Map fusion deviation-error correction method based on deep learning

Abstract

The invention discloses a map fusion deviation-error spectrum correction method based on deep learning. The method comprises the following steps: 1) acquiring an LIBS spectrum and a plasma image of a sample under the same time window; 2) extracting spectral line intensity of an analysis element from the LIBS spectrum; 3) calculating standard spectral intensity in an ideal plasma state based on a plasma radiation formula; 4) calculating a spectral deviation and a spectral error based on the actual spectral intensity and the standard spectral intensity; 5) estimating the spectral deviation and the spectral error based on the LIBS spectrum and the plasma image through a deep learning method; 6) preliminarily correcting the spectral intensity according to the estimated spectral deviation; 7) further correcting the spectral intensity according to the estimated spectral error; 8) performing normalization processing on the corrected spectral intensity; (9) establishing a calibration curve of the element content in the sample based on the corrected normalized spectral intensity; and 9) correcting the spectrum of the unknown sample based on the steps, and predicting the element content of the unknown sample.

Description

Technical Field

[0001] The present invention belongs to the technical field of LIBS element analysis, and more specifically, relates to a spectrum fusion deviation-error correction method based on deep learning. Background Art

[0002] Laser-induced breakdown spectroscopy (LIBS) is an emerging element analysis technology based on atomic emission spectroscopy theory. It can identify the type and content of elements in a sample by analyzing the plasma radiation spectrum generated by ultrashort pulse laser ablation of the sample. Since its advent, it has been widely used in metallurgy, environment, coal, biomedicine, aerospace and other fields due to its advantages of rapid, in-situ, remote, low-loss, and full-element synchronous analysis.

[0003] However, the poor spectral stability and low quantitative accuracy of LIBS technology have always been one of the key obstacles to its further development and application. In response to this technical problem, domestic and foreign researchers have proposed a series of solutions, which can be summarized into two categories: experimental methods and data processing methods. Among them, the experimental methods mainly include sample pretreatment, parameter optimization, dual pulse method, spatial constraint and beam shaping, etc. Although they can improve the spectral stability to a certain extent, the disadvantages of expensive price, complex instruments, cumbersome operation and harsh environmental requirements make LIBS technology lose its advantages such as in-situ, fast and real-time, so it is not suitable for online monitoring; while the data processing methods mainly include internal standard method, spectral intensity normalization method, plasma parameter correction and external reference signal correction method, etc. These methods can not only effectively improve the spectral stability, but also are suitable for in-situ online monitoring, but due to the cumbersome calculation process and unclear physical meaning, there is still room for further improvement. Summary of the invention

[0004] In view of the problems existing in the prior art methods, the present invention provides a spectrum fusion deviation-error spectrum correction method and application based on deep learning to solve the technical problems of poor spectral stability and low quantitative accuracy of the existing LIBS technology.

[0005] The solution of the present invention to solve the above technical problems is as follows:

[0006] A spectrum fusion bias-error spectrum correction method based on deep learning, comprising the following steps:

[0007] S1. Simultaneously obtain the LIBS spectrum and plasma image of the sample to be tested in the same time window;

[0008] S2, extracting the spectral line intensity of the analyzed element from the acquired LIBS spectrum;

[0009] S3. Based on the plasma radiation formula, calculate the standard spectrum intensity under the ideal plasma state;

[0010] S4, calculating the spectrum deviation and the spectrum error based on the actual spectrum intensity and the standard spectrum intensity;

[0011] S5. Estimate spectral deviation and spectral error based on LIBS spectrum and plasma image by deep learning method;

[0012] S6. Preliminarily correcting the spectral intensity according to the estimated spectral deviation;

[0013] S7, further correcting the spectral intensity according to the estimated spectral error;

[0014] S8, normalizing the corrected spectral intensity;

[0015] S9, establishing a calibration curve related to the element content in the sample based on the corrected normalized spectral intensity;

[0016] S10, calibrating the spectrum of the unknown sample according to steps S1-S8, and predicting the content of the element to be analyzed in the unknown sample based on the calibration curve established in step S9.

