Method for detecting polluted vegetable oil through deep learning assisted three-dimensional fluorescence spectrum

Abstract

The invention belongs to the technical field of food safety detection, and particularly relates to a method for detecting polluted vegetable oil by using a deep learning assisted three-dimensional fluorescence spectrum. According to the method, firstly, the optimal adsorption condition of magnesium silicate to the vegetable oil is predicted based on a linear regression model and an L-BFGS-B algorithm, and targeted separation of mineral oil in the polluted vegetable oil is achieved through adsorption of magnesium silicate to the vegetable oil under the optimal condition. Then, on the basis of mineral oil pollutant samples obtained through separation, characteristic fingerprint spectrums of different mineral oil types are extracted from the collected three-dimensional fluorescence spectrums by applying multiple pre-training network models, and therefore several common mineral oil pollutant types are distinguished. And decomposing and judging fluorescent components of corresponding pollutants in the mineral oil by combining a parallel factor data dimension reduction algorithm. And finally, building a support vector regression model of the corresponding component based on the corresponding relationship between the fluorescence signal and the concentration of each pollution component, thereby realizing the quantitative detection of each pollution component in the mineral oil.

Description

Technical Field

[0001] The present invention belongs to the technical field of food safety detection, and specifically relates to a method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy. Background Art

[0002] Edible vegetable oils are contaminated by mineral oils, which seriously violates food safety and endangers public health. Usually, gas chromatography-mass spectrometry (GC-MS) is the first choice for the analysis of mineral oil contaminants in vegetable oils. This method has high sensitivity and accuracy in the laboratory and can effectively identify and quantify mineral oil contaminants and their components. However, vegetable oil itself is a complex matrix, containing a large amount of fatty acids and other organic compounds, which are similar to the composition of mineral oil and are prone to matrix effects, affecting the accuracy of GC-MS test results. In addition, the operation process of this method is complicated and time-consuming, which hinders the food safety testing process that requires a rapid response. Therefore, it is extremely important to develop an efficient detection method.

[0003] Highly alkylated aromatic compounds in mineral oil have fluorescence response due to the large π bond in their molecular structure. Excitation-emission matrix fluorescence (EEMF) technology has been increasingly used in food analysis due to its high sensitivity and the advantages of obtaining rich information about the fluorescence excitation wavelength, emission wavelength and fluorescence intensity of fluorescent compounds in a short time. However, the collected fluorescence data are numerous and complex, and simple linear regression, multivariate statistical analysis and other methods cannot fully explore and effectively utilize these fluorescence information, so a more outstanding high-dimensional data processing and complex pattern recognition method is needed.

[0004] The rapid development of artificial intelligence (AI) will reshape almost every field of work. As a subfield of artificial intelligence, deep learning has powerful information mining capabilities, and its application research in the field of food safety testing has emerged in an endless stream in recent years. Xia et al. achieved a classification accuracy of 97.5% for different types of pesticide residues on the surface of vegetables and fruits based on three-dimensional fluorescence spectroscopy combined with convolutional neural networks (CNN). Zhang's research team successfully achieved the classification of heavy metals such as lead, cadmium, and arsenic in different food samples based on the same method, with a classification accuracy of up to 98.2%. On the other hand, by predicting the optimal conditions for magnesium silicate to adsorb vegetable oil through machine learning, targeted separation of mineral oil can be achieved, which largely eliminates the matrix effect and collects more effective spectral data, thereby improving the accuracy of classification and quantification. At present, deep learning has achieved good results in studying classification problems. However, for consumers, they are more concerned about which components in mineral oil pollutants will cause harm to the human body and whether the content of these components is below the safety threshold. Therefore, how to analyze the harmful components and their content in each mineral oil pollutant is extremely important.

[0005] Parallel factor analysis (PARAFAC) is a multivariate data analysis technique that can decompose fluorescence signals into independent individual fluorescence phenomena. This technique can track the fluorescent components of different parts of mineral oil and separate fluorescence signals with specific excitation and emission spectra. However, the relationship between the fluorescence intensity and concentration obtained by analysis is not always linearly related due to factors such as self-absorption and fluorescence quenching. Support Vector Regression (SVR) is a regression model extended from Support Vector Machine (SVM). Based on the kernel function, it can handle a wide range of nonlinear relationships well and has been frequently used by researchers in nonlinear relationship environments.

