A standard substance-based wavelength calibration method for LIBS spectrometer

Abstract

The application discloses a LIBS spectrometer wavelength calibration method based on a standard substance, and comprises the following steps: a plurality of sub-wave bands are divided according to the wavelength direction, and corresponding standard substances are selected; the standard substances are excited by using a LIBS technology, and atomic emission spectral lines are collected; airPLS is used to correct the baseline of the screened spectral data; the window data is subjected to three times spline interpolation, and a spectral curve is reconstructed; a two-stage Lorentz fitting algorithm is applied to realize accurate modeling, extract the center wavelength, intensity, half-peak width and baseline offset value of each characteristic peak; a pixel-wavelength mapping model is constructed by combining the adaptive segmented polynomial fitting strategy of the characteristic peak density distribution and the fitting residual, and the pixel-wavelength mapping lookup table of the spectrometer is written. By exciting the standard substance to generate wide-spectrum and dense characteristic spectral lines, and by combining the two-stage Lorentz fitting algorithm and the adaptive segmented polynomial fitting technology, the wavelength calibration precision in the full-wave band range of the spectrometer is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of spectrometer calibration, and particularly relates to a LIBS spectrometer wavelength calibration method based on standard substances. BACKGROUND

[0002] In the prior art, a LIBS (Laser-Induced Breakdown Spectroscopy) spectrometer usually adopts a standard light source method for wavelength calibration, such as a mercury argon lamp, corresponding to characteristic spectral lines of 253.65 nm, 365.01 nm, 696.54 nm, such as a neon lamp, corresponding to characteristic spectral lines of 540.06 nm, 585.25 nm, and the spectral line coefficient, resulting in limited wavelength calibration accuracy. In this wavelength calibration method, the spectral lines are sparse, and depend on standard light sources such as mercury argon lamps and neon lamps, which require stable light source conditions in a specific environment as a reference. However, gas discharge light sources are easily affected by factors such as temperature changes, power fluctuations and use aging, and are prone to spectral line wavelength drift or intensity attenuation, resulting in unstable calibration results. When using a standard light source for calibration, the number of available characteristic spectral lines is limited, especially in the ultraviolet band (200-400 nm), which is sparse and difficult to cover the entire detection band. This uneven distribution of calibration points makes the overall fitting model susceptible to local data, resulting in insufficient fitting accuracy in some bands, limiting the quantitative accuracy of high-resolution spectrometers in the full wavelength range.

[0003] In the Chinese patent document with the publication number CN111998944A, a wide-band spectrometer accurate calibration device and method based on multiple light sources are disclosed, including a standard light source, a light source controller, a light switch and a driving circuit; the working principle is as follows: the standard light source is controlled by the light source controller to generate spectral lines of a specific wavelength; the standard light source is connected to any channel of the light switch by an optical fiber, the light switch selects a channel according to the control instruction, and the output light of a selected standard light source passes through the adjustable optical attenuator; the adjustable optical attenuator adjusts the attenuation value of the output power of the standard light source, and the power enters the spectrometer; the optical signal is attenuated by the adjustable optical attenuator, enters the light switch, and the optical signal enters the standard light power meter, measures the real light power of the optical signal entering the spectrometer, and then obtains the polynomial coefficient according to the characteristic peak position matching and the fitting of the wavelength calibration curve. In the above disclosed technology, the wavelength calibration of the wide-band spectrometer is realized by the standard light source and the light source controller, but the problem of few spectral lines of the standard light source and uneven distribution of calibration points cannot be effectively overcome, the wavelength calibration accuracy of the high-resolution spectrometer in the full wavelength range is limited by the fluctuation of the spectral peak position and the accuracy of the fitting curve, and the overall fitting model is susceptible to local data. SUMMARY

[0004] The application aims to provide a LIBS spectrometer wavelength calibration method based on standard substance, which generates wide-spectrum and dense characteristic spectral lines by exciting standard substance, and improves the accuracy of wavelength calibration in the full waveband range of the spectrometer by combining an improved two-stage Lorentz fitting algorithm and an adaptive segmented polynomial fitting technology combining characteristic peak density distribution and fitting residual.

