A method for identifying characteristic peaks of SERS spectra based on a negative feedback database

Abstract

A method for identifying characteristic peaks of SERS spectra based on a negative feedback database proposed by the present invention includes: screening the peak shapes of quasi-peaks in the SERS spectra and retaining the quasi-peaks with small deviations from the Gaussian distribution; matching and selecting the SERS spectra with the standard SERS spectra in the database to select the standard SERS spectra with a large degree of matching; performing negative feedback on the quasi-peaks with small deviations from the Gaussian distribution in the SERS spectra based on the standard SERS spectra with a large degree of matching, that is, obtaining the characteristic peaks in the SERS spectra. A method for identifying characteristic peaks of SERS spectra based on a negative feedback database proposed by the present invention can automatically identify characteristic peaks without any artificial parameter setting, and can also identify weak characteristic peaks and filter out false peaks.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral analysis, and particularly to a method for identifying SERS spectral characteristic peaks based on a negative feedback database. Background Art

[0002] Raman detection has unique advantages in on-site detection due to its fast speed, high sensitivity, and non-destructive fingerprint characteristics, and has also been widely applied in fields such as agriculture, medicine, food, petrochemicals, etc. Since the SERS spectral signals of the test objects can be amplified by more than a million times through a nano-enhanced substrate, and even single-molecule level detection can be achieved, the SERS spectral characteristic peak information can be used to identify the tested objects. Generally, SERS spectral analysis includes steps such as spectral preprocessing, characteristic peak extraction, and characteristic peak classification. Among them, characteristic peak extraction is the core link of SERS spectral analysis. Since each substance has a corresponding SERS spectral characteristic peak distribution, the position and size of the SERS spectral characteristic peaks directly reflect the structural and content information of the substance. That is to say, whether the characteristic peaks in the SERS spectrum can be well identified will directly determine the accuracy of the characteristic classification of the tested object.

[0003] The commonly used spectral peak recognition methods in the past are as follows: The amplitude method sets a threshold, regards the first point greater than the threshold as the starting point of the spectral peak, regards the subsequent maximum point as the peak point of the spectral peak, and then regards the first point less than the threshold next as the final point of the spectral peak. The principle of this method is relatively simple, and the calculation speed is relatively fast, but it is easily affected by baseline drift, and the selection of the comparison threshold has a great influence on the accuracy of spectral peak detection. The continuous wavelet transform method decomposes the signal by superimposing a series of wavelet functions, and changes the peak seeking from finding the extreme value in the time domain to finding the peak of the ridge line of the wavelet coefficient matrix. Although the peak seeking accuracy of this method is relatively high and it has a strong suppression ability for noise and background, the calculation amount is relatively large and it is really not suitable for real-time operation. At the same time, since the length of the ridge line is closely related to the selection of the wavelet scale, it is also necessary to determine the ridge line length threshold and the ridge line signal-to-noise ratio threshold, so the meaning of the ridge line signal-to-noise ratio is not very clear. Therefore, the continuous wavelet transform method is not robust enough and is not easy to be used in the recognition of SERS spectral characteristic peaks with a relatively large signal-to-noise ratio. The derivative method is based on the basic idea of regarding the spectral line as a continuous curve, taking the derivative of each point on the spectral line, and determining the position of the spectral peak according to the properties of the derivative. This method has a relatively high search accuracy for smooth curves, and the calculation speed can basically meet the real-time requirements, but it is easy to generate false peaks for complex spectral lines with relatively large noise, and it is necessary to set threshold parameters to filter out false peaks. Therefore, the selection of the threshold has a great influence on the analysis result. According to the definition of the local signal-to-noise ratio, the lower limit of the local signal-to-noise ratio at the spectral peak should be 6 times the noise standard deviation. However, the actual spectral data contains not only characteristic peaks and baseline noise, but also baseline drift. How to estimate the noise standard deviation from it is a problem worthy of research.

