A method for correcting laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics

Abstract

The present application discloses a method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics. It includes three steps: "simulation analysis-phantom verification-actual sample verification". The interference law and pattern of fluorescence are determined by simulation analysis, and the parameters required for subsequent fluorescence correction are determined; the optimal machine learning model is determined by phantom experiments; the optimal machine learning model is trained by actual vegetable oil samples, and the content of aflatoxin B1 in the vegetable oil to be tested is predicted using the trained model, which verifies the feasibility of the present invention. The fluorescence correction method proposed in the present invention can effectively eliminate the interference of the complex matrix of the sample on the laser-induced fluorescence signal of AFB1 in vegetable oil, and can realize the accurate prediction of the AFB1 content in different types of vegetable oils. Modeling only requires one or more absorption and scattering characteristics under dual wavelengths of excitation and emission, and no sample pre-treatment is required, with high detection accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of optical property detection of liquid food, and in particular to a method for correcting laser-induced fluorescence signals of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics. Background Art

[0002] Vegetable oil plays an irreplaceable role in food production. From the growth of crops to the final refined oil, every link may affect the quality and safety of edible vegetable oil, among which mycotoxins are the main source of harm. Studies have shown that aflatoxin B1 (AFB1) is the most toxic of all aflatoxins and the one with the highest content in peanut oil, accounting for about 65%-80% of the total amount of aflatoxins. AFB1 is the most carcinogenic natural substance discovered so far (about 10 times that of cyanide and 68 times that of arsenic). In 1993, the World Health Organization's International Agency for Research on Cancer classified AFB1 as a Class I carcinogen. It mainly has strong carcinogenic, teratogenic and mutagenic destructive effects on human and animal liver tissues. In recent years, food contaminated with aflatoxin has become common. Therefore, exploring aflatoxin detection methods and improving detection capabilities have become a hot topic in research in various countries.

[0003] The principle of laser induced fluorescence (LIF) technology for detecting AFB1 is that AFB1 will emit blue (450 nm) fluorescence under UV excitation (340-400 nm). This method uses laser as the excitation light source, which improves the detection sensitivity. It is based on the characteristic fluorescence peak of AFB1 to judge the contaminated sample, which can realize online and rapid non-destructive detection, and is a direct measurement method. However, the fluorescence signal obtained by LIF is interfered by many factors, such as the environment (temperature, humidity, etc.), instruments (laser power, detector accuracy, etc.), sample-related factors (physical and chemical indicators), etc. Therefore, in existing research, when establishing a discrimination model, it is targeted at a single variety of samples in a certain batch, and samples of the same size and appearance are selected as much as possible, which limits the widespread application of LIF technology.

[0004] The chemical composition of vegetable oil is triglyceride. Vegetable oils produced from different raw materials (peanuts, corn, rapeseed, etc.) and processes (cold or hot pressing, extraction solvents, alkali washing, etc.) have different molecular size distributions, which will affect the light scattering of vegetable oils. Fatty acids and pigments in vegetable oils (natural pigments and brown, red and yellow objects produced by oxidation, degradation or polymerization of proteins, sugars, phenols and vitamins) will affect light absorption. In addition, normal vegetable oils contain some components that can produce fluorescent signals, such as vitamin B2, carotenoids, chlorophyll and its derivatives, and oxidation products of fatty acids. Absorption coefficient (μ a ) is used to represent the absorption intensity of the sample, while the reduced scattering coefficient (μ' s) is used to represent the light scattering of the sample. Absorption is related to the chemical composition of the sample, and scattering is related to the physical properties of the sample. By analyzing the interference mechanism of absorption and scattering on the fluorescence signal and studying fluorescence correction methods or multiple optical feature fusion methods, it is expected to eliminate the influence of the sample matrix on the LIF measurement results. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting aflatoxin B1 in different types of vegetable oils based on optical characteristic parameters of very few wavelengths, so as to solve the problem that the existing laser induced fluorescence spectroscopy method is seriously interfered by the sample matrix and requires many wavelengths for modeling. The method is simple to operate, rapid and non-destructive, does not require sample pretreatment, is environmentally friendly, and has high detection accuracy.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] The method of the present invention comprises three steps: "simulation analysis-pseudo-verification-actual sample verification".