[0017] As a further improvement of the present invention, in step S3, the standard spectrum intensity under the ideal plasma state is calculated, and its expression formula is as follows:

[0018]

[0019] Among them, I ideal is the standard spectrum intensity; F0 is the instrument and environmental factors under the ideal plasma state; C0 is the content of the element to be analyzed in the sample; l0 is the radiation path length under the ideal plasma state; g i , A ij and E i are degeneracy, transition probability and upper level energy respectively; T0 is the plasma temperature under ideal plasma conditions; U(T0) is the partition function under ideal plasma conditions; k is the Boltzmann constant; n e0 is the electron number density under ideal plasma state; m e is the mass of the electron; h is Planck's constant; E ion ΔE is the ionization energy of the neutral atom of the element to be analyzed; ion is the ionization potential reduction factor.

[0020] As a further improvement of the present invention, in step S4, the calculation formulas of the spectral deviation and the spectral error are respectively as follows:

[0021]

[0022] Among them, e b is the spectral deviation, which is defined as the difference between the actual measured spectral intensity and the average spectral intensity of multiple measurements; e e is the spectral error, which is defined as the difference between the average spectral intensity of multiple measurements and the standard spectral intensity under the ideal plasma state; I real is the actual spectral intensity; is the average spectral intensity.

[0023] As a further improvement of the present invention, in step S5, the estimation formulas of the spectral deviation and the spectral error are respectively as follows:

[0024]

[0025] in, and are the estimated spectral bias and spectral error, respectively; p pi is the plasma image feature; p s LIBS spectral characteristics.

[0026] As a further improvement of the present invention, in step S6, the spectral intensity of the preliminary correction is:

[0027]

[0028] Among them, I corr1 is the spectral intensity of the preliminary correction; I real is the actual spectral intensity; is the estimated spectral deviation; p pi is the plasma image feature; p s LIBS spectral characteristics.

[0029] As a further improvement of the present invention, in step S7, the spectral intensity to be further corrected is:

[0030]

[0031] Among them, I corr2 is the spectral intensity for further correction; I corr1 is the spectral intensity of the preliminary correction; I real is the actual spectral intensity; is the estimated spectral deviation; is the estimated spectral error; p pi is the plasma image feature; p s LIBS spectral characteristics.

[0032] As a further improvement of the present invention, in step S8, the formula for normalization processing is:

[0033]

[0034] Among them, I c ' orr2 is the normalized spectral intensity; I corr2 is the spectral intensity for further correction.

[0035] As a further improvement of the present invention, in step S9, the formula of the calibration curve is:

[0036] C s =kI c ' orr2 +b

[0037] Among them, C s is the element content in the sample; I c ' orr2 is the normalized spectral intensity.

[0038] The beneficial effects of the present invention are:

[0039] (1) Information fusion: The present invention uses both LIBS spectrum and plasma image signals, and uses a deep learning model to fully explore and utilize the effective features of both, providing richer information for the spectral correction process;

[0040] (2) Model fusion: The present invention firstly derives the standard spectrum intensity under the ideal plasma state based on the plasma radiation physics model, and further restores the actual spectrum intensity to the spectrum under the ideal plasma state with the help of external reference signals through the deep learning model, which has clear physical meaning and improves the reliability of the spectrum correction method;

[0041] (3) The present invention only requires an external plasma image acquisition module on the traditional LIBS optical path. The overall device is very simple and easy to operate. It does not require complicated parameters and a large amount of calculations. The method is simple and easy to implement. At the same time, it does not require complex sample pretreatment processes. Therefore, the spectral intensity after correction by this method is highly stable and robust, and is very suitable for application in complex industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1A flowchart of a method for correcting LIBS spectral fluctuation by plasma image-spectrum fusion provided in an embodiment of the present invention;

[0044] Figure 2 Typical LIBS spectra and plasma images of aluminum alloys obtained under different laser energies and defocus amounts provided in the embodiments of the present invention; wherein (a) is a LIBS spectra, and (b) is a plasma image;

[0045] Figure 3 A comparison diagram of calibration experimental results before and after correction of Mg, Mn and Si elements in the aluminum alloy sample provided in an embodiment of the present invention; wherein, (a) is the calibration curve before Mg element correction, (b) is the calibration curve before Mn element correction, (c) is the calibration curve before Si element correction, (d) is the calibration curve after Mg element correction, (e) is the calibration curve after Mn element correction, and (f) is the calibration curve after Si element correction. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0048] The present invention aims to solve the problems of drastic spectral fluctuations and poor quantitative accuracy caused by changes in experimental parameters such as laser energy and defocus in LIBS technology. The plasma image and LIBS spectrum in the same time window are simultaneously collected, and the effective features in the plasma image and LIBS spectrum are fully mined based on the deep learning method. Based on this, an estimation model of spectral deviation and spectral error is established, thereby realizing the correction of spectral intensity.

