preprocessed region of interest. For example, calculate the goodness-of-fit change before and after removing each component, and discard components whose goodness-of-fit change is less than a preset threshold as pseudo-components to obtain the optimized region of interest decomposition result.

[0079] In one possible implementation, a fitting residual analysis is performed on the preprocessed region of interest to obtain the analysis results, including:

[0080] For the region of interest after preprocessing, a null hypothesis and an alternative hypothesis are established; wherein, the null hypothesis is the hypothesis that the region of interest after preprocessing has no missing Vogit spectral lines, and the alternative hypothesis is the hypothesis that the region of interest after preprocessing contains Vogit spectral lines.

[0081] Obtain the residual sequence corresponding to the region of interest after preprocessing. If the residual data in the residual sequence does not satisfy the null hypothesis but satisfies the alternative hypothesis, the analysis result is determined to be that the residual contains Vogit spectral lines; otherwise, the analysis result is determined to be that the residual has no missing Vogit spectral lines.

[0082] In one possible implementation, based on the analysis results, a residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain LIBS spectral signal decomposition results, including:

[0083] If the analysis result shows that the residual contains Vogit spectral lines, the NIST atomic emission spectrometry database is queried to confirm whether there are real element spectral lines at the wavelength positions corresponding to the residuals. If so, the residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition result. Otherwise, it is determined to be a false positive signal caused by noise and is removed.

[0084] For example, the wavelength position can be used as the initial peak position of the new component, the initial intensity can be set to 80% of the residual peak amplitude, and the initial broadening parameter can be set to the regional average broadening. This component is then added to the original component set as a new seed peak. Subsequently, the updated component set is substituted back into the preprocessed region of interest for full fitting, and the parameters of all components are optimized using the non-negative least squares method to update the decomposition results. The above residual analysis and supplementary fitting process is repeated until the residuals meet any of the following convergence conditions: the maximum amplitude of the residuals is lower than a preset threshold; there is no obvious Vogit line shape feature in the residual sequence; the preset maximum number of iterations is reached. At this point, it is determined that the residuals have no significant Vogit structure, and the final LIBS spectral signal decomposition result is obtained.

[0085] In one possible implementation, a residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition results, including:

[0086] New pseudo-Vogit components are added at the wavelength positions corresponding to the residuals, and then substituted back into the preprocessed region of interest for full fitting. The parameters of all components are optimized, and the decomposition results are updated until the residuals have no significant Vogit structure, thus obtaining the LIBS spectral signal decomposition results.

[0087] To enable those skilled in the art to more easily understand the technical solutions described in the embodiments of this application, examples are provided in conjunction with the above technical solutions, for the following reasons.

[0088] Step 1: Data Acquisition and Preprocessing;

[0089] This embodiment used seven steel alloy samples, provided by the National Steel Materials Testing Center and the Shandong Metallurgical Research Institute. To reduce the impact of sample surface contamination on the analytical results and improve the detection efficiency of LIBS, five sampling points were selected for each sample, and 50 sets of inductively coupled plasma spectra were collected at each sampling point. Finally, 1750 sets (50×5×7) of spectral data were synchronously acquired for each beam type.

[0090] The LIBS system used in the experiment included: a Q-switched Nd:YAG laser (Quantel, Ultra100) with an output center wavelength of 1064 nm, a pulse width of approximately 10 ns, and a repetition frequency of 1 Hz. Laser energy was modulated using a combination of a half-wave plate (HWP) and a Glan prism (GP). The laser beam was expanded by a concave lens L1 (f = −50 mm) and a convex lens L2 (f = 100 mm), and then focused onto the sample surface by a plano-convex lens L3 (f = 50 mm) to generate plasma. The plasma emission light was collected at a 45° angle by a convex lens L4 with a focal length of 100 mm and transmitted via fiber to a four-channel spectrometer (AvaSpec-Mini4096CL, Netherlands), with a spectral range of 186–797 nm, a resolution of approximately 0.05 nm, and a minimum gate width of 9 μs. In this experiment, the delay time and gate width were set to 1.5 μs and 9 μs, respectively.

