Three-dimensional fluorescence spectrum benzene series identification method combining spectral line fitting and total variation denoising

Abstract

The invention discloses a three-dimensional fluorescence spectrum benzene series identification method combining spectral line fitting and total variation denoising, and belongs to the field of analysis and detection. Comprising the following steps: collecting and preprocessing three-dimensional fluorescence spectrum data of pure water background sample solutions with different concentrations of single benzene series pollutants, a pollution-free normal river water sample and a to-be-detected river water sample with unknown target benzene series pollutant existence conditions; carrying out pollutant fluorescence peak information extraction on the preprocessed data by adopting a spectrum peak searching method, and establishing a pollutant spectrum peak information base; feature vectors are extracted through principal component analysis, and a support vector machine classification model is constructed; doppler-Gaussian spectral line fitting is combined with total variation de-noising and feature extraction on preprocessed spectral data of a river water sample to be detected, and then a classification model is used for target discrimination. According to the method, the three-dimensional fluorescence spectrum characteristics of the target pollutants can be effectively extracted, the problem of spectrum peak overlapping is solved, and the recognition accuracy and rapidity are improved.

Description

Technical Field

[0001] The present invention belongs to the field of analytical detection, and relates to a method for identifying benzene series compounds in three-dimensional fluorescence spectra by combining spectral line fitting and total variation denoising. Background Art

[0002] Water quality safety is closely related to people's livelihood. As a river that serves as the water source for many civilian uses, there are various types of pollution occurring in it. Industrial organic pollution represented by benzene series pollutants is particularly prominent among many pollution incidents due to its high occurrence frequency and great pollution hazards. Therefore, how to accurately and efficiently identify benzene series pollutants in rivers has become an urgent problem to be solved.

[0003] Since benzene series compounds generally have obvious and specific three-dimensional fluorescence signals, the three-dimensional fluorescence method can effectively identify benzene series pollutants and has been applied in the detection of benzene series pollutants. The commonly used three-dimensional fluorescence field detection method usually extracts the fluorescence characteristics in the water sample and compares them with the fluorescence library of pollutants established in the laboratory and other environments to determine the types of pollutants. However, in the actually polluted rivers, in addition to the target pollutants, there are other substances that can produce fluorescence reactions, such as organic acids with less harm and amino acids in aqueous solutions. The measured three-dimensional fluorescence spectrum is a linear superposition of the spectral peaks of the target pollutants, the spectral peaks of other substances, and non-spectral peak interference information, that is, the spectral peaks of the target pollutants will overlap with those of other substances, thus affecting the accuracy of detection. In previous studies, the parallel factor method was often used to deal with the problem of spectral peak overlap between the substances to be measured and the water quality background and other substances in three-dimensional fluorescence detection. In other optical detection fields, spectral peak overlap is also an urgent problem to be solved, and there are also some solutions in detection methods such as infrared and terahertz. For example, DU Zhiheng et al. proposed a sharpened error wavelet method in the study of a sharpened error wavelet EDXRF spectral analysis method, which successfully separated the spectral peaks of manganese, iron, and aluminum metals with a high degree of overlap; CHEN Jiwen et al. proposed a spectral peak resolution method based on the particle swarm optimization (PSO) in the rapid spectral peak resolution algorithm based on multi-peak cooperation and normalization of the characteristic peak areas of pure elements, which can achieve the separation of the spectral peaks of toluene, o-xylene, and m-xylene; MICHAEL A et al. proposed using the evolution factor method for spectral peak resolution and achieved the separation of the spectral peaks of nickel and titanium; XIONG J Y et al. proposed a spectral peak resolution method based on the multi-order difference method and genetic algorithm (GA), which can effectively separate the infrared absorption overlapping peaks of methane, ethane, and propane in alkanes.

[0004] These methods can reduce the impact of spectral peak overlap on the accuracy of final pollutant detection to a certain extent. However, due to the lack of support from fluorescence mechanisms based on mathematical statistics and the lack of interpretability in the intermediate process, a large number of sample pre-trainings and prior knowledge are often required. For example, the parallel factor method needs to determine the number of factors in advance, manually distinguish the advantages and disadvantages of the emission spectra and fluorescence spectra obtained by decomposing different numbers of factors, and select the optimal number of components. In a new environment, with the change of the position of overlapping fluorescence peaks, the detection accuracy will be greatly affected. Therefore, it is difficult for such methods and models to meet the requirements of migration between different rivers and on-site detection. Summary of the Invention

[0005] The object of the present invention is to address the problems that the fluorescence spectral peaks of other substances in the river background overlap with those of target pollutants, affecting the detection accuracy, requiring a large number of sample pre-trainings, and the model being difficult to migrate in different river backgrounds. A method for identifying benzene series compounds in three-dimensional fluorescence spectra by spectral line fitting combined with total variation denoising is proposed.

