A method for detecting type iii collagen by spectroscopy

Abstract

This invention belongs to the field of product quality testing technology and provides a spectroscopic method for detecting type III collagen. The method includes a quantitative mode and a qualitative mode. The quantitative mode uses ultraviolet-visible spectroscopy or near-infrared spectroscopy to acquire sample spectra, and uses the Butterfly Optimization Algorithm (BOA) or Whale Optimization Algorithm (WOA) to screen characteristic wavelengths and establish a partial least squares regression (PLS) model to achieve high-precision prediction of the content of recombinant type III collagen in cosmetics. The qualitative mode uses a width learning system (BLS) to classify the preprocessed spectra. The accuracy of the ultraviolet-visible spectroscopy scheme can reach 100%, and the accuracy of the near-infrared spectroscopy scheme can reach over 95.8%. This invention has the advantages of being rapid, non-destructive, interference-resistant, and highly accurate, and can simultaneously meet the needs of quantitative analysis and authenticity identification of recombinant type III collagen in cosmetics. It is suitable for production line quality control and market supervision sampling inspections.

Description

Technical Field

[0001] This invention belongs to the field of product quality testing technology, specifically relating to a method for detecting type III collagen using spectrometry. Background Technology

[0002] Recombinant type III collagen is a high-purity collagen synthesized using genetic engineering technology. Due to its excellent biocompatibility, moisturizing, and repairing functions, it is widely added to various cosmetics, especially water-based products such as toners and serums. However, the quality of recombinant type III collagen cosmetics on the market varies greatly, with issues such as discrepancies between actual and labeled content, the use of inferior products, and even the absence of collagen altogether. These problems seriously harm consumer rights and hinder the healthy development of the industry.

[0003] Currently, the mainstream detection methods for recombinant type III collagen include enzyme-linked immunosorbent assay (ELISA), high-performance liquid chromatography (HPLC), liquid chromatography-tandem mass spectrometry (LC-MS / MS), and sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Although these methods have high accuracy, they generally suffer from problems such as complex sample pretreatment, long detection cycles, expensive instruments, the need for professional operators, the destructive nature of samples, and difficulty in rapid on-site screening. These methods cannot meet the needs of cosmetic manufacturers for large-scale quality control and market regulators for immediate sampling inspections.

[0004] Molecular absorption spectroscopy (such as UV-Vis and near-infrared spectroscopy) has advantages such as speed, non-destructive nature, no need for complex pretreatment, and widespread availability of equipment, and has been explored for use in cosmetic ingredient analysis. However, cosmetic matrices typically contain multiple coexisting components such as moisturizers and preservatives, all of which absorb in the UV / near-infrared region, causing severe masking and interference with the characteristic spectrum of collagen. Traditional single-wavelength or multi-wavelength quantitative methods cannot eliminate matrix interference, resulting in poor detection accuracy and stability, and have long hindered industrial application. Existing methods for detecting collagen using near-infrared spectroscopy, such as CN121606259A (a method, electronic device and computer-readable medium for detecting collagen density) and CN 119470324 A (a method and apparatus for detecting collagen), although they also use infrared spectroscopy and pretreatment, are for detecting collagen density in biological tissues, not the content of recombinant type III collagen in cosmetic aqueous solutions. Their technical solutions do not solve the technical problem that collagen characteristic signals are severely obscured in the case of complex cosmetic matrices, high water content, and coexistence of multiple excipients, nor do they achieve integrated quantitative and qualitative detection.

[0005] In recent years, the development of chemometric methods has provided new approaches for complex spectral analysis. Spectral preprocessing techniques (such as multivariate scattering correction, wavelet transform, derivative processing, etc.) can effectively remove noise, baseline drift, and light scattering interference; variable selection algorithms (such as Butterfly Optimization Algorithm (BOA) and Whale Optimization Algorithm (WOA)) can screen out characteristic wavelengths highly correlated with the target component; and nonlinear regression or classification models (such as partial least squares regression, width learning systems, etc.) can achieve robust modeling of spectral signals and component concentrations / categories.

[0006] In the prior art, BOA and WOA have been used for the selection of spectral variables in other fields such as the analysis of Chinese medicinal materials, the detection of blood components, and the identification of edible oil quality. However, applying them to the detection of recombinant type III collagen in cosmetic aqueous solutions requires overcoming technical challenges unique to this field: (1) About 90% of the water in cosmetic aqueous solutions produces strong background absorption in the ultraviolet / near-infrared region, which seriously interferes with the extraction of characteristic signals of collagen; (2) The spectra of coexisting excipients such as moisturizers and preservatives highly overlap with the characteristic peaks of collagen, making it difficult for conventional variable selection methods to effectively distinguish between the target signal and background interference; (3) The concentration of collagen added in cosmetics is low (0.05%-3.0%), and the signal intensity is weak, which puts forward higher requirements for the sensitivity and anti-interference ability of the model.