[0091] Step 2: Adaptive ROI partitioning;

[0092] 1. Gaussian smoothing for noise reduction: Constructing a smoothing kernel Convolution operation on the original spectrum This eliminates noise interference while preserving spectral characteristics. The smoothed spectral intensity, This represents the original spectral intensity.

[0093] 2. Peak candidate point identification: Calculate the first and second derivatives of the smoothed spectrum, and identify local maxima points through the extreme values ​​of the derivatives, which are used as candidate peak points for potential spectral lines.

[0094] 3. Threshold Filtering Anchor Point: Set the absolute intensity threshold. With relative significance threshold Only retain those that simultaneously meet the condition of "strength higher than". "and "significance higher than The candidate peaks are defined as "anchor points" (core markers of effective spectral lines).

[0095] 4. Anchor point clustering: Adjacent anchor points are clustered based on wavelength spacing. When the distance between two peaks is less than... When, they are considered to be in the same peak cluster, among which The coefficients are used to control the clustering scale of the anchor points. The left and right boundaries of each peak cluster are determined by the adjacent local minima.

[0096] 5. Boundary optimization and adjustment:

[0097] Sentinel windowing: If the spectral energies at both ends of the initial boundary are still higher than the threshold. An energy integral criterion is used to expand the window, ensuring complete coverage of spectral information.

[0098] Segmentation and Merging: If the width of the region corresponding to a single peak cluster exceeds the maximum allowable value The valley is divided into multiple sub-regions based on its location; if the distance between adjacent regions is less than a small threshold... If they are combined into one ROI, then they are merged.

[0099] 6. Final output: The entire spectrum is adaptively divided into multiple independent ROIs, with independent spectral structure, smooth background changes, and consistent broadening mechanisms within each ROI.

[0100] Step 3: Vogit decomposition of physical constraints;

[0101] 1. Seed peak identification (determining the decomposition starting point);

[0102] Main peak identification: Within each ROI, through the first derivative extremum condition ( Locate the main peak position and introduce a baseline amplification factor. This ensures that the intensity of the main peak is significantly higher than the background intensity.

[0103] Acromion identification: Judging potential acromions by the trend of derivative changes.

[0104] Right acromion: present This causes the first derivative to exhibit a "decreasing-minimum-increasing" trend. , For local minimum, );

[0105] Left acromion: present This causes the first derivative to exhibit an "increasing-maximum-decreasing" trend. , For local maxima, ).

[0106] 2. Physical constraint fitting;

[0107] Using a pseudo-Vogit line shape (combining Gaussian and Lorentz broadening properties), the observed spectra within the ROI are represented as a non-negative linear superposition of multiple pseudo-Vogit components: ( For component intensity, Peak position, To broaden Gaussian, To broaden Lorenz, (This refers to the mixing coefficient). Intensity of each component. Non-negative least squares constraints are used to ensure physical non-negativity and suppress overfitting. Simultaneously, the residuals are minimized. Finally, the change in goodness of fit after removing the j-th component is calculated as follows: It is used to quantify the independent contribution of each spectral line, to eliminate spurious components, and to evaluate the reliability of the fit.

[0108] Step 4: Residual analysis and completion;

[0109] 1. Residual extraction: The difference between the observed spectrum and the fitted spectrum after Vogit decomposition is defined as the residual sequence, and the segment with local structure (suspected missing spectral lines) is focused in the residual.

[0110] 2. Statistical tests (t-test);

[0111] Establish the hypothesis:

[0112] Null hypothesis : No missing Vogit spectral lines in the residual, i.e. ( It is a constant. (This is random noise).

[0113] Alternative Hypothesis The residual contains Vogit spectral lines, i.e. ( For spectral line intensity, (This is a Vogit template). Satisfying the H1 hypothesis only indicates that the residuals statistically exhibit Vogit linear characteristics. Further false positive detection using the NIST database is required: if the database contains the true elemental spectral line corresponding to the wavelength, it is determined to be a true missing spectral line and iterative updates are initiated; if there is no corresponding spectral line, it is determined to be a false positive signal caused by noise and is removed.