[0006] To achieve the object of the present invention, the technical solution adopted by the present invention is: to provide a method for identifying benzene series compounds in three-dimensional fluorescence spectra by spectral line fitting combined with total variation denoising, including the following steps: Collect the three-dimensional fluorescence spectra of the pure water background sample solution of the target benzene series pollutant and the non-polluted normal river water sample, and use them as the target benzene series pollutant spectrum and the background spectrum respectively; Adopt a spectral peak search method to extract the fluorescence spectral peak position information of the target benzene series pollutant spectrum; extract the eigenvectors of the target benzene series pollutant spectrum and the background spectrum through principal component analysis, establish a data set, use the presence or absence of pollution as a label, and construct a classification model based on machine learning; Collect the three-dimensional fluorescence spectrum of the sample to be tested, adopt a spectral peak search method to extract the fluorescence spectral peak position information of the three-dimensional fluorescence spectrum of the sample to be tested, and remove the fluorescence spectral peaks whose distances from the fluorescence spectral peak positions of the target benzene series pollutant spectrum are less than the threshold; use the collision broadening formula to fit the excitation spectrum and remove the fluorescence spectral peaks whose fitting degrees are lower than the threshold to obtain the interference spectral peaks; take the excitation wavelength or emission wavelength of the interference spectral peaks as the boundary, use the Doppler broadening formula for the part where the excitation wavelength or emission wavelength is lower than the boundary, and use the collision broadening formula for the part where it is higher than the boundary to fit the excitation spectra at each emission wavelength and the emission spectra at each excitation wavelength respectively to obtain the three-dimensional fluorescence spectrum of the interference substance; based on total variation denoising, strip the three-dimensional fluorescence spectrum of the interference substance from the three-dimensional fluorescence spectrum of the sample to be tested to obtain a characteristic spectrum; extract the eigenvector of the characteristic spectrum through principal component analysis and input it into the classification model to determine whether the target benzene series pollutant exists.

[0007] Further, the specific process of sample processing after collection is as follows: after the river water sample is collected, it is filtered, stored at 4°C and protected from light, and the three-dimensional fluorescence spectrum data of the river water sample is recorded within one hour.

[0008] Preferably, after acquiring the three-dimensional fluorescence spectrum, preprocessing is further included; the preprocessing includes: removing Raman scattering, removing Rayleigh scattering, and smoothing.

[0009] Further, the spectral peak search method is specifically: searching for local maxima in the three-dimensional fluorescence spectrum, calculating the average fluorescence intensity and variance within a 15nm×15nm region near the local maximum; if the local maximum is greater than the sum of the average value and three times the variance, extracting the fluorescence spectral peak position corresponding to the local maximum, and the fluorescence spectral peak position is represented by the excitation wavelength and emission wavelength corresponding to the fluorescence spectral peak.

[0010] Preferably, the support vector machine is used for constructing the classification model based on machine learning.

[0011] Further, the fluorescence spectral peaks with a distance less than the threshold from the fluorescence spectral peak position of the target benzene series pollutant spectrum are removed by using the Euclidean distance formula.

[0012] Further, the collected pure water background sample solution of the target benzene series pollutant contains multiple concentrations. In the step of extracting the fluorescence spectral peak position information of the three-dimensional fluorescence spectrum of the sample to be measured and removing the fluorescence spectral peaks with a distance less than the threshold from the fluorescence spectral peak position of the target benzene series pollutant spectrum, the fluorescence spectral peaks with a distance less than the threshold from the fluorescence spectral peak position of the target benzene series pollutant spectrum at any concentration are removed.