[0005] To solve the above technical problems, the embodiment of the application provides a technical solution as follows: a LIBS spectrometer wavelength calibration method based on standard substance, comprising the following steps: step S1: according to the detection capability of the spectrometer in the ultraviolet to near-infrared waveband, a plurality of sub-wavebands are divided, and standard substance capable of generating rich characteristic spectral lines in each sub-waveband is selected; step S2: the standard substance is excited by using the LIBS technology, the integral start time of the spectrometer is controlled by a time sequence controller, the delay is set to avoid the high-intensity continuous background radiation of the initial stage of the plasma, and stable and clear atomic emission spectral lines are collected as original spectral data; step S3: the collected original spectral data is screened, and the airPLS is used to correct the baseline of the screened spectral data; step S4: after the baseline correction, the main peak center pixel of each spectral line is taken as a base point, the data in the range of ±3-10 pixels is taken as window data, the window data is taken as an interpolation input interval, the point interpolation is performed at a density of 2-5 times, the cubic spline interpolation is performed, and the spectral curve is reconstructed; step S5: the spectral peak data after the reconstructed spectral curve is applied to the improved two-stage Lorentz fitting algorithm to realize accurate modeling: including the first stage of single-peak fitting, if the fitting residual exceeds the extreme value of 1.5 times the residual standard deviation, the second stage of double-peak fitting is triggered; the second peak parameter is dynamically generated according to the residual, and the model complexity is controlled through the fitting residual drop rate, and finally the center wavelength, intensity, half-peak width and baseline offset value of each characteristic peak are extracted; step S6: a plurality of groups of spectral curves with the peak intensity saturation degree of 80%-90% are selected, the maximum half-peak width value in each group of spectral curves is extracted, and the average value is taken as the resolution evaluation index of the spectrometer; step S7: based on the extracted characteristic peak center wavelength and the corresponding pixel position, an adaptive segmented polynomial fitting strategy combining the characteristic peak density distribution and the fitting residual is used to construct a pixel-wavelength mapping model, including the following steps: based on the resolution, the characteristic peak is matched and locked, the pixel interval corresponding to the successfully matched characteristic spectral peak data is divided into a plurality of sub-sections, and the sub-sections are combined or subdivided according to the characteristic peak coverage rate and the resolution threshold; the polynomial fitting is performed in each sub-section, and when the fitting residual RMSE exceeds the threshold, the regularization and / or model order promotion strategy is introduced to optimize the fitting stability; finally, each segment fitting parameter is written into the pixel-wavelength mapping lookup table of the spectrometer, and the full-segment high-precision wavelength calibration of the spectrometer is realized.

[0006] Further, in step S1, the standard substance is standard metal target copper and tin.

[0007] Further, in step S2, the original spectral data includes not less than 10 groups of spectral data repeatedly collected at the same excitation position for each standard substance.

[0008] Further, in step S3, the baseline correction of the screened spectral data by airPLS includes: initializing the baseline as the low-frequency trend of the original spectrum; adjusting the weight coefficient by iteration to suppress the influence of the peak region on the baseline fitting, obtaining the smooth baseline signal and completing the baseline correction by deducting from the original spectral data.

[0009] Further, in step S5, the first stage single peak fitting includes establishing a single peak model based on local window data, and the mathematical expression is: , wherein A is the peak intensity, is the center wavelength, is the half peak width, and B is the baseline offset.

[0010] Further, in step S5, the second stage double peak fitting includes double peak Lorentz model fitting, and the mathematical expression is: ; wherein the second peak parameters are dynamically determined by the residual extreme value: the center wavelength initial value is taken from the position corresponding to the residual maximum value, the intensity initial value is set as about 2 times the residual peak value, and the half peak width inherits from the main peak parameters.

[0011] Further, in step S6, the average value is calculated as the resolution evaluation index of the spectrometer, and the mathematical expression of the resolution is: ; wherein is the resolution, N is the total number of data groups, is the maximum half peak width of the selected characteristic peak in the ith data group.

[0012] Further, in step S7, the characteristic peak matching and locking based on the resolution specifically includes searching for each fitting peak center wavelength in the standard spectral line library, and calculating the candidate matching items within the tolerance range according to the resolution matching threshold; wherein the resolution matching threshold is:

[0013] The sorting function including the intensity weight coefficient 0.7 and the wavelength deviation weight coefficient 0.3 is constructed, and the strong peak signal is preferentially selected for matching and locking. The characteristic peak matching process implements bidirectional exclusive constraint, that is, each fitting peak corresponds to a unique standard wavelength and each standard wavelength is matched at most once.

[0014] Further, in step S7, the characteristic peak coverage criterion is: the entire effective pixel range is preliminarily segmented into multiple candidate subsegments, and the number of effective standard spectral lines matched in each candidate subsegment is counted , and the minimum number of spectral lines is dynamically set according to the length of each segment , if , the segment is merged with one of the adjacent segments, and the process is repeated until the new subsegment generated after the merging , wherein

[0015]

[0016] In the formula: is the length of the current segment of pixels; is the average pixel spacing of the characteristic spectral line; is the reference spectral line density.

[0017] Further, in step S7, the fitting residual criterion in the adaptive segmented polynomial fitting strategy is: a cubic polynomial fitting is performed on the subsegment that meets the characteristic peak coverage requirement, and the RMSE of the fitting result of the subsegment is calculated; when the RMSE is greater than a set threshold, the following adjustment strategy is performed according to the pixel span of the subsegment : if pixels, the subsegment is uniformly divided into two new subsegments, and the fitting is performed again; otherwise, the original segment is maintained, the order of the fitting model is increased from three to four, and an L2 regularization term is introduced to constrain the fitting parameters.

