A method for removing abnormal data from laser induced breakdown spectroscopy based on spectral feature fusion
By using a spectral feature fusion method to remove outlier data from LIBS spectra, the problem of interference from outlier data in LIBS analysis is solved, achieving higher analytical accuracy and precision.
Patent Information
- Application Number
- CN202210958165.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-11
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-08-11
AI Technical Summary
The presence of numerous anomalous data in LIBS spectral data leads to a decrease in detection accuracy and precision, especially in harsh industrial environments, affecting the analysis of sample composition and content.
A spectral feature fusion-based approach is adopted, which involves acquiring LIBS spectral data matrix, selecting elemental analysis lines, calculating characteristic index values, removing abnormal spectral data, retaining valid spectral data, and calculating average spectral data to improve data stability and analytical accuracy.
It effectively eliminates abnormal spectral data, improves the accuracy and precision of LIBS analysis, ensures the stability of spectral data, and reduces errors.
Smart Images

Figure CN115420715B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of spectrum analysis, and in particular to a method for eliminating abnormal data of laser induced breakdown spectrum based on spectrum feature fusion. Background Art
[0002] Laser-induced breakdown spectroscopy (LIBS) uses high-energy pulsed lasers to focus on the sample surface, excite a small amount of sample at the focus to produce a laser-induced plasma spectrum, and analyze and detect the plasma spectrum through a spectrometer and a detector. The elements contained in the detection object are determined based on the wavelength of the spectrum line in the plasma spectrum data, and the content of the elements is determined based on the intensity of the spectrum line. LIBS can detect almost all elements in the periodic table and can analyze multiple elements at the same time. The analysis is fast and the sample pretreatment is minimal. The objects to be measured can be gases, solids, liquids, aerosols, etc. Therefore, LIBS technology has a huge application market in industrial processes such as metallurgy, coal and cement production. It has been widely used in biomedicine, environmental monitoring and mineral exploration, and has significant advantages in real-time in-situ material analysis and detection in harsh environments such as radiation and high temperature, and unmanned environments such as deep space and deep sea.
[0003] With the continuous advancement of science and technology and the increasing demand for on-site detection applications, LIBS instruments are gradually developing in the direction of miniaturization and portability. Due to the harsh industrial on-site detection environment, such as large instrument vibration, high temperature, large dust, uneven samples, uneven sample surface, low spectral resolution of miniaturized instruments, poor laser stability and other factors, the spectral data detected by LIBS contains a large amount of abnormal data. These abnormal spectral data are not conducive to the subsequent analysis of sample composition and content, greatly increasing the difficulty of LIBS detection samples and reducing the accuracy and precision of sample detection.
[0004] Therefore, it is urgent to propose a laser induced breakdown spectroscopy abnormal data removal method based on spectral feature fusion to effectively suppress the abnormal spectrum in the LIBS spectrum, ensure the stability of LIBS spectral data, and improve the accuracy and precision of LIBS analysis. Summary of the invention
[0005] Aiming at the problem that abnormal spectral data in LIBS spectrum seriously interferes with the accuracy and precision of LIBS analysis, the present invention proposes a laser induced breakdown spectroscopy abnormal data removal method based on spectral feature fusion, which realizes the effective removal of abnormal spectral data, ensures the stability of spectral data, and effectively improves the accuracy and precision of LIBS analysis.
[0006] The present invention adopts the following technical solutions:
[0007] A method for removing abnormal data of laser induced breakdown spectroscopy based on spectral feature fusion specifically comprises the following steps:
[0008] Step 1: Collect and save the LIBS spectrum data matrix S(W m , I n );
[0009] The LIBS spectral data matrix S(W m , I n ) includes n spectra at the same position of the measured sample and m wavelength variables W in each spectrum m , I n is the wavelength variable W of each spectrum m The spectral intensity corresponding to the location;
[0010] Step 2: In the LIBS spectral data matrix S(W m , I n ) select α element analysis lines;
[0011] Step 3, selecting characteristic indices and calculating characteristic index values of each element analysis line at each wavelength variable;
[0012] Step 4, according to the characteristic index value of each element analysis line at each wavelength variable, determine the maximum value and the minimum value of each characteristic index for each element analysis line, and calculate the score of each characteristic index at each wavelength variable of each element analysis line;
[0013] Step 5, based on the scores of all characteristic indicators at each wavelength variable of each element analysis line, calculate the total score of all characteristic indicators at each wavelength variable of each element analysis line;
[0014] Step 6, according to the total score of all characteristic indicators at each wavelength variable of each element analysis line, the abnormal spectrum data in the LIBS spectrum data matrix S of the tested sample is eliminated, and the valid spectrum data in the LIBS spectrum data matrix S of the tested sample is retained to obtain the valid spectrum data matrix of the tested sample;
[0015] Step 7: Calculate the average spectral data of the tested sample based on the effective spectral data matrix of the tested sample.