[0013] Further, taking the excitation wavelength of the interference spectral peak as the boundary, the following Doppler broadening formula is used for the part where the excitation wavelength is lower than the boundary: ; The following collision broadening formula is used for the part where the excitation wavelength is higher than the boundary: ; where, is the fluorescence intensity, is the excitation spectrum peak value corresponding to the emission wavelength, is the spectral peak excitation wavelength, is the spectral peak emission wavelength, is the spectral half-line width, is the independent variable excitation wavelength, is the independent variable emission wavelength.

[0014] Further, based on total variation denoising, removing the three-dimensional fluorescence spectrum of the interference substance from the three-dimensional fluorescence spectrum of the sample to be measured to obtain the characteristic spectrum is specifically: The following total variation model is adopted and Bregman iteration is used for denoising: ; ; Among them, is the energy functional, is a non - negative parameter, is the target characteristic spectrum to be obtained, is the spectrum of external interferents, is the spectrum directly measured by the instrument, is the regularization term used to maintain the smoothness of the spectrum; At this time, the functional to be minimized is: .

[0015] The beneficial effects of the present invention are as follows: Aiming at the problems that the detection accuracy of the existing technology is easily affected by the overlapping of fluorescence peaks of interferents and the poor migration of the model among different rivers, by using the spectral line broadening theory of natural optics, the Doppler - Gaussian spectral line fitting combined with total variation denoising method is proposed to strip the spectral peaks of the fluorescence peaks of interferents in the river background, improving the anti - interference ability of the detection method; the technical solution based on the spectral line broadening theory can have high migration applicability in different river scenarios. Aiming at the problems that the current benzene - series pollutant identification method lacks the support of fluorescence mechanism and requires a large number of samples for preliminary training, the present invention uses the three - dimensional fluorescence spectra of different - concentration samples of target pollutants and the three - dimensional fluorescence spectra of river background for preliminary model training, reducing the requirements for preliminary training samples. The spectral interference signals of the samples to be measured are stripped by using the Doppler - Gaussian spectral line fitting combined with the total variation denoising algorithm, and then the principal component analysis and the existing support vector machine classifier are used to achieve efficient and accurate identification of benzene - series pollutants under different complex river backgrounds, while making the establishment of the model simpler and more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the flow chart of the method of the present invention; Figure 2 is the three - dimensional fluorescence spectra diagram of four typical benzene - series pollutants; among them, Figure 2 in (a) is the three - dimensional fluorescence spectra diagram of ethylbenzene, Figure 2 in (b) is the three - dimensional fluorescence spectra diagram of styrene, Figure 2 in (c) is the three - dimensional fluorescence spectra diagram of phenanthrene, Figure 2 in (d) is the three - dimensional fluorescence spectra diagram of phenol; Figure 3 is the flow chart of spectral peak search; Figure 4 is the schematic diagram of spectral peak extraction results taking styrene as an example; Figure 5 is the three - dimensional fluorescence spectra diagram of two background water samples; among them, Figure 5 in (a) is the three - dimensional fluorescence spectra diagram of normal river water,Figure 5 In (b) is the three-dimensional fluorescence spectrum of river water with high-concentration dissolved organic matter; Figure 6 is the flow chart of the feature extraction method combining Doppler-Gaussian spectral fitting and total variation denoising (DCF-TV); Figure 7 is the schematic diagram of the effect of excitation spectrum fitting taking phenol as an example; Figure 8 is the comparison chart of the fitting effect of three-dimensional fluorescence spectral peaks taking styrene as an example; among them, Figure 8 In (a) is the original three-dimensional fluorescence spectrum, Figure 8 In (b) is the three-dimensional fluorescence spectrum after fitting. Specific implementation manner

[0017] The present invention proposes a classification model for rapid discrimination of benzene series pollutants. The fluorescence peak information of pollutants is extracted from the collected three-dimensional fluorescence spectrum by the local maximum method, the average value and variance of the spectral intensity are calculated, a fluorescence spectral peak information library is established, the principal component analysis (PCA) is used to extract feature vectors, and combined with pollution and non-pollution labels, a support vector machine (SVM) classification model is constructed. At the same time, a method combining spectral fitting and total variation denoising for solving fluorescence spectral peak overlap is proposed: the excitation spectrum and emission spectrum of the spectral peak are fitted using the Doppler broadening and collision broadening models, and extended to the three-dimensional spectrum by interpolation. The total variation model is used to denoise and strip the interference spectral peaks to obtain a pure characteristic spectrum, and the SVM classification model is used to judge the presence of pollutants. The core technology is to use the Doppler-Gaussian spectral fitting combined with the total variation denoising algorithm to strip the interference spectral peaks in the river background to obtain a relatively pure characteristic spectrum, and perform principal component analysis on it with the pollutant spectrum of pure water solvent and the non-polluted river background spectrum to extract feature vectors, and use the established support vector machine to identify the presence of target benzene series pollutants.