[0049] like Figure 1 As shown, after the method of the present invention is started, the plasma image and LIBS spectrum are simultaneously acquired through the LIBS system, and the spectrum deviation and spectrum error are further estimated based on the deep learning method, and finally the intensity of the spectrum to be analyzed is corrected. The specific steps are as follows:

[0050] Step 1: Simultaneously collect LIBS spectra and plasma images of aluminum alloy samples at different laser energies (58.1mJ, 60.4mJ, and 63.0mJ) and defocus distances (+1.5mm, 0mm, and -1.5mm), such as Figure 2As shown. The change of laser energy is achieved by adjusting the energy attenuator; the change of defocus is achieved by adjusting the height of the sample stage;

[0051] Step 2: Extract the actual spectral intensities of the three spectral lines of Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm from the LIBS spectrum of the aluminum alloy sample. real .

[0052] Step 3: According to the plasma radiation model, the standard spectral intensities of the three spectral lines of MgI 516.73nm, MnII 294.92nm and SiI 288.16nm under the ideal plasma state are calculated. The expression formula is as follows:

[0053]

[0054] Among them, F0 is the instrument and environmental factors under the ideal plasma state, that is, the average value of the instrument and environmental factors obtained based on the aluminum alloy sample; C0 is the element content in the sample under the ideal plasma state, that is, the actual content of Mg, Mn and Si elements in the aluminum alloy sample; T0 and n e0 are the plasma temperature and electron number density under ideal plasma conditions, defined as the average values ​​of the plasma temperature and electron number density obtained from the aluminum alloy sample; l0 is the radiation path length under ideal plasma conditions, defined as the unit length.

[0055] Step 4: Actual spectral intensity I based on Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm real and the standard spectral intensity I ideal , calculate the spectral deviation and spectral error of the three spectral lines. First, calculate the spectral deviation e b , and its calculation formula is as follows:

[0056]

[0057] Furthermore, the spectral errors e of the three spectral lines Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm were calculated. e , and its calculation formula is as follows:

[0058]

[0059] Step 5: Estimate the spectral deviation and spectral error of the three spectral lines of Mg I 516.73nm, MnII 294.92nm and Si I 288.16nm based on LIBS spectra and plasma images through deep learning methods:

[0060]

[0061] Step 6: Based on the estimated spectral deviation Preliminary calibration of the spectral intensities of the three spectral lines Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm:

[0062]

[0063] Step 7: Based on the estimated spectral error Further calibrate the spectral intensities of the three spectral lines of Mg I 516.73nm, Mn II 294.92nm and SiI 288.16nm:

[0064]

[0065] Step 8: Normalize the spectral intensities of the corrected Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm lines:

[0066]

[0067] Step 9: Establish the normalized spectral intensity I of the three spectral lines of Mg I 516.73nm, Mn II 294.92nm and Si I 288.16nm after correction c ' orr2 The content of Mg, Mn and Si in the sample s The calibration curve of:

[0068] C s =kI c ' orr2 +b

[0069] Step 10: After calibrating the spectra of the three spectral lines of Mg I 516.73nm, Mn II 294.92nm and SiI 288.16nm of the unknown sample according to steps 1-8, predict the contents of Mg, Mn and Si elements in the unknown sample based on the calibration curve established in step 9.

[0070] According to the above steps, a spectral intensity correction model was established to correct the spectral intensities of the three characteristic spectral lines of Mg I516.73nm, Mn II294.92nm and Si I 288.16nm of 12 aluminum alloy samples collected at laser energies of 58.1mJ, 60.4mJ and 63.0mJ and defocus amounts of +1.5mm, 0mm and -1.5mm. The corrected spectral line intensities were used to calibrate the three elements Mg, Mn and Si.

