A pure substance identification method based on Raman physicochemical characteristics

Abstract

The present invention discloses a method for identifying pure substances based on Raman physical and chemical characteristics, including the steps of establishing a standard library and identifying a sample, wherein the establishment of the standard library includes preprocessing the Raman spectrum data of the pure substance; based on a peak-finding algorithm, dynamically adjusting the peak height, peak spacing and peak prominence through iteration, pre-finding features and marking the starting and ending Raman shifts; adjusting the pre-finding features according to the actual physical and chemical information of the pure substance, and adding corresponding vibration / rotation modes to the features; building a library for each feature, and combining the pure substance with the feature into the library; matching the sample spectrum data with the substance and feature combination table in the standard library through the matching feature set during identification, calculating the feature similarity, and adding the result to the matching substance-regression coefficient set. The present invention can improve the interpretability of the identification results; reduce the interference of noise signals on the identification results; and improve the data applicability and identification accuracy of different spectral pixel lengths.

Description

Technical Field

[0001] The present invention relates to a pure substance identification method, in particular to a pure substance identification method based on Raman physicochemical characteristics. Background Art

[0002] The existing sample identification methods mainly include full spectrum similarity calculation and feature extraction matching.

[0003] Full spectrum similarity calculation: The complete Raman spectrum of the sample to be identified is used to calculate the similarity with the Raman spectrum in the known sample library, and the sample in the library with the highest similarity is selected as the identification result of the sample to be identified. For example, patent document CN102507532A selects a sample to measure its Raman spectrum, establishes a database, and then detects the Raman spectrum of the unknown sample. The spectrum comparison method is used to compare it with the standard spectrum to calculate its related hit rate index HQI (Hit Qualification Hit). The closer the HQI is to 0, the better the matching relationship between the two spectra, that is, the sample is more likely to be consistent with the corresponding component in the library. By setting a stricter threshold, accurate qualitative judgment of chemicals can be made. The related hit rate index method is an important method of digital judgment, which compares the spectrum with the source machine spectrum at all wavelength points.

[0004] The main problems of full spectrum similarity calculation are as follows:

[0005] ① The actually measured Raman spectra often contain background noise and other interference factors, while the technology of full-spectrum similarity calculation contains a large amount of non-effective information area data, which may affect the results of the similarity calculation and lead to a decrease in recognition accuracy.

[0006] ② For large databases, calculating all possible matches and determining the best match may require a lot of computing resources and time. The technology of full-spectrum similarity calculation contains a large amount of data in non-valid information areas, which will waste a lot of computing resources and time.

[0007] Feature extraction and matching are mainly divided into unsupervised feature extraction based on peak detection and supervised dimensionality reduction feature extraction such as machine learning and deep learning. For example, patent document CN101629909A determines whether there is a spectral signal related to chemical or biochemical substances near a predetermined wavelength in the Raman spectrum.

[0008] For another example, patent document CN101285773A marks the wavenumber corresponding to the extracted characteristic peak, and records the position and intensity of the extracted characteristic peak. The characteristic table of the sample to be tested is matched with the characteristic peak position table set of known samples, and the characteristic table of the sample to be tested is matched with each one in the set. For each wavenumber point where the characteristic peak is located, m is the number of characteristic peaks of the known sample i. Check whether the characteristic table of the sample to be tested also has peaks at these m wavenumber points. If so, it is considered that the position match between the sample to be tested and the i-th known sample is successful, otherwise, the match fails.

[0009] Patent document CN114993891A develops an applicable cosine similarity matching method based on the cosine similarity calculation method of characteristic peaks, for matching spectrum characteristic peak vectors P1 and database spectrum characteristic peak vectors P2 with different characteristic peak positions, and is used to calculate the similarity between the matching spectrum characteristic peak vectors P1 and database spectrum characteristic peak vectors P2.

[0010] Patent document CN108241846A has a first identification step: comparing the peak information of the measured spectrum with the peak information of the pre-stored standard spectrum to identify whether the peak information of the measured spectrum matches the peak information of the standard spectrum; a second identification step: when it is identified in the first identification step that the peak information of the measured spectrum matches the peak information of the standard spectrum, comparing the data of the measured spectrum with the data of the pre-stored standard spectrum to further identify whether the measured spectrum matches the standard spectrum.