[0016] Preferably, in step 2, the analysis line of the matrix element in the sample to be tested or the analysis line of other elements in the sample to be tested except the matrix element is selected as the element analysis line, and the element analysis line L(W α , I α ) includes the wavelength W of the element analysis line α and intensity I α .
[0017] Preferably, in step 3, the characteristic indicators include but are not limited to peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio; the characteristic indicators also include analysis line waveform fitting parameters.
[0018] Preferably, in step 4, if a larger characteristic index indicates better spectral data, the characteristic index score calculation formula is:
[0019]
[0020] If the smaller the characteristic index is, the better the spectral data is, then the characteristic index score calculation formula is:
[0021]
[0022] In the formula, G i is the score of the i-th feature index, F i is the value of the i-th characteristic index in the element analysis line, F i,min is the minimum value of the i-th characteristic index in the element analysis line, F i,max is the maximum value of the i-th characteristic index in the element analysis line.
[0023] Preferably, in step 5, the total score calculation formula of all characteristic indicators at each wavelength variable of each element analysis line is:
[0024] G total =∑a i G i (3)
[0025] In the formula, G total G is the total score of all characteristic indicators at any wavelength variable in the element analysis line. i is the score of the i-th characteristic index, a i is the coefficient of the i-th characteristic index.
[0026] Preferably, the step 6 specifically includes the following steps:
[0027] Step 6.1, for each element analysis line, determine the maximum value of the total score of the element analysis line according to the total score of all characteristic indicators at each wavelength variable on the element analysis line;
[0028] Step 6.2, for each element analysis line, setting the ω% of the maximum value of the total score of the element analysis line as the threshold for screening the element analysis line spectrum data;
[0029] Step 6.3, LIBS spectrum data matrix S(W m , I n) are screened. For each element analysis line, according to the threshold of the element analysis line spectrum data screening, the spectrum data of the element analysis line that is not less than the threshold is recorded as 1, and the spectrum data of the element analysis line that is less than the threshold is recorded as 0, until the LIBS spectrum data matrix S(W) is traversed. m , I n ) in all element analysis lines, completing the LIBS spectral data matrix S(W m , I n ) screening of all spectral data;
[0030] Step 6.4, for each element analysis line, sum the number of times all spectral data are recorded on the element analysis line to obtain the total number of times t of spectral data recorded on the element analysis line;
[0031] Step 6.5, set the number threshold t of elemental analysis line spectrum data TH ;
[0032] Step 6.6, using the number threshold t of the elemental analysis line spectrum data TH For the LIBS spectral data matrix S(W m , I n ) The valid data in each element analysis line are screened, and when the total number of spectral data records of the element analysis line t is not less than the number threshold t TH When , the spectral data on the element analysis line is marked as valid spectral data. When the total number of spectral data records t of the element analysis line is less than the number threshold t TH When , the spectral data on the element analysis line is marked as abnormal spectral data;
[0033] Step 6.7, remove the LIBS spectral data matrix S(W m , I n ) Abnormal spectral data in each element analysis line, retain the LIBS spectral data matrix S(W m , I n ) The effective spectral data in each element analysis line is obtained to obtain the effective spectral data matrix of the tested sample.
[0034] Preferably, in step 7, in the process of calculating the average spectrum of the tested sample, n% pieces of effective spectrum data are randomly selected from the effective spectrum data matrix of the tested sample for multiple times and the average value of the effective spectrum data is calculated to obtain the average spectrum data of the corresponding number of times.