[0004] Currently, noise estimation is mainly carried out through traditional manual or semi-manual methods. To estimate the noise standard deviation, it is first necessary to find a section of data in the spectrum that does not contain characteristic peaks, outliers, and obvious baseline slopes. If the spectral data collected in different scenarios are processed in this way, due to the influence of human subjective factors, it is easy to lead to unreliable estimation results, which is extremely unfavorable to the automated operation of handheld Raman spectrometers. And because this SERS spectrum has more interference peaks than traditional SERS spectra, it is very difficult to distinguish the correct SERS spectral characteristic peaks from the interference peaks only by simple spectral peak recognition. The interference peaks generally include noise interference peaks, substrate interference peaks, and non-material characteristic peaks, etc. Therefore, the SERS spectral characteristic peaks not only need to carry out traditional spectral peak recognition, but also need to carry out characteristic peak screening on the basis of traditional spectral peak recognition. Summary of the Invention

[0005] Based on the technical problems existing in the background art, the present invention proposes a method for identifying characteristic peaks of SERS spectra based on a negative feedback database. This identification method can automatically identify characteristic peaks without any artificial parameter setting, and can also identify weak characteristic peaks and filter out false peaks.

[0006] A method for identifying characteristic peaks of SERS spectra based on a negative feedback database proposed by the present invention includes:

[0007] Perform peak shape screening on the quasi-peaks in the SERS spectrum, and retain the quasi-peaks with small deviation from the Gaussian distribution;

[0008] Match and select the SERS spectrum with the standard SERS spectrum in the database, and select the standard SERS spectrum with a large matching degree; based on the standard SERS spectrum with a large matching degree, perform negative feedback on the quasi-peaks in the SERS spectrum with small deviation from the Gaussian distribution, that is, obtain the characteristic peaks in the SERS spectrum.

[0009] In the present invention, first perform peak shape screening on the quasi-peaks of weak Raman signals, use the deviation degree between the quasi-peaks and the Gaussian distribution to find all quasi-peaks with weak Raman signals that satisfy the Gaussian distribution, and then perform peak position screening on these quasi-peaks, and use database negative feedback for characteristic peak selection to find the characteristic peaks in the weak Raman signals.

[0010] Preferably, the SERS spectrum is obtained after preprocessing the original SERS spectrum based on a convolutional neural network.

[0011] In the present invention, a convolutional neural network trained according to the Raman spectrum characteristics is introduced to eliminate the influence of baseline drift and common noise. Compared with the traditional preprocessing methods such as median filtering, high-pass filtering, and wavelet transform, the convolutional neural network has the characteristics of fast speed, good real-time performance, and signal amplification, improving the recognition rate of Raman spectra.

[0012] Preferably, the "preprocessing based on a convolutional neural network" specifically includes: selecting "Glorot" initialization for the weights of the convolutional kernel and the output layer, and using a random grid search cross-validation framework to select the configuration with the highest accuracy during the training stage.

[0013] In the present invention, for the weight initialization of the convolutional kernel and the output layer, the reason for choosing "Glorot" initialization is that the initialization can be replicated by tracking the seeds used for randomization.

[0014] Preferably, the quasi-peaks are obtained by extracting the SERS spectrum using the supersymmetric characteristic peak extraction method.

[0015] The "supersymmetric feature peak extraction method" specifically includes: using the supersymmetric function as a transformation function to perform convolution transformation with the data of the SERS spectrum, extracting the quasi-peak in the SERS spectrum, and including obtaining the peak position, peak height, and full width at half maximum of the quasi-peak;

[0016] Preferably, the supersymmetric function is a symmetric transformation function, and its formula is as follows:

[0017]

[0018] where A and B are constants, and x represents a variable.

[0019] In the present invention, the supersymmetric feature peak extraction method is introduced for peak detection and extraction. Compared with the amplitude method, slope method, and area method in the traditional peak detection methods, the supersymmetric feature peak extraction method combines a special design of the peak-shaped function and is combined with the method of selecting a matching filter, improving the detection speed and accuracy.

[0020] Preferably, the "peak shape screening" specifically includes: calculating the deviation values α and β between the quasi-peak in the SERS spectrum and the Gaussian distribution. If the deviation values α and β are not within the set threshold range, then filter out the quasi-peak. If the deviation values α and β are within the set threshold range, then retain the quasi-peak, so as to retain the quasi-peak with a small deviation from the Gaussian distribution.

[0021] Preferably, the deviation value α is the skewness α, and its formula is as follows:

[0022]

[0023]

[0024]

[0025] where D represents the variance, n represents the total number of ordinate values of the peak, represents the average value of the spectral ordinate, and y i is the i-th spectral ordinate value;

[0026] The deviation value β is the kurtosis β, and its formula is as follows:

[0027] μ = EY

[0028] σ 2 = EXY 2 - μ 2

[0029]

[0030] μ represents the central moment, E represents the expectation operator, Y represents the array composed of the spectral ordinates, X represents the array composed of the spectral abscissas in the database, and σ represents the standard deviation.