[0011] As can be seen from the above, the existing feature extraction and matching mainly have the following problems:

[0012] ① Feature extraction-based methods have poor interpretability. It may not be possible to clearly explain why a specific sample is identified as a certain substance.

[0013] ②Machine learning and deep learning models usually require a large amount of labeled data for training to ensure good generalization ability. However, obtaining enough high-quality training data can be difficult. Summary of the invention

[0014] The technical problem to be solved by the present invention is to provide a pure substance identification method based on Raman physical and chemical characteristics, which can improve the interpretability of identification results; reduce the interference of noise signals on identification results; and improve the data applicability and identification accuracy of different spectral pixel lengths.

[0015] The technical solution adopted by the present invention to solve the above technical problems is to provide a pure substance identification method based on Raman physical and chemical characteristics, including the steps of establishing a standard library and identifying a sample, wherein the steps of establishing the standard library are as follows: S11) obtaining Raman spectrum data of pure substances for library building; S12) preprocessing the Raman spectrum data of pure substances, including noise reduction, baseline removal and interpolation processing; S13) based on a peak-finding algorithm, dynamically adjusting the peak height, peak spacing and peak prominence through iteration, pre-finding features and marking the starting and ending Raman shifts; S14) adjusting the pre-finding features according to the actual physical and chemical information of the pure substance, and adding corresponding vibration / rotation modes to the features; S15) building a library for each feature, and combining the pure substance with the feature into the library;

[0016] The sample identification steps are as follows: S21) obtaining Raman spectrum data R of the sample to be identified R ; S22) pre-processing the Raman spectrum data of the sample to be identified, including noise reduction, baseline removal and interpolation processing, to obtain the pre-processed Raman spectrum data R P ; S23) obtain the characteristic data set F in the standard library, match each standard library characteristic data with the current sample to be identified in turn, and when matching with a certain standard library feature, intercept R P Corresponding Raman range data, calculate feature similarity; add standard library features with feature similarity higher than feature threshold to matching feature set M F , if M F If it is empty, go to step S27, otherwise go to step S24; S24) by matching feature set M F Matching with the material and feature combination table in the standard library, specifically including: finding the feature set S of each material in the standard library F And matching feature set M F The intersection size C M , and then C M and the characteristic set S of the corresponding substance F Size C S The ratio of is added to the matching material-matching feature ratio set M SF ; S25) from M SF Items whose matching feature ratio is less than the feature ratio threshold are screened out. If M SF If the sign is empty, go to step S27, otherwise go to step S26; S26) traverse M SF For each matching substance, all the features of the substance in the standard library are used to construct the regression spectrum R F and use all its features ranging from R P The spectral data of the corresponding Raman shift is intercepted and the matching spectrum is constructed, the feature similarity is calculated, and the result is added to the matching material-regression coefficient set M SRS27) sorting the matching material-regression coefficient set in descending order of the regression coefficient, selecting the first item as the final recognition result, and calibrating the Raman spectrum of the sample to be tested according to the vibration / rotation mode corresponding to all feature start and end displacements in the standard feature library.

[0017] Furthermore, the step S12 uses the Whittaker smoothing algorithm to perform noise reduction processing, sets the number of smoothing windows to 5, and the fitting order to 2, to obtain baseline-removed pure substance Raman spectrum data.

[0018] Furthermore, the step S12 uses the airPLS algorithm to perform baseline removal processing, sets the fitting error to 1000, the fitting order to 2, and the maximum number of iterations to 15, to obtain baseline-removed pure substance Raman spectrum data.

[0019] Furthermore, the interpolation process of step S12 adopts cubic spline interpolation method, the interpolation interval is 1, and the interpolation range is 158-3412.

[0020] Furthermore, in steps S23 and S26, the feature similarity is calculated using the following formula:

[0021]

[0022] Further, the characteristics include out-of-plane rocking vibration, bending vibration, stretching vibration, in-plane rocking vibration, asymmetric deformation vibration, symmetric deformation vibration, asymmetric stretching vibration and symmetric stretching vibration.

[0023] Furthermore, the pure substance is ethanol, and the corresponding characteristics are OH out-of-plane rocking vibration, CC bending vibration, CO stretching vibration, OH in-plane rocking vibration, CH3 and CH2 asymmetric deformation vibration, CH3 and CH2 symmetric deformation vibration, CH3 and CH2 asymmetric stretching vibration, and CH3 and CH2 symmetric stretching vibration.