[0006] Although the mineral oil contamination problem of vegetable oil has gradually attracted attention in the field of food safety, existing research mainly focuses on the source and total amount of mineral oil detection. After searching, no relevant research and reports on the classification of mineral oil in contaminated vegetable oil and the qualitative and quantitative analysis of its contamination components have been found.

[0007] In view of this, the inventors hope to provide a deep learning-assisted three-dimensional fluorescence spectroscopy detection method for contaminated vegetable oil, which can be used to quickly identify the contamination category of mineral oil in contaminated vegetable oil (gasoline, diesel, industrial boiler oil, engine oil, double neutral industrial boiler and aviation kerosene), and simultaneously determine the harmful components and their content in each mineral oil pollutant. Summary of the invention

[0008] The purpose of the present invention is to overcome the above-mentioned problems existing in traditional technologies and to provide a method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy.

[0009] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0010] The present invention provides a method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy, comprising the following steps:

[0011] Step 1: Prediction of the optimal adsorption conditions of magnesium silicate for different vegetable oils;

[0012] Step 2: Sample preparation and data preprocessing;

[0013] Step 3: Identify the types of mineral oil contaminants;

[0014] Step 4: Qualitative and quantitative analysis of mineral oil pollution components.

[0015] Furthermore, in step one, the adsorption data of magnesium silicate on different vegetable oils under different adsorption conditions are used, the adsorption amount is used as the dependent variable, and the different adsorption conditions are used as independent variables, a linear regression model is established, the regression equation is fitted and the determination coefficient of the model is calculated; then the optimization algorithm is used to limit the experimental conditions, and through the optimization process, the optimal conditions that meet the adsorption target are finally obtained.

[0016] Furthermore, the regression equation is in the form of:

[0017] y=β0+β1x1+β2x2+β3x3+β4x4

[0018] Among them, y is the adsorption amount, x1, x2, x3, x4 represent the amount of magnesium silicate, time, water content and temperature, β0, β1, β2, β3, β4 are regression coefficients;

[0019] In order to find the best experimental conditions, the L-BFGS-B optimization algorithm was used to solve the regression equation, set the adsorption target, and minimize the deviation between the predicted adsorption and the target value;

[0020] The range of optimized variables is limited to the actual experimental conditions, and through the optimization process, the optimal conditions that meet the adsorption target are gradually approached;

[0021] After the optimization was completed, the optimal experimental conditions for each vegetable oil included four variables: magnesium silicate dosage, time, water content and temperature.

[0022] Furthermore, in step 2, mineral oils of different concentration gradients are added to different vegetable oils, and then a certain amount of n-hexane is added to each mixture to achieve effective adsorption of vegetable oil under the predicted optimal adsorption conditions of magnesium silicate for different vegetable oils. After the adsorption is completed, the magnesium silicate is filtered to obtain the corresponding mineral oil test solution; the test solution is transferred to a quartz cuvette, and a fluorescence spectrophotometer is used to collect spectral data; first, the background signal of the solvent is removed from the collected fluorescence spectral data to eliminate the interference of the solvent on the spectral data; based on the characteristic that the excitation light wavelength and the emission light wavelength are close and prone to scattering, the specific wavelength region clipping method is used to remove the scattering noise to obtain a higher signal-to-noise ratio, and then the interpolation method is used to supplement the missing data to ensure the integrity of the data; the mineral oil sample data of the same type and concentration are averaged to improve the stability and representativeness of the data, and the generalization ability of the model is enhanced through data enhancement technology.

[0023] Furthermore, in step three, based on the separated mineral oil contaminant samples, a variety of pre-trained network models are used to extract characteristic fingerprints of different mineral oil categories from the collected three-dimensional fluorescence spectra. Through hyperparameter optimization and comparison of the performance of each model, the optimal model is finally selected to identify several common mineral oil contaminant categories.

[0024] Furthermore, a variety of pre-trained network models including artificial neural networks (ANN), convolutional neural networks (CNN), and recurrent neural networks (RNN) were used to extract characteristic fingerprints of different mineral oil categories from the collected three-dimensional fluorescence spectra. The hyperparameters including learning rate, batch size, number of training rounds, optimizer type, and activation function were optimized through grid search, random search, and Bayesian optimization techniques. The performance of each model was compared, and finally the optimal model was selected to identify several common types of mineral oil pollutants.

[0025] Furthermore, in step four, the parallel factor data dimensionality reduction algorithm is combined to decompose and determine the fluorescent components of the corresponding pollutants in the mineral oil; based on the corresponding relationship between the fluorescence signal and concentration of each pollutant component, a support vector regression model of the corresponding component is built to achieve quantitative detection of various pollutants in the mineral oil.