[0031] In the present invention, the set threshold ranges of α and β are obtained by statistically analyzing a large amount of real case data and combining the characteristic that the Raman characteristic peaks satisfy the Gaussian distribution.

[0032] Preferably, the "matching selection" specifically includes: calculating the matching coefficient γ between the SERS spectrum and the standard SERS spectrum in the database. If the matching coefficient γ is greater than the set threshold, the standard SERS spectrum is retained to select the standard SERS spectrum with a high degree of matching.

[0033] Preferably, the formula of the matching coefficient γ is as follows:

[0034]

[0035] Wherein, A is the matrix composed of the abscissas and ordinates of the SERS spectrum after peak shape screening, and B is the matrix composed of the abscissa and ordinate data of the standard SERS spectrum in the database. And is the mean of A and B, m is the abscissa, and n is the ordinate.

[0036] In the present invention, the set threshold range of the matching coefficient γ is obtained by statistically analyzing a large amount of real case data and combining the characteristic that the Raman characteristic peaks satisfy the Gaussian distribution.

[0037] Preferably, the "negative feedback" specifically includes: calculating the difference θ between the peak positions of the quasi-peak in the SERS spectrum and the peak positions of each peak in the standard SERS spectrum. If the difference θ is within the set threshold range, the quasi-peak is used as a characteristic peak to obtain the characteristic peaks in the SERS spectrum.

[0038] In the present invention, a characteristic peak screening step based on negative feedback of the precursor drug database is introduced during the peak detection process, which can not only further improve the detection speed and accuracy, but also enhance the accuracy and ability to detect weak peaks.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] A method for identifying characteristic peaks of SERS spectra based on a negative feedback database proposed by the present invention can obtain the positions and intensities of weak signal characteristic peaks without the need to pre-set a series of thresholds for SERS spectra, avoiding the distortion caused by the preprocessing of weak Raman spectral data in the prior art and improving the accuracy of weak Raman spectral signal analysis. Description of the Drawings

[0041] Figure 1 It is the flowchart of the steps of the recognition method described in the embodiment;

[0042] Figure 2 It is the main program flowchart of the recognition method described in the embodiment;

[0043] Figure 3 It is the flowchart of the first peak shape screening in the recognition method described in the embodiment;

[0044] Figure 4 It is the flowchart of the second peak position screening in the recognition method described in the embodiment;

[0045] Figure 5 It is the schematic diagram before and after the pretreatment of the SERS spectrum to be recognized described in the embodiment;

[0046] Figure 6 It is the schematic diagram of the quasi-peak extracted from the pretreated SERS spectrum described in the embodiment;

[0047] Figure 7 It is the schematic diagram of the quasi-peak selected by the first peak shape screening described in the embodiment;

[0048] Figure 8 For Embodiment according to CV 10 -7 It is the schematic diagram of the characteristic peak selected by the second peak position screening based on the spectrum;

[0049] Figure 9 For Embodiment according to CV 10 -8 It is the schematic diagram of the characteristic peak selected by the second peak position screening based on the spectrum. Specific implementation manner

[0050] Embodiment

[0051] Referring to Figure 1-4 , the present invention proposes a method for identifying characteristic peaks of SERS spectra based on a negative feedback database, including the following steps:

[0052] S1. Take the collected crystal violet (CV) SERS spectrum as the SERS spectrum to be recognized, perform spectral pretreatment on the SERS spectrum to be recognized based on a convolutional neural network, select "Glorot" initialization for the weights of the convolutional kernel and the output layer, and use a random grid search cross-validation framework to select the configuration with the highest accuracy during the training phase to obtain the pretreated SERS spectrum;

[0053] Among them, the parameters of the convolutional neural network model and their value ranges are as follows:

[0054] The number of convolutional kernels in the convolutional layer: #kernels ∈ {1, 4, 6};

[0055] The size of the convolutional kernel: N ∈ [2, 50];

[0056] Convolution step size: S ∈ [1, 50];

[0057] Parameters in the regularization term: λ1, λ2 = 10n, where n ∈ [-1, 1];

[0058] Momentum in the SGD update rule: momentum ∈ 0.1 * [3, 12];

[0059] Learning rate: lr = 10n, where n ∈ [0, 4];