[0035] The present invention has the following beneficial effects:
[0036] The present invention is based on the LIBS spectrum data matrix S(W) of the tested sample. m , I n), by fusing multiple characteristic indicators of each spectral data on the element analysis line, the total score of all characteristic indicators at each wavelength variable on the element analysis line is obtained, and according to the total score of all characteristic indicators at each wavelength variable of each element analysis line, the abnormal spectral data in the LIBS spectral data matrix S of the tested sample is eliminated to obtain the effective spectral data of the tested sample, and the effective spectral data matrix is averaged to obtain the average spectral data. The present invention solves the problem that the accuracy and precision of LIBS spectral analysis are seriously affected by abnormal spectral data, realizes the elimination of abnormal spectral data, effectively avoids the errors caused by abnormal spectral data, improves the stability of spectral data, and greatly improves the accuracy and precision of LIBS spectral analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a flow chart of a method for removing abnormal data from laser induced breakdown spectroscopy based on spectral feature fusion.
[0038] Figure 2 The original spectral data.
[0039] Figure 3 This is the spectrum diagram of Fe near 440.475nm in the original spectrum data diagram.
[0040] Figure 4 This is a spectrum graph near Fe 440.475nm in an original spectrum data.
[0041] Figure 5 This is the spectral data diagram after removing abnormal spectral data.
[0042] Figure 6 This is the spectrum near Fe 440.475nm after removing abnormal spectral data
[0043] Figure 7 is the standard deviation of the original spectral data and the spectral data after removing abnormal spectral data.
[0044] Figure 8 Calculate the standard deviation after averaging the original spectral data and the spectral data after removing abnormalities.
[0045] Fig. 9 is a calibration curve established based on the original spectral data.
[0046] Fig.10 It is a calibration curve established based on the spectral data after removing abnormal spectra. DETAILED DESCRIPTION
[0047] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings.
[0048] Example 1
[0049] When the sample is detected based on LIBS, there are abnormal spectral data in the spectral data obtained at the same position of the sample, which seriously affects the analysis result of LIBS spectral analysis. It is necessary to remove the abnormal spectral data in the LIBS spectral data and retain the valid spectral data in the LIBS spectral data. Therefore, the present invention proposes a laser induced breakdown spectroscopy abnormal data removal method based on spectral feature fusion, such as Figure 1 As shown, the specific steps include:
[0050] Step 1: Collect and save the LIBS spectrum data matrix S(W m , I n ).
[0051] The LIBS spectral data matrix S(W m , I n ) includes n spectra at the same position of the measured sample and m wavelength variables W in each spectrum m , I n is the wavelength variable W of each spectrum m The spectral intensity corresponding to the location.
[0052] Step 2: In the LIBS spectral data matrix S(W m , I n ) select α element analysis lines.
[0053] Select the analysis line of the matrix element in the sample being tested or the analysis line of other elements in the sample being tested except the matrix element as the element analysis line, and select the element analysis line manually or automatically. The element analysis line L(W α , I α ) includes the wavelength W of the element analysis line α and intensity I α .
[0054] Step 3: Select characteristic indicators and calculate the characteristic indicator values of each element analysis line at each wavelength variable.
[0055] The characteristic indicators include but are not limited to peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio. The characteristic indicators may also include analysis line waveform fitting parameters, such as the determination coefficient of the analysis line waveform fitting, the half-height width of the spectrum line, etc.
[0056] Step 4: According to the characteristic index value of each element analysis line at each wavelength variable, the maximum value and the minimum value of each characteristic index are determined for each element analysis line, and the score of each characteristic index at each wavelength variable of each element analysis line is calculated.
[0057] According to the relationship between the characteristic index changes and the spectral data, the score calculation formula of the characteristic index is selected.
[0058] If the larger the characteristic index is, the better the spectral data is, then the characteristic index score calculation formula is:
[0059]
[0060] If the smaller the characteristic index is, the better the spectral data is, then the characteristic index score calculation formula is:
[0061]
[0062] In the formula, G i is the score of the i-th feature index, F i is the value of the i-th characteristic index in the element analysis line, F i,min is the minimum value of the i-th characteristic index in the element analysis line, F i,max is the maximum value of the i-th characteristic index in the element analysis line.
[0063] Step 5: Based on the scores of all characteristic indicators at each wavelength variable of each element analysis line, the total score of all characteristic indicators at each wavelength variable of each element analysis line is calculated, as shown in formula (3):
[0064] G total =∑a i G i (3)
[0065] In the formula, G total G is the total score of all characteristic indicators at any wavelength variable in the element analysis line. i is the score of the i-th characteristic index, a i is the coefficient of the i-th characteristic index.