[0007] Therefore, developing a rapid detection method based on the combination of molecular absorption spectroscopy and chemometrics, capable of simultaneously identifying the presence and absence of recombinant type III collagen in cosmetics and quantifying it with high precision, is of great practical significance and has broad application prospects for the quality monitoring of recombinant type III collagen in cosmetics. Summary of the Invention

[0008] To address the problems in existing technologies, this invention provides a rapid detection method for recombinant type III collagen based on molecular absorption spectroscopy and chemometrics. This method can simultaneously or separately achieve: rapid qualitative identification of whether cosmetics contain recombinant type III collagen; and accurate quantitative analysis of the content of recombinant type III collagen in cosmetics.

[0009] To achieve the above objectives, the present invention adopts the following technical solution.

[0010] A method for detecting type III collagen using spectrophotometry includes the following steps: S1. Add type III collagen or interfering components with a mass concentration not exceeding 3% to the matrix to obtain standard samples; The matrix was prepared by the following method: 4% 1,3-butanediol, 3% glycerol, and 0.02% EDTA-2Na were weighed according to their mass percentages, and water was added to make up to a fixed volume. The pH was then adjusted to 5-6 with citric acid or citrate. S2. Collect the molecular absorption spectra of the standard sample and the sample to be tested at room temperature to obtain spectral data; S3. Based on the purpose of type III collagen detection, the following processing steps are performed: S3-1. Content Detection: All spectral data are screened using one or more of the following algorithms: Competitive Adaptive Reweighted Sampling (CARS), Butterfly Optimization (BOA), Whale Optimization (WOA), and Grey Wolf Optimization (GWO) to identify characteristic wavelength variables highly correlated with type III collagen content. The spectral intensity corresponding to the screened characteristic wavelength variables is used as input, and the known concentration of type III collagen in the standard sample is used as output. A partial least squares regression (PLS) calibration model is established, and the optimal number of principal components is determined through cross-validation to obtain the calibration model. S3-2, Detecting Presence / Absence: Output all spectral data as either containing or not containing collagen, and establish a width learning system (BLS) binary classification model; S4. Based on the purpose of type III collagen detection, the following processing steps are performed: S4-1, Content Detection: Substitute the value of the optimal characteristic wavelength variable of the sample to be tested into the calibration model, and output the content of type III collagen in the sample to be tested; S4-2, Detection of Presence / Absence: Substitute the molecular absorption spectrum of the sample to be tested into the binary classification discrimination model, and output the result of whether the sample to be tested contains recombinant type III collagen.

[0011] The mass concentration of type III collagen in the matrix is ​​0 or 0.05%-2.5%.

[0012] The interfering component is at least one of sodium hyaluronate, dipotassium glycyrrhizate, phenoxyethanol, and acetyl hexapeptide-8.

[0013] The concentration of hyaluronic acid in the matrix is ​​0.1%-0.2%; the concentration of phenoxyethanol in the matrix is ​​0.5%-1.0%; and the concentration of acetyl hexapeptide-8 in the matrix is ​​0.005%-0.01%.

[0014] The molecular absorption spectrum is at least one of ultraviolet absorption spectrum, visible light absorption spectrum, and near-infrared absorption spectrum. In some embodiments, the molecular absorption spectrum is an ultraviolet-visible spectrum with a wavelength range of 185-800 nm and an optical path length of 0.5 mm or 1 mm. In some embodiments, the molecular absorption spectrum is a near-infrared spectrum with a wavelength range of 900-2200 nm and an optical path length of 1 mm or 2 mm.

[0015] Preferably, S2 further includes a step of preprocessing the spectral data to eliminate baseline drift, light scattering, and random noise; the preprocessing method is selected from at least one of multivariate scattering correction (MSC), continuous wavelet transform (CWT), standard normal variable transform (SNV), Savitzky-Golay smoothing, first derivative, and second derivative; The criteria for mandatory preprocessing are as follows: preprocessing is required when the baseline drift of the original spectrum exceeds 5% of the maximum signal value or the spectral signal-to-noise ratio is less than 30 dB; preprocessing is not required when the baseline drift does not exceed 5% of the maximum signal value and the signal-to-noise ratio is not less than 30 dB.