[0060] The spectral images of the SERS spectra to be recognized before and after spectral preprocessing by the convolutional neural network can be referred to Figure 5 as shown in Figure 5 which is a schematic diagram of the SERS spectra to be recognized before and after preprocessing. As can be seen from Figure 1 this, after the SERS spectra to be recognized are subjected to spectral preprocessing based on the convolutional neural network, the interference of the baseline background and ordinary noise is reduced, providing stable conditions for the subsequent extraction of quasi-peak positions;

[0061] S2. Use the supersymmetric feature peak extraction method to extract the quasi-peak positions from the preprocessed SERS spectra in step S1. By selecting the supersymmetric function as the transformation function and performing convolution transformation with the data of the preprocessed SERS spectra, the quasi-peak positions in the preprocessed SERS spectra are extracted, and spectral data such as the peak positions, peak heights, full-width at half-maximum, and starting positions of the quasi-peak positions are obtained;

[0062] The supersymmetric function is a symmetric transformation function, and its formula is:

[0063]

[0064] where A and B are constants, and x represents the variable;

[0065] The quasi-peak positions extracted from the preprocessed SERS spectra using the supersymmetric feature peak extraction method can be referred to Figure 6 as shown in Figure 6 which is a schematic diagram of the quasi-peak positions extracted from the preprocessed SERS spectra;

[0066] The detailed data of the peak positions, peak heights, full-width at half-maximum, and starting positions of the quasi-peak positions extracted from the preprocessed SERS spectra can be referred to Table 1 below:

[0067] Table 1 List of spectral data of the quasi-peak positions in the preprocessed SERS spectra

[0068] Serial number Peak position Peak height Full width at half maximum Start position End position 1 522 2991.08 20.14 488 556 2 542 886.48 1.64 539 545 3 555 2020.67 10.40 537 573 … … … … … … 54 1803 821.62 5.58 1794 1812 55 1950 1037.29 13.54 1927 1973

[0069] S3. Calculate the deviation values of the quasi-peak extracted in step S2 from the Gaussian distribution, including skewness α and kurtosis β, and screen out the quasi-peaks with skewness α and kurtosis β within the set threshold range: If the skewness α and kurtosis β are not within the set threshold range, directly filter out the quasi-peak and do not count it as the SERS spectral characteristic peak. If the skewness α and kurtosis β are within the set threshold range, retain the quasi-peak. Thus, all the quasi-peaks that satisfy the Gaussian distribution are screened out, and the quasi-peaks after the first peak shape screening are obtained.

[0070] The calculation formulas for the above skewness α and kurtosis β are as follows:

[0071]

[0072]

[0073]

[0074] where D represents variance, n represents the number of ordinate values of the entire peak, represents the average value of the spectral ordinate, and y i is the i-th spectral ordinate value;

[0075] μ = EY

[0076] σ 2 = EXY 2 - μ 2

[0077]

[0078] where μ represents the central moment, E represents the expectation operator, Y represents the array composed of spectral ordinates, X represents the array composed of spectral abscissas, and σ represents the standard deviation;

[0079] The skewness α and kurtosis β of the quasi-peaks calculated according to the above formulas are shown in Table 2 below:

[0080] Table 2 List of calculation results of skewness α and kurtosis β of quasi-peaks

[0081] Serial number Peak position Peak height Full width at half maximum Skewness α Kurtosis β 1 522 2991.08 20.14 1.52 1.57 2 542 886.48 1.64 6.42 6.48 3 555 2020.67 10.40 -0.3 2.77 … … … … … … 54 1803 821.62 5.58 2.76 4.81 55 1950 1037.29 13.54 -1.32 -2.9

[0082] The quasi-peaks screened out by the first peak shape can be referred to Figure 7 as shown, Figure 7 which is a schematic diagram of the quasi-peaks screened out by the first peak shape;

[0083] In this embodiment, the set threshold of skewness α is defined as between -1 and 3, and the set threshold range of kurtosis β is defined as between -3 and 3. Quasi-peaks with skewness α and kurtosis β satisfying α ∈ [-1, 3] and β ∈ [-3, 3] are screened out. The specific screening results are shown in Table 3 below:

[0084] List of spectral data of quasi-peak positions selected by the first peak shape screening in Table 3

[0085] Serial number Peak position Peak height Full width at half maximum Skewness α Kurtosis β 1 522 2991.08 20.14 1.52 1.57 2 555 2020.67 10.40 -0.3 2.77 3 603 603.21 8.78 -0.55 0.75 … … … … … … 21 1584 1430.13 11.71 -0.96 2.73 22 1616 5121.67 12.69 0.93 1.33