[0018] The present invention proposes a method for identifying benzene series substances in three-dimensional fluorescence spectra by combining spectral fitting and total variation denoising. The step process is as Figure 1 shown, and the specific steps are as follows: S1: Obtain pure water background sample solutions of single benzene series pollutants (ethylbenzene, styrene, phenanthrene, and phenol) at different concentrations, and collect the three-dimensional fluorescence spectrum data of each sample; obtain non-polluted normal river water samples, and after sample treatment, collect the three-dimensional fluorescence spectrum data of the river water samples. The pollutant concentration in the pure water background solutions of the four benzene series pollutants: the concentration range of ethylbenzene is 0.2 - 1 mg / L, and the concentration ranges of phenol, styrene, and phenanthrene are all 10 - 50 μg / L. The specific sample treatment is as follows: after the river water sample is collected, it is filtered, stored at 4°C and shielded from light, and the three-dimensional fluorescence spectrum data of the river water sample is recorded within one hour.

[0019] In this embodiment, normal river water samples are taken at seven sampling points. The specific locations are shown in Table 1. The sampling points for normal river water samples are not limited to the locations shown in Table 1.

[0020] Table 1: Longitude and Latitude Coordinate Table of River Water Sampling Points

[0021] S2: Data preprocessing: Perform spectral preprocessing on the three-dimensional fluorescence spectra and background spectra of each benzene series pollutant under pure water background to create a data set. The specific operation is as follows: Perform Raman scattering removal, Rayleigh scattering removal, and smoothing on the three-dimensional fluorescence spectra and background spectra of the benzene series pollutants.

[0022] As Figure 2 shown is the three-dimensional fluorescence spectrogram of four typical benzene series pollutants. Figure 2 In Figure 2 it, (a) is the three-dimensional fluorescence spectrogram of ethylbenzene. Figure 2 In Figure 2 it, (b) is the three-dimensional fluorescence spectrogram of styrene.

[0023] S3: Use the spectral peak search method to extract the fluorescence peak information of the three-dimensional fluorescence spectra of the preprocessed benzene series pollutant samples to obtain the fluorescence peak position information (l is an integer, 1 ≤ l ≤ K), establish a pollutant spectral peak information library, where K is the total number of fluorescence peaks searched. is the excitation wavelength of the l-th fluorescence peak. is the emission wavelength of the l-th fluorescence peak.

[0024] The specific spectral peak search method is as follows: First, search for the local maximum in the three-dimensional fluorescence spectrum ; calculate the average fluorescence intensity in the 15nm × 15nm area near the local maximum , and the variance ; determine whether the searched local maximum meets the requirements to obtain the fluorescence spectral peak position information.

[0025] Figure 3 is the spectral peak search flow chart, whose purpose is to screen out the fluorescence peak information of all possible target pollutants in the three-dimensional fluorescence spectrum, including the spectral peak position and its spectral peak fluorescence intensity, providing prior information for subsequent spectral peak stripping.

[0026] Figure 4 is a schematic diagram of the spectral peak extraction result taking styrene as an example. The positions marked by the red circles in the figure are the fluorescence spectral peak positions searched by the spectral peak search method.

[0027] S4: Extract the eigenvectors of the three-dimensional fluorescence spectra and background spectra of benzene series pollutants using principal component analysis, and construct a support vector machine classification model using the eigenvectors and their corresponding pollution labels (present or absent).

[0028] S5: Obtain a sample of the river water to be tested for the presence of unknown target benzene series pollutants, process the sample according to the sample processing method in step S1, collect its three-dimensional fluorescence spectral data, and perform spectral preprocessing according to step S2 to obtain the spectral data of the sample to be tested after preprocessing.

[0029] In this embodiment, the unknown sample is composed of a single mixture of two backgrounds and four benzene series pollutants. The specific sample conditions are shown in Table 2. Among them, background I is normal river water, and background II is the background of river water with high-concentration dissolved organic matter (fulvic acid with a final concentration of 20 mg / L and tryptophan with a final concentration of 2 mg / L were added to normal river water).