[0016] In some embodiments, when the molecular absorption spectrum is a UV-Vis spectrum, the preprocessing is a combination of MSC and CWT: the original spectral data is first subjected to multivariate scattering correction (MSC), and then the corrected spectrum is subjected to continuous wavelet transform (CWT); the wavelet basis function of the CWT is Haar or db3, and the decomposition scale is 36-37.

[0017] In some embodiments, when the molecular absorption spectrum is a UV-Vis spectrum, the Butterfly Optimization Algorithm (BOA) is used to screen characteristic wavelength variables, retaining 40-200 characteristic wavelength variables.

[0018] In some embodiments, when the molecular absorption spectrum is near-infrared, the Whale Optimization Algorithm (WOA) is used to screen characteristic wavelength variables and retain 40-200 characteristic wavelengths.

[0019] In some embodiments, the network parameters of the width learning system (BLS) are: 30-50 feature nodes, 20-40 windows, and 100-300 augmentation nodes; preferably, 41 feature nodes, 31 windows, and 201 augmentation nodes.

[0020] The present invention has the following advantages: This invention applies the Butterfly Optimization Algorithm (BOA) and the Whale Optimization Algorithm (WOA) to the spectral detection of recombinant type III collagen in cosmetic aqueous solutions for the first time, and reconstructs and optimizes the parameter system for this specific application scenario.

[0021] Through systematic experiments and parameter optimization, this invention has discovered and determined non-obvious technical solutions to the above-mentioned technical problems: (1) For ultraviolet-visible spectroscopy, a combination of 0.5 mm short optical path cuvette, MSC-CWT preprocessing, BOA characteristic wavelength selection, and PLS modeling is used to achieve a prediction set accuracy of R>0.998; (2) For near-infrared spectroscopy, a scheme of WOA characteristic wavelength selection and PLS modeling is used, which only requires 41 characteristic wavelengths to achieve R>0.998. 2The prediction performance was >0.993; (3) The width learning system (BLS) achieved a discrimination accuracy of 100% (ultraviolet) and 95.88% (near infrared) with a specific combination of parameters (N1=41, N2=31, N3=201). The above parameter selection is not something that can be expected by those skilled in the art through conventional optimization, but rather the optimal solution obtained through a large number of experimental screenings and comparative verifications.

[0022] Therefore, this invention is not a simple application of known algorithms, but rather an in-depth analysis of the unique technical problems in the detection of recombinant type III collagen in cosmetic aqueous solutions. It deeply integrates known algorithms with specific technical scenarios and systematically reconstructs parameters and optimizes processes for these scenarios, forming an integrated solution of algorithms and scenarios.

[0023] The detection method provided by this invention has a single sample spectral acquisition time of less than 1 minute, enabling high-throughput batch detection. The detection process does not consume chemical reagents, does not damage the sample, and the sample can be recycled for other tests. Through optimized multi-stage preprocessing and intelligent variable selection, spectral interference from coexisting components such as moisturizers, preservatives, and peptides in cosmetic matrices is effectively eliminated. It boasts high accuracy; qualitative identification using UV-Vis spectroscopy combined with a BLS model achieves 100% accuracy, while qualitative identification using near-infrared spectroscopy combined with a BLS model achieves over 95.8% accuracy. It is applicable to recombinant type III collagen aqueous cosmetics of different brands and concentrations, and can simultaneously achieve both quantitative and qualitative detection modes. Raw data is obtained using a spectroscopic absorption instrument, requiring versatile detection equipment and requiring minimal operator skill. This method offers advantages such as speed, efficiency, non-destructive testing, strong anti-interference ability, high accuracy, versatility, and low cost, solving the problem of existing recombinant type III collagen detection requiring expensive instruments or specific antibody kits. It is suitable for production line quality monitoring and large-scale sample quality inspection. Attached Figure Description

[0024] Figure 1 The original UV-Vis spectra (185-800 nm) of 141 samples containing different concentrations of recombinant type III collagen. Figure 2 A scatter plot showing the fitted values ​​of the predicted and actual values ​​from the quantitative model for ultraviolet-visible spectroscopy. Figure 3 This is a scatter plot showing the predicted values ​​and actual values ​​from the near-infrared spectroscopy quantitative model. Detailed Implementation

[0025] The present invention will be further described below with reference to the embodiments and accompanying drawings, but the present invention is not limited to the following embodiments.