[0114] Linear Regression: Constructing a Regression Model for Suspected Segments The coefficient was determined by the t-test. The significance of, if Then refuse It was confirmed that there were missing spectral lines.

[0115] NIST database verification: Query the NIST atomic emission spectrometry database to confirm whether the wavelength position corresponding to the residual contains the spectral line of the real element, and eliminate false positive signals caused by noise.

[0116] Supplementation and refitting: If the test passes (true spectral lines exist), add new pseudo-Vogit components at the wavelength position, resubmit them to the ROI for full fitting, optimize the parameters of all components (peak position, intensity, broadening), update the decomposition results, until the residuals have no significant Vogit structure.

[0117] like Figure 2 As shown, based on the same inventive concept, this application provides a LIBS spectral signal decomposition device, comprising:

[0118] The ROI extraction module 201 is used to acquire the LIBS spectral signal to be decomposed and to divide the LIBS spectral signal to be decomposed into regions of interest to obtain multiple regions of interest.

[0119] The signal decomposition module 202 is used to perform sub-peak detection on the spectral signal in the region of interest, and simultaneously use a physically constrained pseudo-Vogit line shape to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest.

[0120] The residual completion module 203 is used to perform fitting residual analysis on the preprocessed region of interest, obtain analysis results, and iteratively optimize the preprocessed region of interest using a residual completion mechanism based on the analysis results to obtain LIBS spectral signal decomposition results.

[0121] This LIBS spectral signal decomposition device can perform the above-mentioned method and technical solution. Its principle and beneficial effects are similar, and will not be described in detail here.

[0122] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0123] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0127] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A LIBS spectral signal decomposition method, characterized in that, include: The LIBS spectral signal to be decomposed is acquired, and the region of interest is divided into multiple regions of interest. Sub-peak detection is performed on the spectral signal in the region of interest, and at the same time, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest. The region of interest after preprocessing is subjected to fitting residual analysis to obtain the analysis results. Based on the analysis results, the region of interest after preprocessing is iteratively optimized using a residual completion mechanism to obtain the LIBS spectral signal decomposition results.

2. The LIBS spectral signal decomposition method according to claim 1, characterized in that, The LIBS spectral signal to be decomposed is acquired, and the LIBS spectral signal to be decomposed is divided into regions of interest to obtain multiple regions of interest, including: The LIBS spectral signal to be decomposed is acquired, and Gaussian smoothing is applied to the acquired LIBS spectral signal to obtain the denoised LIBS spectral signal. Identify all local maxima of the denoised LIBS spectral signal to obtain peak candidate points; Based on preset absolute intensity thresholds and relative significance thresholds, the candidate peak points are screened to obtain significant peak anchor points; Based on the wavelength distance between significant peak anchor points, the significant peak anchor points are clustered to obtain at least one peak cluster; The initial region is determined by the adjacent local minimum or energy integral boundary of each peak cluster, and the initial region is merged or segmented and optimized according to the maximum region width and minimum region spacing criteria to output multiple regions of interest.

3. The LIBS spectral signal decomposition method according to claim 2, characterized in that, Based on the wavelength distance between significant peak anchor points, the significant peak anchor points are clustered to obtain at least one peak cluster, including: Determine the peak spacing threshold. When the wavelength distance between significant peak anchor points is less than the peak spacing threshold, the significant peak anchor points are clustered into the same peak cluster, thus obtaining at least one peak cluster.

4. The LIBS spectral signal decomposition method according to claim 2, characterized in that, The initial region is determined by the adjacent local minimum or energy integral boundary of each peak cluster, and the initial region is optimized by merging or segmenting according to the maximum region width and minimum region spacing criteria, outputting multiple regions of interest, including: The left and right boundaries of each peak cluster are determined by the adjacent local minimum points, thus obtaining the initial boundary corresponding to each peak cluster; If the spectral energies at both ends of the initial boundary are higher than the preset energy threshold, the window formed by the initial boundary is expanded using the energy integration criterion to ensure complete coverage of spectral line information and obtain the initial region. If the width of the initial region corresponding to a single peak cluster exceeds the maximum allowable value, it is divided into multiple sub-regions according to the valley position. If the distance between adjacent initial regions is less than a small threshold, they are merged into one region, and multiple regions of interest are output.