[0030] Figure 5 are the three-dimensional fluorescence spectra of two background water samples; Figure 5 In (a) is the three-dimensional fluorescence spectrum of normal river water, Figure 5 In (b) is the three-dimensional fluorescence spectrum of river water with high-concentration dissolved organic matter.

[0031] Table 2: Composition table of unknown samples

[0032] S6: Use the Doppler-Gaussian spectral fitting combined with total variation denoising (DCF-TV) algorithm for the spectral data after preprocessing in step S5 to obtain the characteristic spectrum for the target benzene series pollutants, eliminating the interference of background spectral information. Perform principal component analysis on the characteristic spectrum together with the three-dimensional fluorescence spectral data of the target pollutants and normal river water in step S2 to extract eigenvectors, and use the support vector machine classification model established in step S4 to determine whether the target benzene series pollutants are present in the eigenvectors of the sample to be tested.

[0033] Figure 6 is the flow chart of the Doppler-Gaussian spectral fitting combined with total variation denoising feature extraction method, including the following sub-steps: S6.1: Perform spectral peak search based on local maximum discrimination on the three-dimensional fluorescence spectrum of the sample to be tested after preprocessing according to the spectral peak search method in step S3 to obtain the positions of fluorescence spectral peaks that meet the requirements , represents the excitation wavelength corresponding to the fluorescence spectral peak, represents the emission wavelength corresponding to the fluorescence spectral peak.

[0034] S6.2: Compare the fluorescence peak positions extracted in step S6.1 with the fluorescence spectral peak positions of the target benzene series pollutants extracted in step S3.

[0035] The specific operation is as follows: Calculate the minimum distance between the fluorescence peak positions extracted in step S6.1 and the fluorescence spectral peak positions of the target benzene series pollutants extracted in step S3. The calculation process is as follows: ; and are the excitation and emission wavelengths corresponding to the fluorescence peak of the sample to be measured extracted in step S6.1, and are the fluorescence spectral peak position information of the target benzene series pollutants, corresponding to the excitation wavelength and emission wavelength respectively, is the threshold for spectral peak screening. If the distance is less than the threshold, it is considered to be the spectral peak of the target benzene series pollutants, and the threshold size is set according to the actual situation.

[0036] Among them, the target benzene series pollutants for comparison include all concentrations used in step S1. If the calculation result of any concentration is lower than the threshold, it can be considered that the fluorescence peak position information belongs to the target benzene series pollutants.

[0037] S6.3: For the spectral peaks screened in step S6.2, use the collision broadening formula of spectral line broadening to fit the excitation spectrum of each spectral peak. For the spectral peaks with a fitting degree lower than the threshold ( ), they are identified as non-substance interference peaks such as instrument noise, and spectral peak secondary screening is carried out; the remaining spectral peaks after removing the spectral peaks of the target benzene series pollutants and non-substance interference peaks in the two screenings are identified as the fluorescence spectral peaks of other substances other than benzene series pollutants.

[0038] The specific operation is as follows: For the spectral peaks screened in step S6.2 (removing the spectral peaks of the target benzene series pollutants), after obtaining their fluorescence spectral peak position information and fluorescence intensity , perform spectral line fitting on their excitation spectra, calculate the fitting degree . For the spectral peaks with a fitting degree lower than the threshold ( ), they are identified as non-substance interference peaks such as instrument noise and are ignored; the remaining spectral peaks after screening are identified as the fluorescence spectral peaks of other substances other than benzene series pollutants (i.e., interference spectral peaks).

[0039] The calculation method for the excitation spectrum fitting of collision broadening in step S6.3 is as follows: ; Among them, is the spectral peak excitation wavelength, is the spectral half-line width, The independent variable excitation wavelength for fitting the curve.

[0040] S6.4: For the fluorescence spectral peaks of other substances screened in step S6.3, use the spectral line fitting method to fit the excitation spectrum and emission spectrum of each spectral peak, and then interpolate and expand it to a three-dimensional fluorescence spectrum to achieve the mathematical characterization of the interference spectral peaks; use total variation denoising to strip the spectral peaks of other substances to obtain a relatively pure characteristic spectrum of the sample to be tested for subsequent classification and discrimination.