[0026] Example 1: Quantitative analysis of type III collagen content using UV-Vis spectroscopy combined with the Butterfly Optimization Algorithm 1. Preparation of standard samples Prepare a blank matrix simulating commercially available toner: Weigh 80 g of 1,3-butanediol, 60 g of glycerin, and 0.4 g of disodium EDTA, dissolve them in deionized water, and bring the volume to 2000 mL. Adjust the pH to 5.5 with citric acid / sodium citrate. Different concentrations of recombinant type III collagen (concentration points: 0.05%, 0.07%, 0.09%, 0.11%, 0.13%, 0.15%, 0.17%, 0.19%, 0.20%, 0.30%, 0.50%, 0.80%, 1.00%, 1.50%, 2.00%, 2.50%, a total of 16 concentrations) were added to the blank matrix. Interfering components were also added: hyaluronic acid (0%, 0.1%, 0.2%), dipotassium glycyrrhizate (0%, 0.3%, 0.8%), phenoxyethanol (0%, 0.5%, 1.0%), and acetyl hexapeptide-8 (0%, 0.005%, 0.01%). Using an L9 orthogonal array, a total of 141 collagen-containing samples were obtained. Simultaneously, 150 blank control samples containing the same interfering components but without collagen were prepared. Each sample had a total volume of 5 mL, and the stock solutions were added sequentially and mixed thoroughly.

[0027] 2. Spectral Acquisition The absorbance spectra of each sample were measured using a UV-Vis spectrophotometer with a 0.5 mm quartz cuvette in the wavelength range of 185-800 nm. Each sample was measured three times, and the average spectrum was used to obtain the raw spectral data; the spectra are shown below. Figure 1 As shown in the figure, the absorbance increases overall with increasing collagen concentration, but baseline drift and noise interference are present.

[0028] 3. Sample Division The Kennard-Stone algorithm was used to divide the 141 samples into a calibration set (99 samples) and a prediction set (42 samples) in a 7:3 ratio.

[0029] 4. Calibration Model Construction (1) Modeling after variable selection using the butterfly optimization algorithm with original data (Model 1) The Butterfly Optimization Algorithm (BOA) was used to screen characteristic wavelengths from the original spectrum. BOA parameters were set as follows: population size 20, number of iterations 50, and fitness function was the root mean square error of cross-validation (RMSECV) of the PLS model. Ultimately, 102 characteristic wavelengths were selected, mainly concentrated in the UV absorption region of collagen (approximately 200-280 nm), while other redundant wavelengths were effectively removed. Using the spectral intensity of 102 selected characteristic wavelengths as independent variables and type III collagen concentration as the dependent variable, a partial least squares regression (PLS) model was established. Leave-one-out cross-validation determined the optimal number of principal components to be 14, resulting in the corrected model 1. (2) Modeling after variable selection using the butterfly optimization algorithm for preprocessed data (Model 2) The original spectral data were subjected to multivariate scattering correction (MSC) and continuous wavelet transform (CWT) sequentially. After parameter optimization, the CWT adopted the Haar wavelet basis and the decomposition scale was 36; the preprocessed spectrum was obtained; then the preprocessed spectrum was subjected to characteristic wavelength screening and optimal principal component number screening according to the method in Model 1 to obtain correction model 2.

[0030] 5. Result Prediction The spectra of 42 samples in the prediction set were substituted into calibration model 1 to predict their collagen concentration. Model 1 results showed: correlation coefficient R = 0.9981, root mean square error RMSEP = 0.0341, and mean absolute percentage error MAPE = 1.23%. A scatter plot of predicted and true values ​​is shown. Figure 2 The data shows a close distribution near the diagonal of y=x, indicating that the model has extremely high prediction accuracy.

[0031] The spectra of 42 samples in the prediction set were substituted into calibration model 2 to predict their collagen concentration. The results of model 2 showed that the correlation coefficient of the prediction set was R=0.9781, the root mean square error (RMSEP) was 0.0569, and the mean absolute percentage error (MAPE) was 2.87%.

[0032] Comparative analysis of Model 1 and Model 2: The prediction accuracy of Model 1 (original spectrum + BOA-PLS) (R=0.9981, RMSEP=0.0341) is better than that of Model 2 (MSC-CWT+BOA-PLS, R=0.9781, RMSEP=0.0569), indicating that in UV-Vis spectroscopy, when the spectral quality is good, directly using the BOA variable selection can obtain the best prediction effect. Model 2 is suitable for sample systems with severe baseline drift and can significantly improve model stability.