[0026] Furthermore, parallel factor analysis was used to detect outliers in the preprocessed three-dimensional fluorescence spectra, and the optimal component number of the relevant fluorescent components was screened and verified through core consistency, variance explanation rate, factor matching score and split-half analysis method; based on the correspondence between the excitation and emission wavelengths of each fluorescent component obtained by parallel factor analysis and each standard substance of mineral oil, the category of the main mixed pollution components in the mineral oil was qualitatively determined; further, based on the correspondence between the fluorescence signal and concentration of each pollution component, a support vector regression model of the corresponding component was established, thereby realizing the quantitative detection of various pollution components in mineral oil.

[0027] The beneficial effects of the present invention are:

[0028] The present invention provides a method for detecting contaminated vegetable oil by deep learning-assisted three-dimensional fluorescence spectroscopy, which is used to quickly identify the type of mineral oil in the contaminated vegetable oil and its contaminating components qualitatively and quantitatively. The optimal adsorption conditions of magnesium silicate on vegetable oil are predicted by machine learning algorithms, and the targeted separation of mineral oil in the contaminated vegetable oil is achieved through the adsorption of magnesium silicate on vegetable oil under optimal conditions, which can effectively avoid the complex matrix effect in the vegetable oil and mineral oil environment. The characteristic fingerprints of different mineral oil categories are extracted from the three-dimensional fluorescence spectrum using a deep learning pre-training network model, which can achieve accurate identification of the categories of mineral oil pollutants in vegetable oil. Combined with the parallel factor data dimensionality reduction algorithm, the fluorescent components of the corresponding pollutants in the mineral oil are decomposed and determined, and the support vector regression model of the corresponding components is built based on the corresponding relationship between the fluorescence signal and concentration of each contaminating component, which can achieve quantitative detection of various contaminating components in the mineral oil. This method helps to promote the field of food safety detection from experience dependence to data-driven intelligent and automated detection. It is expected to achieve early warning and precise control of mineral oil pollution in food production, transportation and detection, and provide strong protection for food safety.

[0029] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 The diagram is a schematic diagram of the method of the present invention for quickly identifying the types of mineral oils in contaminated vegetable oils and the qualitative and quantitative analysis of their contaminating components;

[0032] Figure 2 Schematic diagram of the effect of different experimental conditions on the adsorption capacity of magnesium silicate;

[0033] Among them, a-magnesium silicate content, b-time, c-magnesium silicate water content, d-temperature;

[0034] Figure 3 The three-dimensional fluorescence spectra of six mineral oils;

[0035] Among them, a-95 gasoline, b-diesel, c-industrial boiler oil, d-engine oil, e-double neutral industrial boiler oil, f-aviation kerosene;

[0036] Figure 4 This is a comparison chart of the accuracy of six deep learning models;

[0037] Among them, ResNet18 is the optimal model, with a classification accuracy of 96.21% for 6 mineral oils;

[0038] Figure 5 It is the regression curve of the support vector regression model SVR established based on the standard substance;

[0039] Among them, a, b, c, and d correspond to components 1 to 4, respectively, and the corresponding standard substances are naphthalene, anthracene, phenanthrene, and fluorene, respectively. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The present invention provides a method for detecting contaminated vegetable oil by deep learning-assisted three-dimensional fluorescence spectroscopy, which is used to quickly identify the pollution category of mineral oil in contaminated vegetable oil and simultaneously determine the harmful components and their contents in each mineral oil pollutant. First, a machine learning algorithm is used to optimize the conditions for magnesium silicate to adsorb vegetable oil to achieve targeted separation of mineral oil, and the mineral oil pollutants after adsorption and separation are tested by fluorescence to obtain a three-dimensional fluorescence spectrum (EEM). Background and scattering removal technology is used, combined with interpolation method to process data, and the data is enhanced by rotation, clipping and noise injection methods. Then, six pre-trained network models are used to extract characteristic fingerprints of different mineral oil categories from the three-dimensional fluorescence spectrum, and the hyperparameters are optimized by grid search, and the performance of each model is compared. Finally, the optimal model is selected to perform qualitative analysis on different types of mineral oil pollutants. At the same time, the pre-processed three-dimensional fluorescence spectrum is detected for outliers, and the optimal component number of the fluorescent component in the three-dimensional fluorescence spectrum data is screened by core consistency, variance explanation rate and minimum factor similarity, and the component information of the corresponding pollutants in the mineral oil is analyzed based on the excitation and emission wavelength information of each component. Finally, a support vector regression model of the corresponding component was built based on the correspondence between the fluorescence signal and concentration of each pollutant component, thereby realizing the quantitative detection of various pollutants in mineral oil.