[0071] The results of the element calibration curve established using the original spectral intensity and the corrected spectral intensity are shown in Figure 3 As shown in Figure 2, the R value of the calibration curve established by the spectral intensity after correction by the spectrum fusion bias-error spectral correction method based on deep learning is 2 The performance indicators such as RMSE, STD of the training set and RMSE, STD of the prediction set are significantly better than the calibration curve established by the original spectral intensity. For more specific indicator comparison, see the table below:

[0072]

[0073] The technical means disclosed in the present invention are not limited to the technical means disclosed in the above-mentioned embodiments, but also include technical solutions composed of any combination of the above-mentioned technical features. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also regarded as the protection scope of the present invention.

Claims

1. A spectrum fusion deviation-error spectrum correction method based on deep learning, characterized in that: The following steps are involved: S1. Simultaneously obtain the LIBS spectrum and plasma image of the sample to be tested in the same time window; S2, extracting the spectral line intensity of the analyzed element from the acquired LIBS spectrum; S3. Based on the plasma radiation formula, calculate the standard spectrum intensity under the ideal plasma state; S4, calculating the spectrum deviation and the spectrum error based on the actual spectrum intensity and the standard spectrum intensity; S5. Estimate spectral deviation and spectral error based on LIBS spectrum and plasma image by deep learning method; S6. Preliminarily correcting the spectral intensity according to the estimated spectral deviation; S7, further correcting the spectral intensity according to the estimated spectral error; S8, normalizing the corrected spectral intensity; S9, establishing a calibration curve related to the element content in the sample based on the corrected normalized spectral intensity; S10, calibrating the spectrum of the unknown sample according to steps S1-S8, and predicting the content of the element to be analyzed in the unknown sample based on the calibration curve established in step S9.

2. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S3, the standard spectrum intensity under the ideal plasma state is calculated, and the expression formula is as follows: Among them, I ideal is the standard spectrum intensity; F0 is the instrument and environmental factors under the ideal plasma state; C0 is the content of the element to be analyzed in the sample; l0 is the radiation path length under the ideal plasma state; g i , A ij and E i are degeneracy, transition probability and upper level energy respectively; T0 is the plasma temperature under ideal plasma conditions; U(T0) is the partition function under ideal plasma conditions; k is the Boltzmann constant; n e0 is the electron number density under ideal plasma state; m e is the mass of the electron; h is Planck's constant; E ion ΔE is the ionization energy of the neutral atom of the element to be analyzed; ion is the ionization potential reduction factor.

3. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S4, the calculation formulas for the spectral deviation and the spectral error are respectively as follows: Among them, e b is the spectral deviation, which is defined as the difference between the actual measured spectral intensity and the average spectral intensity of multiple measurements; e e is the spectral error, which is defined as the difference between the average spectral intensity of multiple measurements and the standard spectral intensity under the ideal plasma state; I real is the actual spectral intensity; is the average spectral intensity.

4. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S5, the estimation formulas of spectral deviation and spectral error are respectively as follows: in, and are the estimated spectral bias and spectral error, respectively; p pi is the plasma image feature; p s LIBS spectral characteristics.

5. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S6, the spectral intensity of the preliminary correction is: Among them, I corr1 is the spectral intensity of the preliminary correction; I real is the actual spectral intensity; is the estimated spectral deviation; p pi is the plasma image feature; p s LIBS spectral characteristics.

6. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S7, the spectral intensity to be further corrected is: Among them, I corr2 is the spectral intensity for further correction; I corr1 is the spectral intensity of the preliminary correction; I real is the actual spectral intensity; is the estimated spectral deviation; is the estimated spectral error; p pi is the plasma image feature; p s LIBS spectral characteristics.

7. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S8, the formula for normalization processing is: Among them, I' corr2 is the normalized spectral intensity; I corr2 is the spectral intensity for further correction.

8. The method for spectrum fusion deviation-error spectrum correction based on deep learning according to claim 1, characterized in that: In step S9, the formula of the calibration curve is: C s =kI' corr2 +b Among them, C s is the element content in the sample; I' corr2 is the normalized spectral intensity.

Images