[0008] A method for correcting laser-induced fluorescence signals of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics, comprising:

[0009] By analyzing the interference rules and patterns of fluorescence through simulation, the parameters required for fluorescence correction are determined, wherein the parameters required for fluorescence correction are one or more absorption and scattering characteristics under the excitation and emission wavelengths;

[0010] According to the parameters required for fluorescence correction, the optimal machine learning model for correcting the laser-induced fluorescence signal of aflatoxin B1 was determined through phantom experiments;

[0011] Different types of vegetable oils were collected and sample sets with different concentrations of aflatoxin B1 contamination were prepared. The sample sets were used to train the optimal machine learning model, and the trained model was used to predict the content of aflatoxin B1 in the tested vegetable oils.

[0012] As a preferred embodiment of the present invention, the simulation analysis process includes:

[0013] 1.1) The simulation is implemented by Monte Carlo (SL Jacques, Monte Carlo simulations of fluorescence in turbid media, Ch. 6 in Handbook of Biomedical Fluorescence, MA Mycek, BW Pogue, publ. Marcel-Dekker, New York, NY, 2003) algorithm, and the fluorescence escape flux density at the excitation and emission wavelengths is obtained by inputting the absorption coefficient, scattering coefficient, anisotropy coefficient, and refractive index of the simulation model sample and the medium outside the sample at the excitation and emission wavelengths;

[0014] 1.2) Calculate the fluorescence flux Φ escaping from the sample surface at the emission wavelength according to the following formula:

[0015]

[0016] Among them, r i is the distance between the i-th point on the simulation model sample and the center of the incident collimated beam, J(r i ) _em is the escape fluorescence flux density at the i-th point under the emission wavelength, i max is the total number of points on the simulation model sample;

[0017] 1.3) Calculate the effective attenuation coefficient at the excitation and emission wavelengths:

[0018]

[0019]

[0020] In the formula, μ eff_ex is the effective attenuation coefficient at the excitation wavelength, μ a_ex is the absorption coefficient at the excitation wavelength, μ′ s_ex is the reduced scattering coefficient at the excitation wavelength, μ eff_em is the effective attenuation coefficient at the emission wavelength, μ a_em is the absorption coefficient at the emission wavelength, μ′ s_em is the reduced scattering coefficient at the emission wavelength;

[0021] 1.4) The absorption coefficient, reduced scattering coefficient and effective attenuation coefficient at the excitation and emission wavelengths are taken as a set of candidate absorption and scattering features, and two or more candidate absorption and scattering features are selected from the set. According to the inhibition or promotion relationship between the candidate absorption and scattering features and the escape fluorescence flux obtained by simulation, a fitting curve is established:

[0022] Φ=A+a1e H1 / t1+a2e H2 / t2 +…+ake Hk / tk

[0023] Wherein, Hk represents the kth candidate absorption scattering feature, k represents the number of candidate absorption scattering features selected from the set, A, a1, a2, ak, t1, t2, tk are fitting parameters, when Hi,i=1,2,…,k has a promoting effect on the escape fluorescence flux, then Hi takes a negative value, when Hi,i=1,2,…,k has an inhibitory effect on the escape fluorescence flux, then Hi takes a positive value;

[0024] Different candidate absorption and scattering feature combinations are selected from the set for fitting, the optimal combination is selected according to the fitting results, and the absorption and scattering features in the optimal combination are used as the parameters required for fluorescence correction.

[0025] As a preferred embodiment of the present invention, the number of absorption and scattering features in the candidate absorption and scattering feature combination is 2-4.

[0026] As a preferred embodiment of the present invention, the parameters required for fluorescence correction are the effective attenuation coefficient at the excitation and emission wavelengths, and the reduced scattering coefficient at the excitation wavelength.

[0027] As a preferred embodiment of the present invention, the phantom experiment comprises:

[0028] A self-made liquid phantom was prepared, with India Ink as the absorbing material, TiO2 as the scattering particles, quinine sulfate as the fluorescent standard substance, and deionized water as the solvent;

[0029] Testing liquid phantoms with different volume concentrations of absorbing materials, scattering particles, and quinine sulfate to obtain the parameters required for fluorescence correction and fluorescence intensity of the liquid phantoms;

[0030] Different machine learning models were established, the parameters required for fluorescence correction of the liquid mimetic and the fluorescence intensity were used as the input values ​​of the machine learning model, the quinine sulfate content in the liquid mimetic was used as the predicted value, different machine learning models were trained and verified, and the model with the best effect was taken as the optimal machine learning model.