[0086] S4. Calculate the matching coefficient γ between the preprocessed SERS spectrum described in step S1 and the standard SERS spectrum in the specially established database, and screen out the standard SERS spectra with the matching coefficient γ greater than the set threshold: If the matching coefficient γ is less than the set threshold, then do not perform negative feedback on the peak positions of this standard SERS spectrum and the quasi-peak positions after the first peak shape screening; if the matching coefficient γ is greater than the set threshold, then perform negative feedback on the peak positions of this standard SERS spectrum and the quasi-peak positions after the first peak shape screening;

[0087] The calculation formula for the above-mentioned matching coefficient γ is as follows:

[0088]

[0089] where A is the matrix composed of the horizontal and vertical coordinates of the SERS spectrum after peak shape screening, and B is the matrix composed of the horizontal and vertical coordinate data of the standard SERS spectrum in the database, and is the mean value of A and B, m is the abscissa, and n is the ordinate;

[0090] The matching coefficient γ calculated according to the above formula is shown in Table 4 below:

[0091] List of calculation results of the matching coefficient γ

[0092]

[0093] In this embodiment, the range of the set threshold of the matching coefficient γ is defined as 90 - 100%, that is, when the matching coefficient γ ≥ 90%, perform negative feedback on the peak positions of the standard SERS spectrum in the database and the quasi-peak positions after the first peak shape screening; according to the above table, the spectra that need to perform peak position negative feedback are CV 10 -7 and CV 10 -8 , that is, it is necessary to perform peak position negative feedback twice;

[0094] According to the above table, the spectra that need to perform peak position negative feedback are CV 10 -7 , compare this CV 10 -7 spectrum with the quasi-peak positions after the first peak shape screening, and calculate CV 10 -7If the difference θ between the peak position of each peak in the spectrum and the peak position of the quasi-peak satisfies θ ∈ [-5, 5], then this quasi-peak is counted as a characteristic peak; if it does not satisfy θ ∈ [-5, 5], it is filtered out. Thus, the second peak position screening is completed, and the characteristic peaks after the second peak position screening are obtained. The specific screening results can be referred to Figure 8 as shown in Figure 8 Figure -7 6, which is a schematic diagram of the characteristic peaks screened by the second peak position according to the CV 10

[0095] The spectral data of the characteristic peaks screened by the second peak position according to the above CV 10 -7 spectrum are shown in Table 5 below:

[0096] Table 5 List of spectral data of the characteristic peaks screened by the second peak position

[0097] Serial number Peak position Peak height Full width at half maximum 1 522 2991.08 20.14 2 555 2020.67 10.40 3 723 7499.89 8.66 4 757 10695.66 25.07 5 914 2313.46 11.79 6 939 2836.14 40.77 7 961 2671.84 9.56 8 1040 1559.63 23.62 9 1062 1037.29 6.63 10 1137 1055.57 14.37 11 1170 4283.62 28.18 12 1296 1558.69 15.78 13 1367 1721.70 3.04 14 1390 2529.16 41.33 15 1445 5702.72 23.16 16 1538 1210.71 15.90 17 1584 1430.13 11.71 18 1616 5121.67 12.69

[0098] As can be seen from the above table, the spectra that need to perform peak position negative feedback are still CV 10 -8 , and this CV 10 -8 spectrum is compared with the characteristic peaks after the second peak position screening, and the difference θ between the peak position of each peak in the CV 10 -8 spectrum and the peak position of the characteristic peak is calculated. If θ ∈ [-5, 5], then this characteristic peak is counted as the final characteristic peak; if it does not satisfy θ ∈ [-5, 5], it is filtered out to obtain the characteristic peaks in the SERS spectrum. The specific screening results can be referred to Figure 9 as shown in Figure 9 Figure -8 7, which is a schematic diagram of the characteristic peaks screened by the second peak position according to the CV 10

[0099] The spectral data of the characteristic peaks screened by the second peak position according to the above CV 10 -8 spectrum are shown in Table 6 below:

[0100] Table 6 List of spectral data of the characteristic peaks screened by the second peak position

[0101] Serial number Peak position Peak height Full width at half maximum 1 723 7499.89 8.66 2 757 10695.66 25.07 3 914 2313.46 11.79 4 939 2836.14 40.77 5 1170 4283.62 28.18 6 1445 5702.72 23.16 7 1584 1430.13 11.71 8 1616 5121.67 12.69

[0102] The comparison between the finally identified characteristic peaks in the SERS spectrum and the actual characteristic peaks is shown in Table 7 below:

[0103] Table 7 List of comparison results between the finally identified characteristic peaks and the actual characteristic peaks in the SERS spectrum

[0104]

[0105]

[0106] As can be seen from the above, through the recognition method of this embodiment, a complete SERS spectral characteristic peak can finally be obtained, and the peak position shift is within the normal range.