[0042] Related specific embodiments of the present invention:

[0043] Example 1

[0044] First, the adsorption amount of 300 groups of magnesium silicate on vegetable oil under different conditions was obtained experimentally. A linear regression model was established for each vegetable oil. The adsorption amount was used as the dependent variable, and the amount of magnesium silicate, time, water content, and temperature were used as independent variables. The regression equation was fitted and the determination coefficient of the model was calculated. The form of the regression equation is usually y=β0+β1x1+β2x2+β3x3+β4x4, where y is the adsorption amount, x1, x2, x3, and x4 represent the amount of magnesium silicate, time, water content, and temperature, respectively, and β0, β1, β2, β3, and β4 are regression coefficients. In order to find the best experimental conditions, the L-BFGS-B optimization algorithm was used to solve the regression equation, set the adsorption amount target to 100 mg, and minimize the deviation between the predicted adsorption amount and the target value. The optimized variable range is limited to the actual experimental conditions, with the amount of magnesium silicate being 1-6 g, the time being 10-50 minutes, the water content being 0-10%, and the temperature being 15-40°C. Through the optimization process, the optimal conditions that meet the adsorption target are gradually approached. After the optimization is completed, the optimal experimental conditions for each vegetable oil will include four variables: magnesium silicate dosage, time, water content and temperature.

[0045] Example 2

[0046] Six different mineral oils (diesel, gasoline, engine oil, aviation kerosene, industrial boiler oil, double neutral industrial boiler oil) with 28 concentration gradients were added to six different vegetable oils (peanut oil, soybean oil, rapeseed oil, corn oil, sesame oil, olive oil), and 10 mL of n-hexane was added to each mixture. Then, magnesium silicate under the optimal adsorption conditions was added, and the mineral oil components in the vegetable oil were adsorbed by magnesium silicate, and the mixture was allowed to stand for stratification. The supernatant after standing was filtered through a glass funnel padded with glass wool, which was pre-soaked with n-hexane and dried. The filtrate was finally collected in a colorimetric tube for subsequent measurement. In this process, a total of 1008 standard samples containing six mineral oil pollution components with 28 concentration gradients were obtained. These samples can simulate actual pollution conditions and construct standard samples with known pollution components, providing a basis for subsequent data acquisition and analysis.

[0047] Example 3

[0048] The obtained samples were first subtracted from the interference of blank solvents, and then the noise in the spectral data was removed by cutting the first and second order Rayleigh scattering to ensure that the measured spectral data was more accurate. Next, the blank values ​​were filled in using cubic spline interpolation to ensure the integrity of the data. In this process, samples belonging to the same mineral oil and the same concentration gradient were added and averaged to reduce experimental errors and improve the stability and representativeness of the data. After this step, the number of data was reduced from the original 1008 samples to 168 new samples (data after averaging), laying the foundation for subsequent analysis. Subsequently, data enhancement technology was applied to further increase the number of samples to 672. The enhanced samples provide a richer sample size for model training, improving the generalization ability and prediction accuracy of the model.

[0049] Example 4

[0050] In the classification process of mineral oil, transfer learning technology is used, combined with multiple deep learning models, to extract rich fingerprint information from the three-dimensional fluorescence spectrum to help achieve accurate identification of mineral oil. Specifically, six deep learning models including SimpleCNN, LeNet-5, AlexNet, GoogLeNet, VGG16 and ResNet18 are applied. The batch size and learning rate of all models are optimized through grid search technology, and the performance of different models is compared. Finally, the optimal model is selected for qualitative analysis of mineral oil.

[0051] In this process, transfer learning makes small sample learning possible, reducing the reliance on large amounts of labeled data, thereby reducing the cost and workload of data collection, while improving the generalization ability of the model, enabling it to achieve high accuracy with limited samples. Through this classification model, researchers can quickly and accurately identify the types of gasoline, diesel, engine oil, aviation kerosene, industrial boiler oil, and dual-neutral industrial boiler oil, significantly improving experimental efficiency, reducing human errors, and providing strong technical support for subsequent analysis.