[0024] Compared with the prior art, the present invention has the following beneficial effects: the pure substance identification method based on Raman physical and chemical characteristics provided by the present invention has the following advantages:

[0025] 1. The present invention identifies unknown samples based on their physical and chemical characteristics, and the final results are regressed to the physical and chemical characteristics to support and explain the identification results;

[0026] 2. Compared with traditional data database construction, the present invention only builds databases for feature areas, excludes interference from non-effective information areas, and reduces the interference of noise signals on recognition results;

[0027] 3. Due to the construction of a database for physical and chemical characteristics, the scope of a single standard data is greatly reduced, so that more flexible feature combination matching can be achieved, which is suitable for Raman spectral data with different spectral pixel lengths;

[0028] 4. By matching features with lower granularity, the influence of signals in non-effective information areas can be eliminated, thereby improving recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart for establishing the standard library of the present invention;

[0030] Figure 2 This is a sample identification flow chart of the present invention;

[0031] Figure 3 This is the Raman spectrum of pure ethanol after interpolation of the present invention;

[0032] Figure 4 The Raman spectrum of the sample to be tested is identified and calibrated by the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0034] The pure substance identification method based on Raman physicochemical characteristics provided by the present invention is divided into two parts: the establishment of a standard library and the identification of samples. The specific process of establishing the standard library is as follows: Figure 1 The specific process of sample identification is as follows: Figure 2 .

[0035] Steps to build the standard library:

[0036] S11) obtaining Raman spectrum data of pure substance for library construction;

[0037] S12) Preprocessing the pure substance Raman spectrum data includes noise reduction, baseline removal and interpolation. The noise reduction process uses the Whittaker smoothing algorithm. In particular, the present invention uses processing parameters with a window of 5 and a fitting order of 2 to obtain the baseline-removed pure substance Raman spectrum data; the baseline removal process uses the airPLS (baseline fitting of adaptive iterative reweighted penalized least squares method) algorithm. In particular, the present invention uses processing parameters with a fitting error of 1000, a fitting order of 2, and a maximum number of iterations of 15 to obtain the baseline-removed pure substance Raman spectrum data; the interpolation process uses the cubic spline interpolation method. In particular, the present invention interpolates it to an interval of 1 and a range of 158-3412 to obtain the interpolated pure ethanol Raman spectrum data;

[0038] S13) based on the peak-finding algorithm, by iteratively and dynamically adjusting the peak height, peak spacing, and peak prominence, pre-finding features and marking the start and end Raman shifts;

[0039] S14) adjusting the pre-searched features according to the actual physical and chemical information of the pure substance, and adding corresponding vibration / rotation modes to the features;

[0040] S15) Building a library for each feature, and combining the pure substance with the feature into the library.

[0041] Sample identification steps:

[0042] S21) Obtaining Raman spectrum data R of the sample to be identified R ;

[0043] S22) Preprocessing the Raman spectrum data of the sample to be identified is consistent with the library preprocessing process, including noise reduction, baseline removal and interpolation, to obtain the preprocessed Raman spectrum data R P ;

[0044] S23) Obtain a feature data set F in the standard library, and sequentially match each feature data of the standard library with the current sample to be tested. Specifically, when matching with a feature of a certain standard library, intercept R P The corresponding Raman range data is used to calculate the feature similarity using Formula 1. For feature similarity higher than the feature threshold T F = 0.75 standard library features are added to the matching feature set M F , if M F If it is empty, go to step S27, otherwise go to step S24;

[0045]

[0046] In Formula 1, X is the spectrum to be currently identified, and Y is the spectrum in the standard library that is currently being matched.

[0047] S24) by matching feature set M F Match the material and feature combination table in the standard library. Specifically, find the feature set S of each material in the standard library. F And matching feature set M F The intersection size C M , and then C M and the characteristic set S of the corresponding substance F Size C S The ratio of is added to the matching material-matching feature ratio set M SF ;

[0048] S25) from M SF Screen out the matching features whose ratio is less than the feature ratio threshold T S =0.6, if M SF If the sign is empty, go to step S27, otherwise go to step S26;

[0049] S26) Traverse M SF For each sub-item (i.e., matching substance), all the features of the substance in the standard library are used to construct the regression spectrum R Fand use its range from R P The spectral data of the corresponding Raman shift is intercepted and the matching spectrum is constructed. The correlation is calculated using formula 1, and the result is added to the matching material-regression coefficient set M SR middle;

[0050] S27) sorting the matching substance-regression coefficient set in descending order of the regression coefficient, selecting the first item as the final recognition result, displaying the recognition result, and calibrating the Raman spectrum of the sample to be tested according to all feature start and end displacements in the standard feature library and the vibration / rotation mode corresponding to the feature.