[0018] Compared with the prior art, the LIBS spectrometer wavelength calibration method based on a standard substance provided by the application generates dense characteristic spectral lines covering the ultraviolet to near-infrared band by exciting the standard substance, ensures that there are enough calibration points in each sub-band, effectively solves the problems of sparse characteristic spectral lines in the ultraviolet band, difficulty in covering the entire detection band, and uneven distribution of calibration points in the prior art, and significantly improves the quantitative accuracy of a high-resolution spectrometer in the full-band range. By using the method of combining cubic spline interpolation and improved Lorentz fitting, the volatility of the spectral peak position is effectively reduced, the continuity and stability of the spectral data are improved, the difficulty in positioning the peak position caused by the volatility of the spectral peak position, the diversity of the peak shape and the uncertainty of the form in the prior art is effectively solved, and the precision and reliability of the wavelength calibration of the spectrometer are significantly improved. By using the adaptive segmented polynomial fitting technology based on the joint criterion of characteristic peak density and fitting residual, combined with the cubic or quartic polynomial model and the L2 regularization optimization mechanism, the fitting error caused by local nonlinearity is effectively reduced, and the consistency and accuracy of the wavelength calibration of the spectrometer in the full-band range are improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] One or more embodiments are illustrated by way of example in the figures that are part of this disclosure and which are illustrative, and not restrictive, of the embodiments, wherein like references numerals refer to like elements, and wherein the notation of the figures is not to scale unless otherwise specified. In the figures:

[0020] Figure 1 A standard substance-based LIBS spectrometer wavelength calibration method step flowchart in embodiments of the present application;

[0021] Figure 2 A LIBS device structure schematic diagram in embodiments of the present application;

[0022] Figure 3 A copper element characteristic peak before and after interpolation spectral peak comparison result schematic diagram in embodiments of the present application;

[0023] Figure 4 A copper element characteristic peak Lorentz function single peak fitting result schematic diagram in embodiments of the present application;

[0024] Figure 5 A multi-peak Lorentz fitting result schematic diagram in embodiments of the present application;

[0025] Figure 6 A whole polynomial fitting result schematic diagram in the prior art;

[0026] Figure 7 An adaptive segmented polynomial fitting result schematic diagram in embodiments of the present application;

[0027] Figure 8 A segmented fitting residual and prior art whole fitting residual comparison curve schematic diagram in embodiments of the present application. DETAILED DESCRIPTION

[0028] To make the objects, technical solutions, and advantages of the present application clearer, the various embodiments of the present application will be described in detail below with reference to the accompanying drawings. However, those of ordinary skill in the art can understand that, in the various embodiments of the present application, many technical details are presented in order to make the reader better understand the present application. However, the technical solutions claimed by the various claims of the present application can be implemented even without these technical details and various changes and modifications based on the following various embodiments.

[0029] It should be noted that if the present application embodiments involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship, movement condition, etc. between the components in a certain posture, and if the certain posture changes, the directional indications also change accordingly.

[0030] As Figure 1As shown, one embodiment of the present application relates to a standard substance-based LIBS spectrometer wavelength calibration method, comprising the following steps:

[0031] Step S1: According to the detection capability of the spectrometer in the ultraviolet to near infrared waveband, multiple sub-wavebands are divided, and standard substances capable of generating rich characteristic spectral lines in each sub-waveband are selected to realize the coverage and calibration support of the entire working waveband of the spectrometer. In one example, the standard substances selected are metal targets copper and tin, which can generate dense characteristic spectral lines in multiple sub-wavebands under the excitation of LIBS technology. The problems of sparse characteristic spectral lines in the ultraviolet waveband, difficulty in covering the entire detection waveband, and uneven distribution of calibration points in the traditional method are effectively solved, which significantly improves the quantitative accuracy of high-resolution spectrometers in the full waveband range, reduces the dependence on specific external standard light sources, and reduces the cost and inconvenience caused by light source replacement and maintenance.

[0032] Step S2: When the standard substance is excited by LIBS technology, the integration start time of the spectrometer is controlled by a time sequence controller, and a delay is set to avoid the high-intensity continuous background radiation of the initial plasma, and stable and clear atomic emission spectral lines are collected as original spectral data; Preferably, each standard substance is repeatedly collected for not less than 10 groups of spectral data at the same excitation position to enhance statistical stability and suppress random errors.

[0033] Step S3: The collected original spectral data is screened, and the screened spectral data is baseline corrected by airPLS;

[0034] Step S4: After baseline correction, the main peak center pixel of each spectral line is taken as the base point, the data in the range of ±3-10 pixels is taken as the window data, the window data is taken as the interpolation input interval, the point interpolation is performed according to 2-5 times the density, the spectral curve is reconstructed by cubic spline interpolation;

[0035] Step S5: The spectral peak data after the reconstructed spectral curve is applied to the improved two-stage Lorentz fitting algorithm to realize accurate modeling: including the first stage to perform single peak fitting, if there is an extreme value in the fitting residual that exceeds 1.5 times the standard deviation, the second stage double peak fitting is triggered; The second peak parameter is dynamically generated by the residual, and the model complexity is controlled by the fitting residual drop rate, and finally the center wavelength, intensity, half peak width, baseline offset value of each characteristic peak are extracted;

[0036] Step S6: Select multiple groups of spectral curves with peak intensity saturation of 80%-90%, extract the maximum half peak width value in each group of spectral curves, and calculate the average value as the resolution evaluation index of the spectrometer;

[0037] Step S7: Based on the extracted characteristic peak center wavelength and corresponding pixel position, combined with the adaptive segmented polynomial fitting strategy of characteristic peak density distribution and fitting residual, the pixel-wavelength mapping model is constructed, including the following steps: based on the above resolution, the characteristic peak matching lock is performed, the pixel interval corresponding to the successfully matched characteristic peak data is preliminarily divided into several subsegments, and the segment merging or subdivision is performed according to the characteristic peak coverage rate and the resolution threshold; the polynomial fitting is performed in each subsegment, and when the fitting residual RMSE exceeds the threshold, the regularization and / or model order promotion strategy is introduced to optimize the fitting stability; finally, the fitting parameters of each segment are written into the pixel-wavelength mapping lookup table of the spectrometer, and the full-segment high-precision wavelength calibration of the spectrometer is realized.