[0031] As a preferred embodiment of the present invention, the machine learning model includes artificial neural network ANN, random forest RF, and least squares support vector machine regression LSSVR model.

[0032] As a preferred embodiment of the present invention, when using the trained model to predict the content of aflatoxin B1 in the vegetable oil to be tested, the parameters required for fluorescence correction and the fluorescence intensity of the vegetable oil to be tested are used as model inputs to output the predicted result of the aflatoxin B1 content.

[0033] The beneficial effects of the present invention are as follows: the present invention provides a method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics, which solves the problem that the existing laser-induced fluorescence spectroscopy technology is seriously interfered by the complex matrix of the sample when detecting the AFB1 content of different varieties of vegetable oil. The method determines that only based on the absorption and scattering characteristics at the dual wavelengths of excitation and emission, and the fluorescence characteristics at the emission wavelength, the AFB1 in different varieties of vegetable oil can be robustly predicted, with high detection accuracy, no sample pre-treatment, and few parameters required for modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0035] Figure 1 It is a flow chart of the main program of fluorescence MC simulation according to an exemplary embodiment.

[0036] Figure 2 is the excitation light and fluorescence escape flux density obtained by MC simulation according to an exemplary embodiment.

[0037] Figure 3 is the calculated escape fluorescence flux with μ according to an exemplary embodiment. a_ex ,μ a_em , μ' s_ex ,μ' s_em Law of change.

[0038] Figure 4 Graph 1 is a diagram showing the fitted value vs. simulated value of the escape fluorescence flux according to an exemplary embodiment.

[0039] Figure 5 According to an exemplary embodiment, eff_ex , μ eff_em , μ' s_ex LSSVR model predicted values ​​of AFB1 content in vegetable oils versus reference values ​​of fluorescence intensity at emission wavelength.

[0040] Figure 6 FIG. 5 is an uncorrected fluorescence intensity of plant oils with different AFB1 contents at 450 nm according to an exemplary embodiment.

[0041] Figure 7 PLSR predicted value vs. reference value of AFB1 content in vegetable oil based on fluorescence intensity at 400-600 nm before fluorescence correction according to an exemplary embodiment.

[0042] Figure 8It is a flow chart showing a method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to an exemplary embodiment. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0044] Figure 8 The main process of a method proposed by the present invention for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics is shown, including three steps of "simulation analysis-pseudo-verification-actual sample verification".

[0045] like Figure 1 As shown, the main program workflow of the fluorescence MC simulation algorithm is:

[0046] Emitting photons: In this embodiment, the light source type used is a collimated beam, which is consistent with the laser light source used in subsequent phantom experiments;

[0047] Photon single movement: by generating a random number and the total attenuation coefficient of the input excitation light (denoted as μ t_ex , is the absorption coefficient μ a_ex and the scattering coefficient μ s_ex The sum of the photons and the initial positions of the photons are used to further determine the positions of the photons after they move.

[0048] Escape judgment link: judge whether the photon escapes from the surface of the simulation model. If it escapes, the photon is judged as "disappeared", and its photon weight is added to the excitation light escape flux density (J(r) _ex ) and continue to emit photons; if the photon does not escape, the photon enters the "photon specific gravity decrease" stage, which is as follows: after the photon moves and does not escape, according to the albedo of the simulation sample (a = μ s_ex / μ t_ex ), its specific gravity will decrease, and the decreased specific gravity of photons is added to the photon energy flux rate (F) inside the sample;

[0049] Photon rotation: When the photon does not disappear and its proportion in the Drop link decreases, the transmission path will also change. By generating random numbers, the Henyey-Green-stein scattering function is used to determine the angle of its path deviation, and the photon is rotated according to the angle value;

[0050] Roulette judgment stage: If the photon's weight decreases after the "photon weight decreases" and "photon rotation" stages, the weight of the photon after the decrease is higher than or equal to the set threshold (1e -4 ), the photon escapes, and its photon weight is added to the excitation light escape flux density (J(r) _ex), if the proportion after the decrease is lower than the set threshold (1e -4 ) is generated, a random number and a chance value (Chance, 0.1) are generated to determine whether the photon "disappears". If the random number is less than 0.1, the photon disappears. If the random number is greater than or equal to 0.1, the photon's weight is reassigned and the "photon single movement" phase is entered.