[0066] Step 6, according to the total score of all characteristic indicators at each wavelength variable of each element analysis line, remove the abnormal spectral data in the LIBS spectral data matrix S of the tested sample, retain the valid spectral data in the LIBS spectral data matrix S of the tested sample, and obtain the valid spectral data matrix of the tested sample, which specifically includes the following steps:
[0067] Step 6.1, for each element analysis line, determine the maximum value of the total score of the element analysis line according to the total score of all characteristic indicators at each wavelength variable on the element analysis line.
[0068] Step 6.2, for each element analysis line, set the ω% of the maximum value of the total score of the element analysis line as the threshold for screening the element analysis line spectrum data.
[0069] Step 6.3, LIBS spectrum data matrix S(W m , I n) are screened. For each element analysis line, according to the threshold of the element analysis line spectrum data screening, the spectrum data of the element analysis line that is not less than the threshold is recorded as 1, and the spectrum data of the element analysis line that is less than the threshold is recorded as 0, until the LIBS spectrum data matrix S(W) is traversed. m , I n ) in all element analysis lines, completing the LIBS spectral data matrix S(W m , I n ) to filter all spectral data.
[0070] Step 6.4, for each element analysis line, sum the number of times all spectral data are recorded on the element analysis line to obtain the total number of times t of spectral data records of the element analysis line.
[0071] Step 6.5, set the number threshold t of elemental analysis line spectrum data TH .
[0072] Step 6.6, using the number threshold t of the elemental analysis line spectrum data TH For the LIBS spectral data matrix S(W m , I n ) The valid data in each element analysis line are screened, and when the total number of spectral data records of the element analysis line t is not less than the number threshold t TH When , the spectral data on the element analysis line is marked as valid spectral data. When the total number of spectral data records t of the element analysis line is less than the number threshold t TH , the spectral data on the element analysis line is marked as abnormal spectral data.
[0073] Step 6.7, remove the LIBS spectral data matrix S(W m , I n ) Abnormal spectral data in each element analysis line, retain the LIBS spectral data matrix S(W m , I n ) The effective spectral data in each element analysis line is obtained to obtain the effective spectral data matrix of the tested sample.
[0074] Step 7: Calculate the average spectral data of the tested sample based on the effective spectral data matrix of the tested sample.
[0075] In the process of calculating the average spectrum of the tested sample in this embodiment, n% pieces of effective spectrum data are randomly selected from the effective spectrum data matrix of the tested sample for multiple times and the average value of the effective spectrum data is calculated to obtain the average spectrum data of the corresponding number of times.
[0076] In this embodiment, it may also include:
[0077] Step 8, analyzing the processing effect of the method of the present invention, by comparing the spectral graphs of the original spectral data before processing and the spectral data after the abnormal spectral data is eliminated by the present invention, comparing the standard deviations of the original spectral data and the spectral data after the abnormal spectral data is eliminated, and establishing a calibration curve, using the determination coefficient of the calibration curve to compare the quantitative analysis capabilities of the original spectral data and the spectral data after the abnormal spectral data is eliminated.
[0078] Example 2
[0079] The laser induced breakdown spectroscopy abnormal data elimination method based on spectral feature fusion proposed in the present invention is applied to actual LIBS spectral analysis and processing, specifically comprising the following steps:
[0080] Step 1: Collect and save the LIBS spectrum data matrix S(W 4094 , I 500 ).
[0081] In this embodiment, the tested samples are 9 kinds of steel samples, and the steel samples are numbered 1# to 9#. Five different positions of each steel sample were tested, and 500 spectra were collected at each position. Each spectrum was collected using a dual-channel spectrometer. The spectral data of each channel consists of 2047 wavelength values, and the wavelength values of each spectrum cover 278.587nm to 484.529nm. The LIBS spectrum of steel sample 1# collected is shown in Figure 1. Figure 2 As shown, the spectrum of Fe in the range of 439.059nm to 441.297nm near 440.475nm is as follows: Figure 3 shown.
[0082] in, Figure 2 The spectral data matrix S(W 4094 , I 500 ) consists of 500 spectra at a certain position of steel sample 1#, where S represents the spectral data matrix, W 4094 Corresponding to 4094 wavelength variables from 278.587nm to 484.529nm, I 500 Represents the spectral intensity of 500 spectra at 4094 wavelength variables.