[0038] Through the above steps, the LIBS spectrometer wavelength calibration method based on standard substance realizes high-density calibration point coverage in the whole working waveband and accurate analysis of overlapping spectral lines, and solves the problems of sparse spectral line dependence, insufficient detection stability and limited fitting accuracy of complex spectral lines in the traditional method.

[0039] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substance is involved, wherein the working full waveband wavelength range of the spectrometer is 200-900 nm, based on the spectral line coverage density, experimental operability and chemical stability requirements, the working full waveband is divided into multiple sub wavebands and standard substances are selected, the standard substances include metal targets copper and tin, and nitrogen and oxygen elements contained in air; as shown in Table 1, the sub waveband and standard substance element, characteristic peak corresponding record table:

[0040] Table 1 Sub waveband and standard substance element, characteristic peak corresponding record table

[0041]

[0042] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substance is involved, which excites standard substance to generate plasma by LIBS (laser-induced breakdown spectroscopy) technology, and collects corresponding spectral data based on the detection working waveband and characteristics of the spectrometer to be calibrated. As shown in Figure 2As shown, the LIBS device includes a laser excitation unit for generating a high-energy pulsed laser to induce plasma on the sample surface, including a laser, a mirror, and a focusing lens group; an optical collection unit for collecting plasma emission light and transmitting it to a spectrometer, which in this embodiment uses optical fiber transmission; a spectrometer for dispersing plasma light into a spectrum and detecting it; a computer for data processing and control, such as spectral analysis and quantitative modeling; and a timing controller for accurately coordinating the action timing of the laser pulse, spectral collection, detector gating, and environmental control unit. During the collection of raw spectral data, the excitation intensity and signal-to-noise ratio of the plasma are optimized by adjusting the laser energy, the focused spot size, and the detector integration time; the integration start time is controlled by a digital delay trigger mechanism to avoid the continuous background radiation in the initial stage of the plasma, and stable and clear atomic emission lines are collected; for the spectrometer to be calibrated, copper target samples and tin target samples are collected respectively to ensure that the signal intensity of the spectral data is stable and the signal-to-noise ratio meets the requirements. In actual operation, it is necessary to ensure that the surfaces of the copper target and the tin target are flat and free of oxidation layer or impurity pollution to improve the excitation efficiency and data quality. Each standard substance is repeatedly collected with not less than 10 groups of spectral data at the same excitation position to ensure the statistical reliability of the data. This multi-group data collection method not only can reduce the influence of random errors, but also can provide more reference data points for subsequent calibration, thereby improving the stability of the calibration results. By optimizing the laser excitation parameters, such as adjusting the laser energy and the focused spot size, the plasma excitation intensity is adapted to the linear response interval of the spectrometer to avoid signal saturation or insufficient signal-to-noise ratio. After the laser pulse is triggered, the integration start time of the spectrometer is controlled by a digital delay generator, and the delay is set to avoid the high-intensity continuous background radiation in the initial stage of the plasma. The background radiation in this stage usually masks the characteristic spectral lines, resulting in a decrease in data quality. By setting an appropriate delay, stable atomic emission lines can be collected, thereby improving the reliability of the spectral data and the calibration accuracy.

[0043] After the data collection is completed, the collected spectral data is preliminarily screened to ensure that the signal intensity and signal-to-noise ratio of each group of data are within a preset reasonable range. By comparing the intensity changes of the spectral characteristic peaks under different excitation conditions, the best excitation parameter combination is selected to ensure the stability and accuracy of the subsequent calibration process. For the screened spectral data, the airPLS (Adaptive Iteratively Reweighted Penalized Least Squares) method is used for baseline correction. This method dynamically distinguishes the effective peak signal and the baseline background in the spectral data through an adaptive iterative weighting strategy, and gradually fits the baseline profile. The specific process includes: initializing the baseline as the low-frequency trend of the original spectrum, adjusting the weight coefficient through iteration to suppress the influence of the peak region on the baseline fitting, and finally obtaining a smooth baseline signal and subtracting it from the original data, effectively eliminating the interference of instrument dark current, stray light and continuous background radiation.