[0051] The present invention is further described in detail below by combining the library building of pure ethanol, acetonitrile and dichloromethane and the identification of ethanol with the accompanying drawings:

[0052] The specific steps to establish the standard library are:

[0053] (1) Obtaining Raman spectrum data E of pure ethanol R ;

[0054] (2) Raman spectrum data of pure ethanol E R Preprocessing is performed, wherein the preprocessing includes three processing steps: noise reduction, baseline removal and interpolation. The noise reduction process uses the Whittaker smoothing algorithm. In particular, the present invention uses processing parameters with a window of 5 and a fitting order of 2 to obtain the baseline-removed Raman spectrum data E of pure ethanol. N The airPLS algorithm is used for baseline removal. In particular, the present invention uses the processing parameters of a fitting error of 1000, a fitting order of 2, and a maximum number of iterations of 15 to obtain the baseline-free Raman spectrum data E of pure ethanol. B The interpolation process uses cubic spline interpolation, and the original Raman shift range of the acquisition instrument is 157.1-3412.5cm -1 In particular, the present invention interpolates it to an interval of 1 cm -1 , ranging from 158-3412cm -1 , and the interpolated Raman spectrum data of pure ethanol E is obtained P ,like Figure 3 ;

[0055] (3) Based on the peak search algorithm, the peak height, peak spacing, and peak prominence are dynamically adjusted iteratively to pre-search the features and mark the start and end Raman shifts. In particular, the pre-search features of ethanol are shown in Table 1:

[0056] Starting displacement Cut-off displacement 344 509 817 935 984 1067 1065 1175 1211 1325 1388 1525 2845 2912 2910 2962 2960 3012

[0057] Table 1

[0058] (4) According to the actual physical and chemical information of ethanol, the pre-searched features are adjusted, and the corresponding vibration / rotation modes are added to the features. In particular, the ethanol features after adjustment are as shown in Table 2:

[0059] Starting displacement Cut-off displacement Feature information 344 509 OH out-of-plane rocking vibration 817 1175 CC bending vibration 1211 1325 CO stretching vibration, OH in-plane rocking vibration 1388 1525 Asymmetric deformation vibration and symmetric deformation vibration of CH3 and CH2 2845 3012 Asymmetric stretching vibration and symmetric stretching vibration of CH3 and CH2

[0060] Table 2

[0061] (5) Building a database for each feature, and storing the pure substance and the feature combination in the database. In particular, the database data for the ethanol Raman feature includes the feature information (F1 to F5) corresponding to Table 2, the feature ID automatically generated when storing, and the ethanol substance ID association data;

[0062] (6) To add interference data, repeat steps (1) to (5) for pure acetonitrile and dichloromethane.

[0063] The specific steps of sample identification are:

[0064] (1) Obtain Raman spectrum data R of the sample to be identified R ,In particular, the sample to be identified collected in this specific implementation is ethanol;

[0065] (2) Preprocess the Raman spectrum data of the sample to be identified, which is consistent with the preprocessing process of library construction, including noise reduction, baseline removal and interpolation, and finally obtain R P ;

[0066] (3) Obtain the characteristic data F in the standard library, and match each characteristic data of the standard library with the current sample in turn. Specifically, when performing the OH out-of-plane rocking vibration characteristics, from R P Intercept 344cm -1 -509cm -1 The Raman data in the range is used to calculate the similarity using Equation 1. The feature similarity is higher than the feature threshold T F = 0.75 standard library features are added to the matching feature set M F middle;

[0067] (4) By matching feature set M F Match the material and feature combination table in the standard library. Specifically, find the feature set S of each material in the standard library. F And matching feature set M F The intersection size C M , and then C M and the characteristic set S of the corresponding substance F Size C S The ratio of is added to the matching material-matching feature ratio set M SF ;

[0068] (5) From M SFScreen out the matching features whose ratio is less than the feature ratio threshold T S =0.6;

[0069] (6) Traverse M SF For each sub-item, all the features of the substance in the standard library are used to construct the regression spectrum R F and use the feature range of all its features from R P The spectral data of the corresponding Raman shift is intercepted and the matching spectrum is constructed. The correlation is calculated using formula 1, and the result is added to the matching material-regression coefficient set M SR middle;

[0070] (7) Sort the matching substance-regression coefficient set in descending order of regression coefficient, select the first item as the final recognition result, display the recognition result, and calibrate the Raman spectrum of the sample to be tested according to the vibration / rotation mode corresponding to all the feature start and end displacements in the standard feature library, such as Figure 4 shown.