[0107] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope of the present invention, according to the technical solution of the present invention and its invention, making equivalent replacements or changes should be covered within the protection scope of the present invention.

Claims

1. A method for identifying characteristic peaks of SERS spectra based on a negative feedback database, characterized in that, Including: Performing peak shape screening on the quasi-peaks in the SERS spectrum, and retaining the quasi-peaks with small deviations from the Gaussian distribution; Matching and selecting the SERS spectrum with the standard SERS spectra in the database, and selecting the standard SERS spectrum with a large matching degree; performing negative feedback on the quasi-peaks with small deviations from the Gaussian distribution in the SERS spectrum based on the standard SERS spectrum with a large matching degree, that is, obtaining the characteristic peaks in the SERS spectrum; The "negative feedback" specifically includes: calculating the difference θ between the peak positions of the quasi-peaks in the SERS spectrum and the peak positions of each peak in the standard SERS spectrum. If the difference θ is within the set threshold range, then taking this quasi-peak as a characteristic peak, and obtaining the characteristic peaks in the SERS spectrum.

2. The SERS spectral characteristic peak recognition method based on a negative feedback database according to claim 1, wherein The SERS spectrum is obtained after preprocessing the original SERS spectrum based on a convolutional neural network.

3. The SERS spectral characteristic peak recognition method based on a negative feedback database according to claim 2, wherein The "preprocessing based on a convolutional neural network" specifically includes: selecting "Glorot" initialization for the weights of the convolutional kernel and the output layer, and using a random grid search cross-validation framework to select the configuration with the highest accuracy during the training phase.

4. The SERS spectral feature peak recognition method based on a negative feedback database according to any one of claims 1 to 3, characterized in that, The quasi-peaks are obtained by convolving the SERS spectrum data with a supersymmetric function as a transformation function, and extracting the quasi-peaks in the SERS spectrum, and include obtaining the peak positions, peak heights, and full-width at half-maximum of the quasi-peaks.

5. The SERS spectral characteristic peak recognition method based on a negative feedback database according to claim 4, characterized in that, The supersymmetric function is a symmetric transformation function, and its formula is as follows: where a and b are constants, and x represents a variable.

6. The SERS spectral characteristic peak recognition method based on a negative feedback database according to any one of claims 1-3, characterized in that The "peak shape screening" specifically includes: calculating the deviation values α and β between the quasi-peaks in the SERS spectrum and the Gaussian distribution. If the deviation values α and β are not within the set threshold range, then filtering out this quasi-peak. If the deviation values α and β are within the set threshold range, then retaining this quasi-peak, so as to retain the quasi-peaks with small deviations from the Gaussian distribution.

7. The SERS spectral characteristic peak recognition method based on a negative feedback database according to claim 6, wherein, The deviation value α is the skewness α, and its formula is as follows: Among them, D represents variance, n represents the number of the entire wave peak ordinate values, represents the average value of the spectral ordinate, y i is the i-th spectral ordinate value; The deviation value β is the kurtosis β, and its formula is as follows: μ = EY σ 2 = EXY 2 - μ 2 where represents the central moment, E represents the expectation operator, Y represents the array composed of the ordinate of the spectrum, X represents the array composed of the abscissa of the spectrum, and σ represents the standard deviation.

8. The SERS spectral characteristic peak recognition method based on a negative feedback database according to any one of claims 1-3, characterized in that The "matching selection" specifically includes: calculating the matching coefficient γ between the SERS spectrum and the standard SERS spectra in the database. If the matching coefficient γ is greater than the set threshold, then retaining this standard SERS spectrum, so as to select the standard SERS spectrum with a large matching degree.

9. The SERS spectral feature peak recognition method based on a negative feedback database according to claim 8, wherein The formula of the matching coefficient γ is as follows: Among them, A is a matrix composed of the horizontal and vertical coordinates of the SERS spectrum after peak shape screening, and B is a matrix composed of the horizontal and vertical coordinate data of the standard SERS spectrum in the database. and is the mean of A and B, m is the abscissa, and n is the ordinate.

Images