[0044] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substances is involved. In the LIBS spectral wavelength calibration process, due to factors such as laser pulse instability, plasma spatial distribution difference, optical system perturbation, and detector sampling accuracy, even if the same standard substance is repeatedly collected at the same position, slight fluctuations in the spectral peak center position often occur. Although this fluctuation is small, it still introduces an error that cannot be ignored in high-precision wavelength calibration scenarios. The stability of the spectral line peak position directly affects the accuracy of the fitting result. To improve the reconstruction accuracy of the spectral line and reduce the interference of the above fluctuations, and to provide a high-resolution data basis for subsequent Lorentz fitting, the embodiment provided in the present application introduces a cubic spline interpolation process after baseline correction of the spectral data. This method uses a cubic spline function with first and second derivative continuity to reconstruct a continuous curve from the original discrete spectrum, significantly improving the data density while preserving the spectral peak profile. Specifically, it includes: spectral peak window selection: taking the center pixel of the main peak of each spectral line as a reference, the data within ±3-10 pixels of the main peak are selected as the interpolation input interval; constructing a spline function: based on the wavelength and corresponding intensity points in the selected interval, a cubic spline curve is constructed to ensure local smoothness and physical reasonableness; high-density interpolation: the original wavelength interval is interpolated with 2-5 times the accuracy (density) to construct a more continuous and smooth spectral curve.

[0045] Figure 3The copper element characteristic peak (236.989 nm) before and after interpolation is shown. The left figure is the original spectrum of 5 repeated acquisitions, and the peak shape has a certain degree of deviation, and the main peak RSD is 0.01%; the right figure is the spectrum after three times of spline interpolation, the spectrum peak edge transition is smoother, the main peak position of multiple repeated spectra is highly consistent, and the RSD is significantly reduced to 0.0004%, effectively improving the consistency and symmetry of the spectrum peak shape. The cubic spline interpolation not only improves the spectrum reconstruction accuracy, suppresses the spectrum peak fluctuation caused by the sampling error, but also provides a more reliable fitting basis for the subsequent Lorentz function fitting, and further promotes the improvement of the wavelength calibration accuracy.

[0046] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substance is involved. In a complex spectrum, the existence of overlapping peaks can cause center wavelength calculation error. The present application realizes the analysis and positioning of complex spectral peaks through an improved Lorentz fitting algorithm, and the technical features mainly lie in the dynamic parameter initialization mechanism, the residual error driven double peak discrimination strategy and the intensity priority bidirectional matching principle. The specific implementation process is as follows:

[0047] When performing feature peak rough positioning on the collected spectral intensity data, a dynamic baseline calculation and adaptive threshold setting strategy is adopted. The 25% quantile of the spectral intensity distribution is taken as the baseline reference value, and the significance threshold is dynamically calculated by the difference between the 95% quantile and the 25% quantile of the spectral intensity, which effectively overcomes the problem of weak peak missed detection caused by traditional fixed threshold. In the peak position determination process, the half peak width characteristic quantity is analyzed synchronously, and the window edge protection mechanism is combined to exclude the pseudo peak signal caused by noise.

[0048] For each candidate peak region, a two-stage Lorentz fitting process is performed. First, a single peak model is established based on local window data, and its mathematical expression is as follows:

[0049] .

[0050] In the formula, A is the peak intensity, is the center wavelength, is the half peak width, and B is the baseline offset. The parameter initialization adopts a dynamic calculation mechanism: the initial center wavelength is determined by statistical analysis of window data, the baseline value takes the average of the intensity of the first three points at the beginning and end of the window, and the half peak width is automatically generated by detecting the wavelength span of the original data at 50% peak height.

[0051] Figure 4The Lorentz function single peak fitting result of the copper element characteristic peak (236.989 nm) is shown. After three spline interpolation and Lorentz fitting processing, the fitting curve is highly consistent with the actual spectrum, accurately restoring the peak position and morphological characteristics.

[0052] When there is a significant extreme value exceeding 1.5 times the residual standard deviation in the single peak fitting residual, a double peak Lorentz model fitting is triggered, and its mathematical expression is:

[0053] .

[0054] The second peak parameter is dynamically determined by the residual extreme value: the center wavelength initial value is taken from the position corresponding to the residual maximum value, and the intensity initial value is about 2 times the residual peak value, and the half peak width inherits from the main peak parameter. By constraining the double peak fitting residual sum of squares to be reduced by more than about 20% compared to the single peak, the model complexity is self-optimized.

[0055] As Figure 5 shown in the comparison chart of the multi-peak Lorentz fitting results achieved based on the above strategy. In the figure, taking the two overlapping characteristic peaks of copper element—224.262 nm and 224.7 nm as an example, the original spectral data and the separation fitting results of the double peak Lorentz model are shown. Through double peak fitting, two closely adjacent peak shapes are successfully separated, effectively solving the fitting error and peak position deviation problem caused by peak overlap in single peak fitting. The fitting curve is highly consistent with the experimental data, further verifying the accuracy and robustness of the method in complex spectral analysis, providing a reliable peak parameter basis for subsequent quantitative analysis. Through this step, we get the basic parameters of each candidate peak, including peak intensity, center wavelength, half peak width and baseline offset. These parameters provide an important basis for further accurate analysis of the spectrum. In particular, the half peak width (FWHM), which not only reflects the width of the spectral peak, but also can be used as an important evaluation index of resolution. By measuring the half peak width of different peaks, we can intuitively understand the resolution capability of the instrument. The higher the resolution, the smaller the half peak width. This index has important significance in practical application, especially when the spectral peaks are close to each other, it can provide a basis for judging whether they can be effectively distinguished.