[0033] Comparative Example 1: Quantitative analysis of type III collagen content using UV-Vis spectroscopy combined with a PLS model (without variable selection) A PLS model was directly built using raw UV-Vis spectral data (without variable selection). The correlation coefficient of the prediction set was R=0.9679, and RMSEP=0.0957. This is significantly worse than the BOA-PLS scheme in Example 1 (R=0.9981), indicating that variable selection plays a crucial role in improving the model's prediction accuracy.

[0034] Example 2: Quantitative analysis of type III collagen content using near-infrared spectroscopy combined with a whale-optimized algorithm. Samples were prepared and spectral data were divided according to the method in Example 1.

[0035] 1. Spectral Acquisition Fourier transform near-infrared spectrometer was used to acquire the spectra of each sample in the wavelength range of 900-2200 nm using a 1 mm quartz cuvette, with a resolution of 8 cm⁻¹. -1 Each sample was scanned 32 times and the average was taken.

[0036] The SPXY algorithm was used to divide the 141 samples into a calibration set (94 samples) and a prediction set (47 samples) at a ratio of 2:1. 2. Calibration Model Construction The Whale Optimization Algorithm (WOA) was used for feature wavelength selection. WOA parameters were: population size 30, number of iterations 100, and fitness function RMSECV of the PLS model. Ultimately, 41 feature wavelengths were retained, mainly distributed in the 1100-1700 nm region, which coincides with the CH, NH, and OH overtone and combination absorption bands of collagen. Using the spectral intensity of the 41 retained characteristic wavelengths as independent variables and collagen concentration as dependent variable, a PLS model was established. After cross-validation, the optimal number of principal components was determined to be 14, and a corrected model was obtained.

[0037] 3. Result Prediction The spectra of 47 samples in the prediction set were substituted into the calibration model to predict their type III collagen concentration. The results showed that the correlation coefficient R = 0.9936, RMSEP = 0.0302, and MAPE = 2.15%. The scatter plot is shown below. Figure 3 As shown.

[0038] Comparative Example 2: Quantitative analysis of type III collagen content using near-infrared spectroscopy combined with a PLS model (without variable selection) A PLS model was directly built using raw near-infrared spectral data (without variable selection), with a prediction set R. 2 =0.8332, RMSEP=0.1530. This is significantly inferior to the WOA-PLS scheme in Example 2 (R... 2 =0.9936, RMSEP=0.0302), indicating that variable selection also plays a crucial role in quantitative analysis of near-infrared spectroscopy.

[0039] Example 3: UV-Vis spectroscopy combined with width learning system for identifying the presence of type III collagen 1. Sample Preparation 141 samples containing recombinant type III collagen (same as in Example 1) and 150 samples without recombinant type III collagen were prepared. The collagen-free samples were identical to the collagen-containing samples in terms of matrix and interfering components, except for the absence of added collagen, and were obtained through fixed blank controls and stratified random sampling. These collagen-free samples served as negative controls to train the BLS model to identify spectral features against a background of interfering substances.

[0040] 2. Spectral Acquisition Same as Example 1.

[0041] 3. Sample Splitting All 291 samples were divided into a training set (194 samples) and a prediction set (97 samples) using the Kennard-Stone algorithm. The ratio of samples containing / not containing collagen in the training set and the prediction set remained consistent with the overall population.

[0042] 4. Qualitative Model Establishment A binary classification model was built using a width learning system (BLS). The BLS network structure consisted of 41 feature nodes (N1), 31 windows (N2), and 201 augmentation nodes (N3). The input was the preprocessed full spectrum (701 wavelengths), and the output was the class label (1: contains collagen, 0: does not contain collagen). The model used ridge regression to solve for the output layer weights.

[0043] 5. Prediction Results All 97 samples in the prediction set were correctly classified with an accuracy of 100% and a Kappa coefficient of 1.000.

[0044] Example 4: Near-infrared spectroscopy combined with width learning system for identifying the presence of type III collagen Sample preparation was the same as in Example 3.

[0045] The spectral acquisition method is the same as in Example 2.

[0046] Qualitative model establishment: BLS parameters are the same as in Example 3 (N1=41, N2=31, N3=201).

[0047] Prediction results: Prediction accuracy was 95.88%, and the Kappa coefficient was 0.956.