[0083] Step 2: In the LIBS spectral data matrix S(W 4094 , I 500 ) to select multiple element analysis lines.
[0084] In this embodiment, 36 analysis lines of 7 elements including Fe, Si, Mn, Cr, Mo, Ni and Cu were selected, as shown in Table 1.
[0085] Table 1 36 analysis lines selected from spectral data
[0086]
[0087] Step 3: Select characteristic indicators and calculate the characteristic indicator values of each element analysis line at each wavelength variable.
[0088] In this embodiment, peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio are selected as characteristic indicators. Figure 4 Content, taking the analysis line Fe 440.475nm as an example, the peak intensity SI is the intensity value at the center wavelength of the analysis line Fe 440.475nm, which is Figure 4 The intensity value I at point D D ; The peak area SPA is the spectral band λ where the central wavelength Fe 440.475nm is located C ~λ E The area covered; the standard deviation SD is the spectral band λ where the central wavelength Fe 440.475nm is located C ~λ E The standard deviation of the intensity values corresponding to all wavelengths in , the calculation of the signal-to-background ratio SBR and the signal-to-noise ratio SNR in this embodiment is shown in formula (4) and formula (5).
[0089]
[0090]
[0091] In formula (4) and formula (5), I A ,I B ,I C ,I E ,I F ,I G They correspond to the intensity values of the three background points on the left and right near the selected central wavelength D point, mean(I A ,I B ,I C ,I E ,I F ,I G ) is the average of the intensity values at the six background points, std(I A ,I B ,I C ,I E ,I F ,I G ) is the standard deviation of the intensity values at the six background points.
[0092] Step 4: According to the characteristic index value of each element analysis line at each wavelength variable, the maximum value and the minimum value of each characteristic index are determined for each element analysis line, and the score of each characteristic index at each wavelength variable of each element analysis line is calculated.
[0093] For the characteristic indicators in this embodiment, the larger the peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio, the better the spectral data. Therefore, the score value of the characteristic indicator is calculated using formula (1). The peak intensity scores of 36 element analysis lines selected from 500 spectra at a certain position of steel sample 1# are shown in Table 2.
[0094] Table 2 The scores of the peak intensity of the analysis line
[0095]
[0096] Step 5: Based on the scores of all characteristic indicators at each wavelength variable of each element analysis line, the total score of all characteristic indicators at each wavelength variable of each element analysis line is calculated. The total score calculation formula of all characteristic indicators at each wavelength variable of each element analysis line is:
[0097] G total =∑a i G i (3)
[0098] In the formula, G total G is the total score of all characteristic indicators at any wavelength variable in the element analysis line. i is the score of the i-th characteristic index, a i is the coefficient of the i-th characteristic index. i It can express the fractional value of peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio. i Represents the coefficient of each fractional value. In this embodiment, a i =1, and the total score of all characteristic indicators of 36 element analysis lines selected from 500 spectra at a certain position of steel sample 1# is calculated, as shown in Table 3.
[0099] Table 3 Total scores of all characteristic indicators
[0100]
[0101] Step 6, according to the total score of all characteristic indicators at each wavelength variable of each element analysis line, remove the abnormal spectral data in the LIBS spectral data matrix S of the tested sample, retain the valid spectral data in the LIBS spectral data matrix S of the tested sample, and obtain the valid spectral data matrix of the tested sample, which specifically includes the following steps:
[0102] Step 6.1, for each element analysis line, determine the maximum value of the total score of the element analysis line according to the total score of all characteristic indicators at each wavelength variable on the element analysis line.
[0103] Step 6.2: for each element analysis line, set 60% of the maximum value of the total score of the element analysis line as the threshold for screening the element analysis line spectrum data.
[0104] Step 6.3, LIBS spectrum data matrix S(W 4094 , I 500 ) are screened. For each element analysis line, according to the threshold of the element analysis line spectrum data screening, the spectrum data of the element analysis line that is not less than the threshold is recorded as 1, and the spectrum data of the element analysis line that is less than the threshold is recorded as 0, until the LIBS spectrum data matrix S(W) is traversed. 4094 , I 500 ) in all element analysis lines, completing the LIBS spectral data matrix S(W 4094 , I 500 ) to filter all spectral data.
[0105] Step 6.4, for each element analysis line, sum the number of times all spectral data are recorded on the element analysis line to obtain the total number of times t of spectral data records of the element analysis line.