[0051] The photon that is judged as "disappeared" in the roulette life and death link needs to be further judged whether the photon is the last photon of the simulation. If not, continue to emit photons; if so, end the cycle, and the calculated excitation light escape flux density (J(r) _ex ) and the excitation light flux (F _ex ).

[0052] In this embodiment, the fluorescence escape flux density (J(r) _em ) and fluorescence energy flux (F _em ) is obtained in the same way, except that the fluorescence simulation subroutine assumes that the fluorescence is produced by uniform fluorescent substances inside the simulated sample. Therefore, when photons are emitted, isotropic fluorescent photons are emitted at a certain position inside the sample.

[0053] By inputting the absorption coefficient (μ a_ex ,μ a_em ), scattering coefficient (μ s_ex ,μ s_em ), anisotropy coefficient (g), refractive index of the simulated sample and the medium outside the sample (n) to obtain Figure 2 The escape flux density (J(r)) of the excitation light and emitted fluorescence is shown, Figure 2 The horizontal axis r in the equation represents the distance between the model sample and the incident light position. The g value is set to 0.9, and the n value is set to 1.33 and 1 respectively. a_ex , μ a_em ,μ s_ex ,μ s_em As shown in the following table:

[0054]

[0055]

[0056] According to the following formula, different μ a_ex ,μ a_em ,μ s_ex ,μ s_em The fluorescence flux (Φ) escaping from the sample surface under the input conditions.

[0057]

[0058] Among them, r i is the distance between the i-th point on the simulation model sample and the center of the incident collimated beam, J(r i ) _em is the escape fluorescence flux density at the i-th point under the emission wavelength, i max is the total number of points on the simulation model sample.

[0059] The calculated escape fluorescence flux Φ varies with μ a_ex , μ a_em , μ s_ex , μ s_em The changing rules are as follows Figure 3 As shown in the figure, the upper left figure is the results of simulation samples 1-7, the upper right figure is the results of simulation samples 8-14, the lower left figure is the results of simulation samples 15-21, and the lower right figure is the results of simulation samples 22-28. From the simulation results, it can be found that the escape fluorescence flux is related to μ a_ex , μ a_em , μ s_em Both show a nonlinear decreasing relationship, and the reduced scattering coefficient μ′ at the excitation wavelength s_ex It promotes the escape fluorescence flux.

[0060] The effective attenuation coefficient (μeff) at the excitation and emission wavelengths is calculated according to the following formula:

[0061]

[0062] In the formula, μ a is the absorption coefficient, μ′ s is the reduced scattering coefficient. Substitute the absorption coefficient and reduced scattering coefficient at the excitation and emission wavelengths into the above formula to obtain the effective attenuation coefficient μ at the excitation wavelength. eff_ex and the effective attenuation coefficient μ at the emission wavelength eff_em The escape fluorescence flux obtained by MC simulation is fitted by the following formula, and the fitting value is similar to the simulation value as shown in Figure 4 shown.

[0063]

[0064] Among them, A, a1, a2, a3, t1, t2, t3 are fitting parameters, μ eff_ex With μ eff_em is the effective attenuation coefficient at the excitation and emission wavelengths.

[0065] Depend on Figure 4 It can be found that the effective attenuation coefficient μ at the excitation wavelength eff_ex , reduced scattering coefficient μ′ s_ex The effective attenuation coefficient μ at the emission wavelength eff_em, the fluorescence signal can be better fitted, so it is determined that these three coefficients will be used to correct the fluorescence signal in the future.

[0066] A self-made liquid phantom was made, with India Ink as the absorption material, TiO2 as the scattering particles, quinine sulfate as the fluorescent standard substance, and deionized water as the solvent. The volume concentrations of India Ink were 0, 0.00375%, 0.0075%, 0.015%, and 0.03%, respectively; the volume concentrations of TiO2 were 0.001875%, 0.00375%, 0.0075%, 0.015%, and 0.03%, respectively; the volume concentrations of quinine sulfate were 0.5%, 0.75%, 1%, 1.25%, and 1.5%, respectively. A total of 135 phantom samples were configured, and the phantom samples were divided into 90 samples in the calibration set and 45 samples in the verification set. The phantom was tested using a self-developed optical platform (patent application number CN202011238253.4) to obtain the phantom μ a_ex , μ a_em , μ s_ex , μ s_em And fluorescence intensity. Further calculation of the phantom μ eff_ex With μ eff_em , and the μ of the phantom eff_ex , μ eff_em , μ s_ex The fluorescence intensity was brought into different machine learning models (ANN, RF, LSSVR) to predict the content of quinine sulfate in the phantom. The prediction effect of different models was analyzed by the correlation coefficient (R c , R v ) and root mean square error (RMSEC, RMSEV) are used for evaluation. The model effect is shown in the following table:

[0067] Modeling approach <![CDATA[R c ]]> <![CDATA[R v ]]> RMSEC RMSEV ANN 0.950 0.515 0.120 0.464 RF 0.939 0.555 0.181 0.347 LSSVR 0.806 0.828 0.227 0.245

[0068] From the modeling results, it can be found that after multiple parameter adjustments, ANN and RF still have difficulty in achieving accurate prediction of the prediction set. In comparison, LSSVR modeling results are optimal, and the correlation coefficient and root mean square error of the calibration set and the validation set are close, with correlation coefficients of 0.806 and 0.828, and root mean square errors of 0.227 and 0.245, respectively. Therefore, in the subsequent prediction of AFB1 in vegetable oil samples, LSSVR will be used to establish the AFB1 prediction model based on the effective attenuation coefficient at the excitation wavelength, the reduced scattering coefficient, and the effective attenuation coefficient at the emission wavelength.

[0069] Prepare different types and brands of vegetable oils, add different volumes of AFB1 standard solution with acetonitrile as solvent, and obtain 1620 vegetable oil samples with different contamination levels (0, 10, 20, 40 μg / kg). The vegetable oil samples are divided into 1080 calibration sets and 540 validation sets. The types of vegetable oils are: peanut oil, rapeseed oil, corn oil, and two brands of each oil. The vegetable oil samples are tested using the self-developed optical platform to obtain the μ a_ex , μ a_em , μ s_ex , μ s_em And fluorescence intensity. Further calculation of μ eff_ex With μ eff_em , and the μ of vegetable oil eff_ex , μ eff_em , μ s_ex The fluorescence intensity was brought into the LSSVR model to predict the AFB1 content in vegetable oil samples. eff_ex , μ eff_em , μ s_ex The prediction results of the LSSVR model for the AFB1 content in vegetable oil based on the fluorescence intensity at the emission wavelength are as follows: Figure 5 As shown, R c , R v Both can reach 1.000, and RMSEC and RMSEV are as low as 0.038 and 0.099.

[0070] In order to reflect the advantages of the method proposed in the present invention for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics, the fluorescence intensity of the uncorrected vegetable oil at a wavelength of 450nm (characteristic fluorescence emission wavelength of AFB1) is plotted as a scatter plot as shown in FIG. Figure 6 As shown, it can be found that for the same AFB1 concentration pollution, the fluorescence intensity of different types and brands of vegetable oils at 450nm is very different, and it is difficult to predict AFB1 by the fluorescence intensity at a single wavelength.

[0071] Further, based on the fluorescence intensity of 400-600nm, partial least squares regression (PLSR) was used to establish a prediction model for the AFB1 content of different types and brands of vegetable oils. The model prediction results are as follows: Figure 7 As shown in the figure, it can be found that the quantitative prediction of AFB1 in different types and brands of vegetable oils can be achieved through the fluorescence spectrum of 400-600nm. However, compared with the method proposed in the present invention, the model accuracy is lower, R c , R v They are only 0.925 and 0.923 respectively, and RMSEC and RMSEV are as high as 6.385 and 6.470.

[0072] The present invention can accurately predict the AFB1 content in different types of vegetable oils based only on the absorption, scattering and fluorescence characteristics under dual wavelengths of excitation and emission. It does not require sample pretreatment, is environmentally friendly, and has a fast detection speed. Compared with wide-band spectral detection, less wavelength information is required.