[0056] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substances is involved. The full width at half maximum (FWHM) of each candidate peak can be used to quantify the resolution of the spectrometer. To ensure the accuracy of the measurement results, 5 sets of test data for each standard sample must be selected, and the conditions of these test data should be consistent, especially the peak intensity saturation should be controlled in a moderate range (e.g. 80%-90%). This range ensures that the spectral line will not be distorted due to excessive saturation, thereby affecting the evaluation of the resolution.

[0057] After obtaining the spectral lines with obvious characteristics in the standard samples, the half peak width of each spectral line can be further analyzed. The half peak width refers to the width when the spectral line intensity reaches half of the maximum value, which directly reflects the degree of detail of the spectral line. To quantify the resolution, statistical analysis can be performed on the half peak width values in multiple data sets. Among the half peak width values of the characteristic spectral lines selected in each data set, we should focus on the maximum half peak width value, because the maximum half peak width value will represent the worst case of resolution in that data set. That is, the maximum half peak width value of each data set reflects the resolution capability of that test. The following formula can be used to calculate the resolution:

[0058] .

[0059] where, is the resolution, N is the total number of data sets, is the maximum half peak width of the selected characteristic peak in the i-th data set. By this formula, by calculating the average of these maximum half peak widths, we get the average resolution of the spectrometer under that condition. The calculation of the resolution of the instrument provides a reliable basis for further analysis of the characteristic peaks. The smaller the value of the resolution, the higher the resolution of the instrument, which can effectively distinguish closer spectral peaks. Therefore, by calculating the resolution, the resolution capability of the instrument can be accurately evaluated, and a reference is also provided for the matching of characteristic spectral lines.

[0060] In one embodiment, a LIBS spectrometer wavelength calibration method based on standard substances is involved. To improve the fitting accuracy and robustness of the spectrometer wavelength calibration, the present invention proposes an adaptive segmented polynomial fitting strategy based on the joint criterion of characteristic peak density and fitting residual in the pixel-wavelength mapping relationship construction stage, which is used to replace the traditional overall cubic polynomial fitting scheme. As Figure 6 shown in the traditional overall polynomial fitting result, the fitting curve is relatively smooth but the fitting effect for local nonlinear regions is poor, and the overall RMSE is high.

[0061] As Figure 7The adaptive segmented polynomial fitting result is shown. The wavelength calibration method of the spectrometer provided by the application ensures the continuity of global fitting, and adaptively segments according to the peak density and fitting residual, locally uses cubic or quartic polynomial fitting, significantly improves the local fitting accuracy, significantly reduces the RMSE, enhances the adaptability to local nonlinear sections, and is particularly suitable for complex spectral systems with uneven spectral line density or dramatic response curve fluctuations. The specific process is as follows:

[0062] Characteristic peak matching: For each fitting peak center wavelength, search for candidate matching items within the tolerance range calculated according to the resolution index in the standard spectral line library. According to the calculated resolution matching threshold:

[0063] .

[0064] An intensity-dominant priority strategy is adopted to construct a ranking function containing an intensity weight coefficient of 0.7 and a wavelength deviation weight coefficient of 0.3, and strong peak signals are preferentially selected for matching and locking. The matching process implements bidirectional exclusive constraints to ensure that each detection peak corresponds to a unique standard wavelength and each standard wavelength is matched at most once, eliminating the multiple mapping problem in dense spectral regions.

[0065] Characteristic peak coverage criterion: First, sort the successfully matched characteristic spectral peak data according to their corresponding pixel positions as the basis data for fitting. Then, the entire effective pixel range is preliminarily segmented into multiple candidate sub-sections, initially set to 100-300 pixels per section. Count the number of matched effective standard spectral lines in each section, and dynamically set the minimum number of spectral line coverage according to the length of each section:

[0066] .

[0067] Where: is the current section pixel length; is the average pixel spacing of the characteristic spectral line; is the reference spectral line density, which can be set according to the instrument characteristics, such as 0.01-0.02 / pixel;

[0068] Get , count the number of matched effective standard spectral lines in each section , if , it means that the data in this section is insufficient to support stable polynomial fitting. At this time, merge this section with the one with more spectral lines in its left or right adjacent section until the total number of spectral lines after merging is not less than .

[0069] Fitting residual criterion: for each sub-section that meets the characteristic peak coverage requirement, perform cubic polynomial fitting, and use the least squares method to determine the fitting coefficients , , , , and calculate the RMSE (Root Mean Square Error) of the fitting result. When the RMSE is less than a threshold value, the wavelength value corresponding to the pixel is calculated according to the fitting result to form a pixel-wavelength mapping; when the RMSE is greater than the threshold value, the pixel span of the sub-section is divided into two new sub-sections, and the fitting is performed again. The following adjustment strategy is performed: if the number of pixels in the sub-section is greater than a threshold value, the sub-section is uniformly divided into two new sub-sections, and the fitting is performed again; otherwise, the original section is maintained, the order of the fitting model is increased from three to four, and an L2 regularization term (the hyperparameter λ can be set to 0.0001 to 0.001) is introduced to constrain the fitting parameters to improve stability. Preferably, the RMSE threshold value is set to 0.002-0.005.