5. The LIBS spectral signal decomposition method according to claim 1, characterized in that, Sub-peak detection is performed on the spectral signal in the region of interest, and simultaneously, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in a preprocessed region of interest, including: Identify the main peak, right shoulder peak, and left shoulder peak of the spectral signal in the region of interest, and use the main peak, right shoulder peak, and left shoulder peak together as the seed peak of the region of interest; Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a physically constrained pseudo-Vogit line shape, thus obtaining the preprocessed region of interest.

6. The LIBS spectral signal decomposition method according to claim 5, characterized in that, Using the seed peak as the starting point for decomposition, a physically constrained pseudo-Vogit line shape is used to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components, resulting in the preprocessed region of interest, including: Using the seed peak as the starting point for decomposition, the spectral signal in the region of interest is modeled as a linear superposition of multiple pseudo-Vogit peak components using a pseudo-Vogit line shape. Non-negative least squares constraints are applied to each component to ensure physical non-negativity and suppress overfitting, while minimizing the residuals to obtain the constrained components. For the constrained components, determine the fitting optimization change corresponding to the component to be removed, and eliminate the components whose goodness of fit change is less than the preset change threshold to obtain the preprocessed region of interest.

7. The LIBS spectral signal decomposition method according to claim 5, characterized in that, Perform fitting residual analysis on the preprocessed region of interest to obtain the analysis results, including: For the region of interest after preprocessing, a null hypothesis and an alternative hypothesis are established; wherein, the null hypothesis is the hypothesis that the region of interest after preprocessing has no missing Vogit spectral lines, and the alternative hypothesis is the hypothesis that the region of interest after preprocessing contains Vogit spectral lines. Obtain the residual sequence corresponding to the region of interest after preprocessing. If the residual data in the residual sequence does not satisfy the null hypothesis but satisfies the alternative hypothesis, the analysis result is determined to be that the residual contains Vogit spectral lines; otherwise, the analysis result is determined to be that the residual has no missing Vogit spectral lines.

8. The LIBS spectral signal decomposition method according to claim 5, characterized in that, Based on the analysis results, a residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition results, including: If the analysis result shows that the residual contains Vogit spectral lines, the NIST atomic emission spectrometry database is queried to confirm whether there are real element spectral lines at the wavelength positions corresponding to the residuals. If so, the residual completion mechanism is used to iteratively optimize the preprocessed region of interest to obtain the LIBS spectral signal decomposition result. Otherwise, it is determined to be a false positive signal caused by noise and is removed.

9. The LIBS spectral signal decomposition method according to claim 5, characterized in that, The region of interest after preprocessing is iteratively optimized using a residual completion mechanism to obtain the LIBS spectral signal decomposition results, including: New pseudo-Vogit components are added at the wavelength positions corresponding to the residuals, and then substituted back into the preprocessed region of interest for full fitting. The parameters of all components are optimized, and the decomposition results are updated until the residuals have no significant Vogit structure, thus obtaining the LIBS spectral signal decomposition results.

10. A LIBS spectral signal decomposition device, characterized in that, include: The ROI extraction module is used to acquire the LIBS spectral signal to be decomposed and to divide the LIBS spectral signal to be decomposed into regions of interest to obtain multiple regions of interest. The signal decomposition module is used to perform sub-peak detection on the spectral signal in the region of interest, and simultaneously use a physically constrained pseudo-Vogit line shape to model the spectral signal in the region of interest as a linear superposition of multiple pseudo-Vogit peak components to obtain the preprocessed region of interest. The residual completion module is used to perform fitting residual analysis on the preprocessed region of interest, obtain the analysis results, and iteratively optimize the preprocessed region of interest using a residual completion mechanism based on the analysis results to obtain the LIBS spectral signal decomposition results.

Images