[0041] The specific method of spectral line fitting for the excitation spectrum and emission spectrum in step S6.4 is as follows: Taking the excitation spectrum as an example, the spectral line shape expression of its Doppler broadening is: .

[0042] The spectral line shape expression of its collision broadening is: .

[0043] The spectral line shape expression of its Voigt broadening is: .

[0044] Among them, is the spectral peak excitation wavelength, is the spectral half-line width, is the wavelength broadening amount, and its calculation formula is: ; Among them is the spectral line width, is the speed of light in vacuum, is the Boltzmann constant, is the absolute temperature, is the gas molecular mass.

[0045] Figure 7 Figure is a schematic diagram of the fitting effect of the excitation spectrum with phenol as an example. Among them, the unit of the excitation wavelength is nm. In the actual fitting process, for the case of a single fluorescence spectral peak, taking the excitation wavelength of the spectral peak as the boundary, the Doppler broadening fitting effect is better at low excitation wavelengths, while the collision broadening fitting effect is better at high wavelengths.

[0046] The specific method of interpolation and expansion of the three-dimensional fluorescence spectrum in step S6.4 is as follows: Assume that the excitation spectral curves at any emission wavelength conform to the spectral line broadening theory. Take the fluorescence intensity value on the emission spectrum corresponding to the spectral peak as the excitation spectral peak at the corresponding emission wavelength, and use a piecewise Doppler-collision broadening model to fit the excitation spectra at each emission wavelength. Taking the excitation wavelength of the spectral peak as the boundary, the Doppler broadening fitting is used at low excitation wavelengths, while the collision broadening fitting is used at high wavelengths, that is, it is obtained by fitting with the following formula: When is adopted: ; When is adopted: ; Among them, is the fluorescence intensity, is the excitation spectral peak at the corresponding emission wavelength, is the spectral peak excitation wavelength, is the spectral peak emission wavelength, is the spectral half-line width, is the independent variable excitation wavelength, is the independent variable emission wavelength.

[0047] Similarly, using a piecewise Doppler-collision broadening model, with the emission wavelength of the interference spectral peak as the boundary, the Doppler broadening formula is used at emission wavelengths below the boundary, and the collision broadening formula is used at wavelengths above the boundary to fit the emission spectra at each excitation wavelength, and finally a three-dimensional fluorescence spectrogram is extended, which is the spectrum of the external interference substance.

[0048] Figure 8 is a comparison diagram of the three-dimensional fluorescence spectral peak fitting effect with styrene as an example; Figure 8 In (a) is the original three-dimensional fluorescence spectrum, Figure 8 In (b) is the three-dimensional fluorescence spectrum after fitting. According to the image comparison, the similarity of the spectral peak fitting is relatively high, and there are fitting differences at the spectral peak edges.

[0049] The specific method of spectral peak stripping by total variation denoising in step S6.4 is as follows: Adopt the following total variation model and use Bregman iteration for denoising.

[0050] ; ; Among them, is the energy functional, is a non-negative parameter, is the characteristic spectrum to be obtained, is the spectrum of the external interference substance, is the spectrum directly measured by the instrument (i.e., the three-dimensional fluorescence spectral data collected in step S5), It is a regularization term for maintaining spectral smoothness.

[0051] At this time, the functional to be minimized is: .

[0052] Among them, the spectra of the interfering substances have been characterized as mathematical functions by the spectral line broadening fitting method: ; Among them, is the central wavelength of the excitation spectrum, is the central wavelength of the emission spectrum, is the standard deviation, which is used to describe the broadening degree of the spectral line, is the excitation wavelength , and the emission wavelength is under the fluorescence intensity.

[0053] In this specific embodiment, accuracy, precision, recall, and F1-score are selected as evaluation indicators.

[0054] For the four benzene series pollutants under background I, the detection effect of the present invention is shown in Table 3.

[0055] Table 3: Table of various evaluation indicators for four benzene series pollutants under background I

[0056] For the four benzene series pollutants under background II, the detection effect of the present invention is shown in Table 4.

[0057] Table 4: Table of various evaluation indicators for four benzene series pollutants under background II

[0058] The detection effect of the present invention under the two backgrounds demonstrates the high mobility of the method of the present invention under different river detection backgrounds. Facing background II that was not involved in training, the detection effect of the method of the present invention changes little and still remains at a relatively high level, showing the anti-interference performance in the detection of complex water sample backgrounds.