[0106] Step 6.5, set the number threshold t of elemental analysis line spectrum data TH In this embodiment, t TH =22.
[0107] Step 6.6, using the number threshold t of the elemental analysis line spectrum data TH For the LIBS spectral data matrix S(W 4094 , I 500 ) The valid data in each element analysis line are screened, and when the total number of spectral data records of the element analysis line t is not less than the number threshold t TH When , the spectral data on the element analysis line is marked as valid spectral data. When the total number of spectral data records t of the element analysis line is less than the number threshold t TH , the spectral data on the element analysis line is marked as abnormal spectral data.
[0108] Step 6.7, remove the LIBS spectral data matrix S(W 4094 , I 500 ) Abnormal spectral data in each element analysis line, retain the LIBS spectral data matrix S(W 4094 , I 500 ) The effective spectral data in each element analysis line is obtained to obtain the effective spectral data matrix of the tested sample.
[0109] Step 7, calculate the average spectrum data of the tested sample according to the effective spectrum data matrix of the tested sample. In this embodiment, the average spectrum data is calculated 50 times, and 20% of the spectra in the effective spectrum data matrix are randomly selected each time for averaging to obtain 50 average spectrum data.
[0110] Step 8, analyzing the processing effect of the laser induced breakdown spectroscopy abnormal data removal method based on spectral feature fusion of the present invention.
[0111] In this embodiment, the spectrum of the original spectrum data is as follows: Figure 2 As shown, the spectrum of the spectral data after removing the abnormal spectral data is as follows Figure 5 As shown, the spectrum of the original spectrum data near Fe 440.475nm is as follows Figure 3 As shown in the figure, the spectrum of Fe 440.475nm after removing the abnormal spectrum data is as follows Figure 6 As shown in the figure, it can be seen that after comparison, the spectral distribution after removing the abnormal spectral data is more concentrated. The standard deviation of the original spectral data and the spectral data after removing the abnormal spectral data is shown in Figure 7 As shown in the figure, it can be seen that the standard deviation of the spectral data is significantly reduced after removing the abnormal spectral data.
[0112] For steel samples 1# to 9#, each sample has 5 groups of spectral data, and each group of original spectral data has 500 spectra. For each group of data, 20% of the spectra are randomly selected for averaging, and the cycle is repeated 50 times. 50 average spectral data are obtained for each group of data. For the valid spectral data after deleting the abnormal spectral data, 20% of the spectra are randomly selected for averaging, and the cycle is repeated 50 times to calculate 50 average spectral data. The standard deviation of the average spectrum before and after the abnormal spectral data is removed at a certain detection position of steel sample 1# is as follows Figure 8 As shown by Figure 8 It can be seen that the standard deviation of the average spectrum is significantly reduced after removing the abnormal spectral data. Figure 7 and Figure 8 The standard deviation can be further reduced by averaging the spectrum. The original spectrum data and the effective spectrum data after removing the abnormal spectrum data are averaged and a calibration curve for quantitative analysis is established, such as Fig. 9 and Fig.10 Analysis Fig. 9 and Fig.10 It can be obtained that the determination coefficient of the calibration curve established based on the original spectral data is 0.9481, while the determination coefficient of the calibration curve established based on the effective spectral data after eliminating the abnormal spectral data is 0.9654, which verifies that the laser induced breakdown spectroscopy abnormal data elimination method based on spectral feature fusion proposed in the present invention can effectively improve the accuracy of model quantitative analysis.
[0113] In order to solve the problem that abnormal spectral data seriously interfere with the accuracy and precision of LIBS analysis, the present invention adopts a laser induced breakdown spectroscopy abnormal data elimination method based on spectral feature fusion to achieve effective elimination of abnormal spectral data, which not only effectively improves the stability of spectral data, but also greatly improves the accuracy and precision of LIBS analysis.
[0114] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.