[0048] Comparative Example 3: Qualitative Analysis Using PCA Only Principal component analysis (PCA) was used to cluster samples containing and without collagen. The scatter plots of the first and second principal components showed that the two types of samples overlapped significantly and could not be effectively distinguished, with an accuracy of only about 56%. This indicates that unsupervised methods are not suitable for this system, and a supervised classification model (such as BLS) must be used.

[0049] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for detecting type III collagen, characterized in that, Includes the following steps: S1. Add type III collagen or interfering components with a mass concentration not exceeding 3% to the matrix to obtain standard samples; The matrix was prepared by the following method: Weigh 4% 1,3-butanediol, 3% glycerol, and 0.02% EDTA-2Na according to the mass percentage, add water to make up the volume, and adjust the pH to 5-6 with citric acid or citrate. S2. Collect the molecular absorption spectra of the standard sample and the sample to be tested at room temperature to obtain spectral data; S3. Based on the purpose of type III collagen detection, the following processing steps are performed: S3-1. Content Detection: All spectral data are screened using one or more of the following algorithms: competitive adaptive reweighted sampling, butterfly optimization algorithm, whale optimization algorithm, and gray wolf optimization algorithm, to identify characteristic wavelength variables that are highly correlated with the content of type III collagen. The spectral intensity corresponding to the screened characteristic wavelength variables is used as input, and the known concentration of type III collagen in the standard sample is used as output. A partial least squares regression correction model is established, and the optimal number of principal components is determined through cross-validation to obtain the correction model. S3-2, Detecting Presence / Absence: Using all spectral data as outputs (containing or not containing type III collagen), establish a binary classification discriminant model for the width learning system. S4. Based on the purpose of type III collagen detection, the following processing steps are performed: S4-1, Content Detection: Substitute the value of the optimal characteristic wavelength variable of the sample to be tested into the calibration model, and output the content of type III collagen in the sample to be tested; S4-2, Detection of presence or absence: Substitute the molecular absorption spectrum of the sample to be tested into the binary classification discrimination model, and output the result of whether the sample to be tested contains type III collagen; The interfering component is at least one of sodium hyaluronate, dipotassium glycyrrhizate, phenoxyethanol, and acetyl hexapeptide-8; the concentration of hyaluronic acid in the matrix is ​​0.1%-0.2%; the concentration of phenoxyethanol in the matrix is ​​0.5%-1.0%; and the concentration of acetyl hexapeptide-8 in the matrix is ​​0.005%-0.01%.

2. The method according to claim 1, characterized in that, The mass concentration of type III collagen in the matrix is ​​0 or 0.05%-3%.

3. The method according to claim 1, characterized in that, The molecular absorption spectrum is at least one of ultraviolet absorption spectrum, visible light absorption spectrum, and near-infrared absorption spectrum.

4. The method according to claim 3, characterized in that, The molecular absorption spectrum is an ultraviolet-visible spectrum with a wavelength range of 185-800 nm and an optical path of 0.5 mm or 1 mm; and a near-infrared spectrum with a wavelength range of 900-2200 nm and an optical path of 1 mm or 2 mm.

5. The method according to claim 1, characterized in that, S2 also includes a step of preprocessing the spectral data to eliminate baseline drift, light scattering, and random noise; the preprocessing method is selected from at least one of multivariate scattering correction, continuous wavelet transform, standard normal variable transform, Savitzky-Golay smoothing, first derivative, and second derivative; The standard for preprocessing is: when the baseline drift of the original spectrum exceeds 5% of the maximum signal value, or when the spectral signal-to-noise ratio is less than 30 dB.

6. The method according to claim 5, characterized in that, When the molecular absorption spectrum is a UV-Vis spectrum, the preprocessing is a combination of multivariate scattering correction and continuous wavelet transform. The original spectral data is first subjected to multivariate scattering correction, and then the corrected spectrum is subjected to continuous wavelet transform. The wavelet basis function of the continuous wavelet transform is Haar or db3, and the decomposition scale is 36-37.

7. The method according to claim 1, characterized in that, When the molecular absorption spectrum is ultraviolet-visible, the butterfly optimization algorithm is used to screen characteristic wavelength variables and retain 40-200 characteristic wavelength variables. When the molecular absorption spectrum is near-infrared, the whale optimization algorithm is used to screen characteristic wavelength variables and retain 40-200 characteristic wavelengths.

8. The method according to claim 1, characterized in that, The network parameters of the width learning system are: 30-50 feature nodes, 20-40 windows, and 100-300 augmentation nodes; the preferred number of feature nodes is 41, the number of windows is 31, and the number of augmentation nodes is 201.

Images