[0071] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make some modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the definition of the claims.

Claims

1. A method for identifying pure substances based on Raman physicochemical characteristics, characterized in that: The method includes the steps of establishing a standard library and identifying a sample. The steps of establishing the standard library are as follows: S11) obtaining Raman spectrum data of pure substances for library construction; S12) preprocessing the pure substance Raman spectrum data, including noise reduction, baseline removal and interpolation processing; S13) Based on the peak-finding algorithm, the peak height, peak spacing and peak prominence are dynamically adjusted iteratively to pre-find features and mark the start and end Raman shifts; S14) adjusting the pre-searched features according to the actual physical and chemical information of the pure substance, and adding corresponding vibration / rotation modes to the features; S15) building a library for each feature, and combining the pure substance with the feature into the library; The sample identification steps are as follows: S21) Obtain Raman spectrum data of the sample to be identified ; S22) Preprocess the Raman spectrum data of the sample to be identified, including noise reduction, baseline removal and interpolation processing, to obtain the preprocessed Raman spectrum data ; S23) Get feature datasets in the standard library , match each standard library feature data with the current sample to be identified in turn, and when matching with a certain standard library feature, intercept Corresponding Raman range data, calculate feature similarity; For standard library features whose feature similarity is higher than the feature threshold, add them to the matching feature set ,like If it is empty, go to step S27, otherwise go to step S24; S24) by matching feature set Matching with the material and feature combination table in the standard library, including: finding the feature set of each material in the standard library Matching feature set The intersection size , and then The characteristic set of the corresponding substance Size ratio, added to the matching substance-matching feature ratio set ; S25) from Filter out items whose matching feature ratio is less than the feature ratio threshold. If at this time If it is empty, go to step S27, otherwise go to step S26; S26) Traversal For each matching substance, all the features of the substance in the standard library are used to construct a regression spectrum , and use all its features ranging from The spectral data of the corresponding Raman shift is intercepted and the matching spectrum is constructed, the feature similarity is calculated, and the result is added to the matching material-regression coefficient set middle; S27) sorting the matching substance-regression coefficient set in descending order of the regression coefficient, selecting the first item as the final identification result, and calibrating the Raman spectrum of the sample to be tested according to all feature start and end displacements in the standard feature library and the vibration / rotation mode corresponding to the feature.

2. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 1, characterized in that: The step S12 uses the Whittaker smoothing algorithm to perform noise reduction processing, sets the number of smoothing windows to 5, and the fitting order to 2, to obtain the baseline-removed pure substance Raman spectrum data.

3. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 1, characterized in that: The step S12 uses the airPLS algorithm to perform baseline removal processing, sets the fitting error to 1000, the fitting order to 2, and the maximum number of iterations to 15, to obtain baseline-removed pure substance Raman spectrum data.

4. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 1, characterized in that: The interpolation process in step S12 adopts cubic spline interpolation method, the interpolation interval is 1, and the interpolation range is 158-3412.

5. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 1, characterized in that: In steps S23 and S26, the feature similarity is calculated using the following formula: 。 6. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 1, characterized in that: The features include out-of-plane rocking vibrations, bending vibrations, stretching vibrations, in-plane rocking vibrations, asymmetric deformation vibrations, symmetric deformation vibrations, asymmetric stretching vibrations, and symmetric stretching vibrations.

7. The method for identifying pure substances based on Raman physicochemical characteristics according to claim 6, characterized in that: The pure substance is ethanol, and the corresponding characteristics are OH out-of-plane rocking vibration, CC bending vibration, CO stretching vibration, OH in-plane rocking vibration, CH3 and CH2 asymmetric deformation vibration, CH3 and CH2 symmetric deformation vibration, CH3 and CH2 asymmetric stretching vibration and CH3 and CH2 symmetric stretching vibration.

Images