[0070] Segmented polynomial fitting: after the above process is completed, all sub-sections are determined as valid fitting sections after judgment, merging, and subdivision operations. A pixel-wavelength mapping model is independently constructed in each section, and a cubic polynomial expression form is usually used as follows:

[0071] .

[0072] wherein x is the pixel position, y is the fitted wavelength value, and a, b, c, and d are the fitting coefficients of the first, second, third, and fourth order terms, respectively. , , , After the fitting is completed, the boundary pixels, corresponding polynomial coefficients, and pixel interval indexes of each section are written into a pixel-wavelength mapping lookup table (LUT) of the spectrometer to realize high-precision conversion from the original pixel position to the physical wavelength value. As shown in FIG. 6, the overall fitting residual and the segmented fitting residual are compared, and the segmented fitting residual provided by the spectrometer wavelength calibration method has a smaller fluctuation range and is closer to zero, verifying the superior performance of the method. Figure 8

[0073] ​​​​​The wavelength calibration method of the LIBS spectrometer based on the standard substance provided in the embodiment of the application generates wide-spectrum dense characteristic spectral lines covering the ultraviolet to near-infrared band by exciting the standard substance, ensures that there are enough calibration points in each sub-band, effectively solves the problems of sparse characteristic spectral lines in the ultraviolet band, difficulty in covering the entire detection band and uneven distribution of calibration points in the prior art, significantly improves the quantitative accuracy of the high-resolution spectrometer in the full-band range, reduces the dependence on specific external standard light sources, and reduces the cost and inconvenience caused by light source replacement and maintenance; the method combining cubic spline interpolation and improved Lorentz fitting is adopted, wherein the cubic spline interpolation effectively reduces the volatility of the spectral peak position, improves the continuity and stability of the spectral data; the improved Lorentz fitting can accurately obtain the peak shape of the spectral peak, ensure that the fitting result truly reflects the physical form of the spectral peak, accurately determine the key parameters such as the peak position, intensity and half-peak width, effectively solve the difficulty in positioning the peak position caused by the volatility of the spectral peak position, the diversity of the peak shape and the uncertain form in the prior art, and significantly improve the accuracy and reliability of the wavelength calibration of the spectrometer; the two-stage fitting technology is adopted, that is, the single-peak model parameters are dynamically initialized first, and then the double-peak model fitting is triggered based on residual analysis, effectively processes the overlapping spectral lines in the spectrum, accurately locks the primary and secondary peak center wavelengths, significantly improves the analysis accuracy of the complex spectrum, enhances the accuracy and stability of the calibration, and effectively solves the problem of decreased calibration accuracy caused by spectral peak overlap in the prior art; the adaptive segmented fitting strategy based on the joint criterion of characteristic peak density and fitting residual is adopted, combined with the cubic or quartic polynomial model and the L2 regular optimization mechanism, the fitting error caused by local nonlinearity is effectively reduced, the overall calibration accuracy and consistency are improved, and the problem of limited calibration accuracy caused by factors such as grating dispersion nonlinearity, detector edge response distortion and optical path geometry deviation in the prior art is solved; by constructing the one-to-one mapping lookup table (LUT) of the pixel position and the physical wavelength and writing it into the device parameter unit, the programmatic calling and real-time compensation of the calibration model are realized, the wavelength conversion in the subsequent spectral detection process is ensured to be efficient and accurate, and the seamless integration of calibration and detection is promoted.

[0074] Although the present application has been disclosed with the preferred embodiments as above, it is not intended to limit the present application, and any person skilled in the art can make some modifications and improvements without departing from the spirit and scope of the present application, therefore the protection scope of the present application shall be defined by the claims.

Claims

1. A standard substance-based LIBS spectrometer wavelength calibration method, characterized in that, The method comprises the following steps: Step S1: According to the detection capability of the spectrometer in the ultraviolet to near infrared band, a plurality of sub-bands are divided, and standard substances capable of generating rich characteristic spectral lines in each sub-band are selected; Step S2: The standard substance is excited by using LIBS technology, the integration start time of the spectrometer is controlled by a time sequence controller, the delay is set to avoid the high-intensity continuous background radiation of the initial plasma, and stable and clear atomic emission spectrum is collected as the original spectral data; Step S3: The original spectral data collected is screened, and the airPLS is used to correct the baseline of the screened spectral data; Step S4: After the baseline correction, the main peak center pixel of each spectral line is taken as the base point, the data in the range of ±3-10 pixels is taken as the window data, the window data is taken as the interpolation input interval, the point interpolation is performed according to 2-5 times the density, the spectral curve is reconstructed by cubic spline interpolation; Step S5: The spectral peak data after the reconstructed spectral curve is applied to the improved two-stage Lorentz fitting algorithm to realize accurate modeling: including the first stage to perform single peak fitting, if the fitting residual exceeds the extreme value of 1.5 times the residual standard deviation, the second stage double peak fitting is triggered; the second peak parameter is dynamically generated by the residual, and the model complexity is controlled by the fitting residual drop rate, and finally the center wavelength, intensity, half peak width and baseline offset value of each characteristic peak are extracted; Step S6: A plurality of groups of spectral curves with peak intensity saturation of 80%-90% are selected, the maximum half peak width value in each group of spectral curves is extracted, and the average value is taken as the resolution evaluation index of the spectrometer; Step S7: Based on the extracted characteristic peak center wavelength and corresponding pixel position, an adaptive segmented polynomial fitting strategy combining characteristic peak density distribution and fitting residual is used to construct a pixel-wavelength mapping model, comprising the following steps: based on the resolution, the characteristic peak matching is locked, the pixel interval corresponding to the successfully matched characteristic spectral peak data is divided into a plurality of sub-sections, and the sub-sections are combined or subdivided according to the characteristic peak coverage rate and the resolution threshold; in each sub-section, polynomial fitting is performed, and when the fitting residual RMSE exceeds the threshold, regularization and / or model order promotion strategy is introduced to optimize the fitting stability; finally, each segment fitting parameter is written into the pixel-wavelength mapping lookup table of the spectrometer, and the full-band high-precision wavelength calibration of the spectrometer is realized.