Claims
1. A method for removing abnormal data from laser induced breakdown spectroscopy based on spectral feature fusion, characterized in that: The specific steps include: Step 1: Collect and save the LIBS spectrum data matrix S(W m , I n ); The LIBS spectral data matrix S(W m , I n ) includes n spectra at the same position of the measured sample and m wavelength variables W in each spectrum m , I n is the wavelength variable W of each spectrum m The spectral intensity corresponding to the location; Step 2: In the LIBS spectral data matrix S(W m , I n ) select α element analysis lines; Step 3, select characteristic indexes and calculate the characteristic index values of each element analysis line at each wavelength variable; Step 4, according to the characteristic index value of each element analysis line at each wavelength variable, determine the maximum value and the minimum value of each characteristic index for each element analysis line, and calculate the score of each characteristic index at each wavelength variable of each element analysis line; Step 5, based on the scores of all characteristic indicators at each wavelength variable of each element analysis line, calculate the total score of all characteristic indicators at each wavelength variable of each element analysis line; Step 6, according to the total score of all characteristic indicators at each wavelength variable of each element analysis line, the abnormal spectrum data in the LIBS spectrum data matrix S of the tested sample is eliminated, and the valid spectrum data in the LIBS spectrum data matrix S of the tested sample is retained to obtain the valid spectrum data matrix of the tested sample; Step 7, calculating the average spectrum data of the tested sample according to the effective spectrum data matrix of the tested sample; In step 2, the analysis line of the matrix element in the sample to be tested or the analysis line of other elements in the sample to be tested except the matrix element is selected as the element analysis line. The element analysis line L(W α , I α ) includes the wavelength W of the element analysis line α and intensity I α ; In step 3, the characteristic indicators include but are not limited to peak intensity, peak area, standard deviation, signal-to-background ratio and signal-to-noise ratio; the characteristic indicators also include analysis line waveform fitting parameters; In step 4, if the larger the characteristic index is, the better the spectral data is, then the characteristic index score calculation formula is: If the smaller the characteristic index is, the better the spectral data is, then the characteristic index score calculation formula is: In the formula, G i is the score of the i-th feature index, F i is the value of the i-th characteristic index in the same element analysis line, F i,min is the minimum value of the i-th characteristic index in the same element analysis line among n spectra, F i,max It is the maximum value of the i-th characteristic index in the n spectra of the same element analysis line; In step 5, the total score calculation formula of all characteristic indicators at each wavelength variable of each element analysis line is: G total =Σa i G i (3) In the formula, G total G is the total score of all characteristic indicators at any wavelength variable in the element analysis line. i is the score of the i-th characteristic index, a i is the coefficient of the i-th characteristic index; The step 6 specifically includes the following steps: Step 6.1, for each element analysis line, determine the maximum value of the total score of the element analysis line according to the total score of all characteristic indicators at each wavelength variable on the element analysis line; Step 6.2, for each element analysis line, setting the ω% of the maximum value of the total score of the element analysis line as the threshold for screening the element analysis line spectrum data; Step 6.3, LIBS spectrum data matrix S(W m , I n ) are screened. For each element analysis line, according to the threshold of the element analysis line spectrum data screening, the spectrum data of the element analysis line that is not less than the threshold is recorded as 1, and the spectrum data of the element analysis line that is less than the threshold is recorded as 0, until the LIBS spectrum data matrix S(W) is traversed. m , I n ) in all element analysis lines, completing the LIBS spectral data matrix S(W m , I n ) screening of all spectral data; Step 6.4, for each element analysis line, sum the number of times all spectral data are recorded on the element analysis line to obtain the total number of times t of spectral data recorded on the element analysis line; Step 6.5, set the number threshold t of elemental analysis line spectrum data TH ; Step 6.6, using the number threshold t of the elemental analysis line spectrum data TH For the LIBS spectral data matrix S(W m , I n ) The valid data in each element analysis line are screened, and when the total number of spectral data records t of the element analysis line is not less than the number threshold t TH When , the spectral data on the element analysis line is marked as valid spectral data. When the total number of spectral data records t of the element analysis line is less than the number threshold t TH When , the spectral data on the element analysis line is marked as abnormal spectral data; Step 6.7, remove the LIBS spectral data matrix S(W m , I n ) Abnormal spectral data in each element analysis line, retain the LIBS spectral data matrix S(W m , I n ) The effective spectral data in each element analysis line is obtained to obtain the effective spectral data matrix of the tested sample.
2. The method for removing abnormal data of laser induced breakdown spectroscopy based on spectral feature fusion according to claim 1 is characterized in that: In step 7, during the process of calculating the average spectrum of the tested sample, n% of the valid spectrum data are randomly selected from the valid spectrum data matrix of the tested sample for multiple times and the average value of the valid spectrum data is calculated to obtain the average spectrum data of the corresponding number of times.