[0059] The above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A three-dimensional fluorescence spectrum benzene identification method based on spectral line fitting combined with total variation denoising, characterized in that: The following steps are involved: Collect three-dimensional fluorescence spectra of pure water background sample solution of target benzene series pollutants and normal river water samples without pollution, which are used as target benzene series pollutant spectrum and background spectrum respectively; The peak search method is used to extract the fluorescence peak position information of the target benzene pollutant spectrum. The characteristic vectors of the target benzene pollutant spectrum and the background spectrum are extracted through principal component analysis, and a data set is established. The presence or absence of pollution is used as a label, and a classification model is constructed based on machine learning. Collect the three-dimensional fluorescence spectrum of the sample to be tested, use the peak search method to extract the fluorescence peak position information of the three-dimensional fluorescence spectrum of the sample to be tested, and remove the fluorescence peak whose distance from the fluorescence peak position of the target benzene pollutant spectrum is less than the threshold; use the collision broadening formula to fit the excitation spectrum, remove the fluorescence peak whose fitting degree is lower than the threshold, and obtain the interference peak; The excitation wavelength or emission wavelength of the interfering substance's spectral peak is taken as the boundary. The Doppler broadening formula is used when the excitation wavelength or emission wavelength is lower than the boundary, and the collision broadening formula is used when it is higher than the boundary. The excitation spectrum at each emission wavelength and the emission spectrum at each excitation wavelength are fitted respectively to obtain the three-dimensional fluorescence spectrum of the interfering substance. Based on total variation denoising, the three-dimensional fluorescence spectrum of the interfering substance is stripped from the three-dimensional fluorescence spectrum of the sample to be tested to obtain the characteristic spectrum; The characteristic vector of the characteristic spectrum is extracted through principal component analysis and input into the classification model to determine whether the target benzene pollutant exists.

2. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: After the three-dimensional fluorescence spectrum is collected, preprocessing is also performed; the preprocessing includes: removing Raman scattering, removing Rayleigh scattering, and smoothing.

3. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The peak search method is specifically as follows: searching for the local maximum in the three-dimensional fluorescence spectrum, calculating the average and variance of the fluorescence intensity in a 15nm×15nm area near the local maximum; if the local maximum is greater than the sum of the average and three times the variance, extracting the fluorescence peak position corresponding to the local maximum, wherein the fluorescence peak position is represented by the excitation wavelength and emission wavelength corresponding to the fluorescence peak.

4. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The classification model constructed based on machine learning adopts support vector machine.

5. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The method of removing the fluorescence peak whose distance from the fluorescence peak position of the target benzene series pollutant spectrum is less than a threshold value adopts the Euclidean distance formula.

6. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The collected target benzene pollutant pure water background sample solution contains multiple concentrations. After the fluorescence peak position information of the three-dimensional fluorescence spectrum of the sample to be tested is extracted, the fluorescence peak whose distance from the fluorescence peak position of the target benzene pollutant spectrum at any concentration is less than a threshold is removed.

7. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The excitation wavelength of the interfering substance spectrum peak is used as the boundary, and the following Doppler broadening formula is used when the excitation wavelength is lower than the boundary: ； When the excitation wavelength is above the limit, the following collision broadening formula is used: ； in, is the fluorescence intensity, is the excitation spectrum peak corresponding to the emission wavelength, is the peak excitation wavelength, is the peak emission wavelength, is the spectral half-line width, is the independent variable excitation wavelength, is the independent variable emission wavelength.

8. The three-dimensional fluorescence spectrum benzene identification method based on line fitting combined with total variation denoising according to claim 1 is characterized in that: The total variation denoising method is based on removing the three-dimensional fluorescence spectrum of the interferent from the three-dimensional fluorescence spectrum of the sample to be tested to obtain a characteristic spectrum, which is specifically: The following total variation model is adopted and Bregman iteration is used for denoising: ； ； in, is the energy functional, is a non-negative parameter, The characteristic spectrum obtained for the target is is the spectrum of the external interferent, is the spectrum measured directly by the instrument, is the regularization term used to maintain the smoothness of the spectrum; The functional to be minimized at this time is: 。

Images