[0073] The above examples are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many variations are possible. All variations that can be directly derived or associated with the contents disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics, characterized in that: include: The interference rules and patterns of fluorescence are analyzed by Monte Carlo algorithm simulation to determine the parameters required for fluorescence correction, wherein the parameters required for fluorescence correction are one or more absorption and scattering characteristics at the excitation and emission wavelengths, including the effective attenuation coefficient at the excitation and emission wavelengths, and the reduced scattering coefficient at the excitation wavelength; According to the parameters required for fluorescence correction, the optimal machine learning model for correcting the laser-induced fluorescence signal of aflatoxin B1 was determined through phantom experiments; the liquid phantom used in the phantom experiment used India Ink as the absorption material, TiO2 as the scattering particle, quinine sulfate as the fluorescence standard substance, and deionized water as the solvent; Different types of vegetable oils were collected and sample sets with different concentrations of aflatoxin B1 contamination were prepared. The sample sets were used to train the optimal machine learning model, and the trained model was used to predict the content of aflatoxin B1 in the tested vegetable oils.

2. The method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to claim 1, characterized in that: The simulation analysis process includes: 1.1) Obtain the fluorescence escape flux density at the excitation and emission wavelengths by inputting the absorption coefficient, scattering coefficient, anisotropy coefficient, and refractive index of the simulation model sample and the medium outside the sample; 1.2) Calculate the fluorescence flux Φ escaping from the sample surface at the emission wavelength according to the following formula: Among them, r i is the distance between the i-th point on the simulation model sample and the center of the incident collimated beam, J(r i ) _em is the escape fluorescence flux density at the i-th point under the emission wavelength, i max is the total number of points on the simulation model sample; 1.3) Calculate the effective attenuation coefficient at the excitation and emission wavelengths: In the formula, μ eff_ex is the effective attenuation coefficient at the excitation wavelength, μ a_ex is the absorption coefficient at the excitation wavelength, μ′ s_ex is the reduced scattering coefficient at the excitation wavelength, μ eff_em is the effective attenuation coefficient at the emission wavelength, μ a_em is the absorption coefficient at the emission wavelength, μ′ s_em is the reduced scattering coefficient at the emission wavelength; 1.4) The absorption coefficient, reduced scattering coefficient and effective attenuation coefficient at the excitation and emission wavelengths are taken as a set of candidate absorption and scattering features, and two or more candidate absorption and scattering features are selected from the set. According to the inhibition or promotion relationship between the candidate absorption and scattering features and the escape fluorescence flux obtained by simulation, a fitting curve is established: Φ=A+a1e H1 / t1 +a2e H2 / t2 +…+ake Hk / tk Wherein, Hk represents the kth candidate absorption scattering feature, k represents the number of candidate absorption scattering features selected from the set, A, a1, a2, ak, t1, t2, tk are fitting parameters, when Hi,i=1,2,…,k has a promoting effect on the escape fluorescence flux, then Hi takes a negative value, when Hi,i=1,2,…,k has an inhibitory effect on the escape fluorescence flux, then Hi takes a positive value; Different candidate absorption and scattering feature combinations are selected from the set for fitting, the optimal combination is selected according to the fitting results, and the absorption and scattering features in the optimal combination are used as the parameters required for fluorescence correction.

3. The method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to claim 2, characterized in that: The number of absorption and scattering features in the candidate absorption and scattering feature combination is 2-4.

4. The method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to claim 1, characterized in that: The phantom experiment includes: A self-made liquid phantom was prepared, with India Ink as the absorbing material, TiO2 as the scattering particles, quinine sulfate as the fluorescent standard substance, and deionized water as the solvent; Testing liquid phantoms with different volume concentrations of absorbing materials, scattering particles, and quinine sulfate to obtain the parameters required for fluorescence correction and fluorescence intensity of the liquid phantoms; Different machine learning models were established, the parameters required for fluorescence correction of the liquid mimetic and the fluorescence intensity were used as the input values ​​of the machine learning model, the quinine sulfate content in the liquid mimetic was used as the predicted value, different machine learning models were trained and verified, and the model with the best effect was taken as the optimal machine learning model.

5. The method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to claim 1, characterized in that: The machine learning models include artificial neural network ANN, random forest RF, and least squares support vector machine regression LSSVR models.

6. The method for correcting the laser-induced fluorescence signal of aflatoxin B1 in vegetable oil based on dual-wavelength absorption and scattering characteristics according to claim 1, characterized in that: When using the trained model to predict the content of aflatoxin B1 in the vegetable oil to be tested, the parameters required for fluorescence correction and the fluorescence intensity of the vegetable oil to be tested are used as model inputs, and the predicted result of the aflatoxin B1 content is output.

Images