2. The standard substance based LIBS spectrometer wavelength calibration method of claim 1, wherein, In step S1, the standard substance is a standard metal target copper and tin.

3. The standard-based LIBS spectrometer wavelength calibration method according to claim 1, wherein, In step S2, the original spectral data includes not less than 10 groups of spectral data of each standard substance collected repeatedly at the same excitation position.

4. The standard-based LIBS spectrometer wavelength calibration method of claim 1, wherein, In step S3, the baseline correction of the screened spectral data by using airPLS includes: initializing the baseline as the low-frequency trend of the original spectrum; the influence of the peak region on the baseline fitting is suppressed by iteratively adjusting the weight coefficient, and the smooth baseline signal is obtained and the baseline correction is completed by deducting from the original spectral data.

5. The standard-based LIBS spectrometer wavelength calibration method of claim 1, wherein, In step S5, the first stage performs a single peak fitting including establishing a single peak model based on the local window data, which is mathematically expressed as: where A is the peak intensity, is the center wavelength, is the half-peak width, and B is the baseline offset.

6. The standard-based LIBS spectrometer wavelength calibration method according to claim 5, wherein, In step S5, the second stage double-peak fitting includes a double-peak Lorentz model fitting, the mathematical expression of which is: ; wherein the second peak parameters are dynamically determined by residual extreme value: the initial value of the center wavelength is taken from the position corresponding to the maximum residual value, the initial value of the intensity is set to be about 2 times the residual peak value, and the half-peak width inherits from the main peak parameters.

7. The standard-based LIBS spectrometer wavelength calibration method of claim 1, wherein, Step S6, the average value of the calculation as the resolution of the spectrometer evaluation index, the mathematical expression of the resolution is: ; In the formula The resolution is N, the total number of data sets, The maximum half-peak width of the selected characteristic peak in the i-th data set.

8. The standard-based LIBS spectrometer wavelength calibration method of claim 1, wherein, The feature peak matching locking based on the resolution in step S7 specifically comprises: searching for each fitting peak center wavelength in a standard spectrum line library, and calculating candidate matching items within a tolerance range according to a resolution matching threshold value; wherein the resolution matching threshold value is: ; The sorting function comprising intensity weight coefficient 0.7 and wavelength deviation weight coefficient 0.3 is constructed, and the strong peak signal is preferentially selected for matching and locking; the characteristic peak matching process implements bidirectional exclusive constraint, that is, each fitting peak uniquely corresponds to a standard wavelength, and each standard wavelength is matched at most once.

9. The standard-based LIBS spectrometer wavelength calibration method of claim 8, wherein, In step S7, the characteristic peak coverage criterion is: the entire effective pixel range is preliminarily segmented into multiple candidate sub-sections, and the number of effective standard spectral lines matched in each candidate sub-section is counted , and the minimum spectral line coverage number is dynamically set according to the length of each section , if , it is merged with an adjacent section, and the new sub-section generated after the merging is ; wherein , in the formula: is the current section pixel length; is the average pixel spacing of the characteristic spectral line; is the reference spectral line density.

10. The standard-based LIBS spectrometer wavelength calibration method of claim 8, wherein, In step S7, the fitting residual criterion in the adaptive piecewise polynomial fitting strategy is: performing a cubic polynomial fitting on the sub-segment satisfying the characteristic peak coverage requirement, and calculating the RMSE of the fitting result of the sub-segment. When the RMSE is greater than a set threshold, according to the pixel span of the sub-segment The following adjustment strategy is performed: if the number of pixels in the sub-segment is greater than a set threshold, the sub-segment is uniformly divided into two new sub-segments, and fitting is performed again respectively; otherwise, the original segment is maintained unchanged, the order of the fitting model is increased from cubic to quartic, and an L2 regularization term is introduced to constrain the fitting parameters. The following adjustment strategy is performed: if the number of pixels in the sub-segment is greater than a set threshold, the sub-segment is uniformly divided into two new sub-segments, and fitting is performed again respectively; otherwise, the original segment is maintained unchanged, the order of the fitting model is increased from cubic to quartic, and an L2 regularization term is introduced to constrain the fitting parameters.

Images