[0052] Example 5

[0053] In the quantitative analysis of mineral oil components, parallel factor analysis (PARAFAC) is first used to decompose the preprocessed three-dimensional fluorescence spectrum data. The optimal number of components is determined by performing outlier detection, core consistency evaluation, variance explanation rate analysis and factor matching scoring on the data, and the optimal number of components is verified by the split-half analysis method, so as to extract the main spectral characteristics of various pollutant components in mineral oil. Next, based on the correspondence between the excitation and emission wavelengths of each standard substance of mineral oil and each fluorescent component obtained by parallel factor analysis, the main mixed pollutant component category in mineral oil is qualitatively determined. Further, based on the correspondence between the fluorescence signal and concentration of each pollutant component, a support vector regression model of the corresponding component is built, thereby realizing the quantitative detection of various pollutant components in mineral oil. This not only improves the efficiency of pollutant detection, but also provides a feasible technical path for pollutant monitoring and environmental protection in related fields.

[0054] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy, characterized in that: The steps include: Step 1: Prediction of the optimal adsorption conditions of magnesium silicate for different vegetable oils; Step 2: Sample preparation and data preprocessing; Step 3: Identify the types of mineral oil contaminants; Step 4: Qualitative and quantitative analysis of mineral oil pollution components.

2. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 1, characterized in that: In step one, the adsorption data of magnesium silicate on different vegetable oils under different adsorption conditions are used, with the adsorption amount as the dependent variable and the different adsorption conditions as the independent variables. A linear regression model is established, the regression equation is fitted and the determination coefficient of the model is calculated; then the optimization algorithm is used to limit the experimental conditions, and through the optimization process, the optimal conditions that meet the adsorption target are finally obtained.

3. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 2, characterized in that: The regression equation is of the form: y=β0+β1x1+β2x2+β3x3+β4x4 Among them, y is the adsorption amount, x1, x2, x3, x4 represent the amount of magnesium silicate, time, water content and temperature, β0, β1, β2, β3, β4 are regression coefficients; In order to find the best experimental conditions, the L-BFGS-B optimization algorithm was used to solve the regression equation, set the adsorption target, and minimize the deviation between the predicted adsorption and the target value; The range of optimized variables is limited to the actual experimental conditions, and through the optimization process, the optimal conditions that meet the adsorption target are gradually approached; After the optimization was completed, the optimal experimental conditions for each vegetable oil included four variables: magnesium silicate dosage, time, water content and temperature.

4. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 3, characterized in that: In step 2, different concentration gradients of mineral oil are added to different vegetable oils, and then a certain amount of n-hexane is added to each mixture to achieve effective adsorption of the vegetable oil under the predicted optimal adsorption conditions of magnesium silicate for different vegetable oils. After the adsorption is completed, the magnesium silicate is filtered to obtain the corresponding mineral oil test solution; The solution to be tested is transferred to a quartz cuvette, and spectral data is collected using a fluorescence spectrophotometer; first, the background signal of the solvent is removed from the collected fluorescence spectral data to eliminate the interference of the solvent on the spectral data; Based on the characteristic that the excitation light wavelength is close to the emission light wavelength and is prone to scattering, the specific wavelength region clipping method is used to remove the scattering noise to obtain a higher signal-to-noise ratio, and then the interpolation method is used to supplement the missing data to ensure the integrity of the data; The sample data of mineral oil of the same type and concentration are averaged to improve the stability and representativeness of the data, and the generalization ability of the model is enhanced through data enhancement technology.

5. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 4, characterized in that: In step three, based on the separated mineral oil contaminant samples, a variety of pre-trained network models are used to extract characteristic fingerprints of different mineral oil categories from the collected three-dimensional fluorescence spectra. Through hyperparameter optimization and comparison of the performance of each model, the optimal model is finally selected to identify several common types of mineral oil contaminants.

6. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 5, characterized in that: A variety of pre-trained network models including artificial neural networks (ANN), convolutional neural networks (CNN), and recurrent neural networks (RNN) are used to extract characteristic fingerprints of different types of mineral oil from the collected three-dimensional fluorescence spectra. The hyperparameters including learning rate, batch size, number of training rounds, optimizer type, and activation function are optimized through grid search, random search, and Bayesian optimization techniques. The performance of each model is compared, and finally the optimal model is selected to identify several common types of mineral oil pollutants.

7. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 6, characterized in that: In step 4, the parallel factor data dimensionality reduction algorithm is combined to decompose and determine the fluorescent components of the corresponding pollutants in the mineral oil; based on the corresponding relationship between the fluorescence signal and concentration of each pollutant component, a support vector regression model of the corresponding component is built to achieve quantitative detection of various pollutants in the mineral oil.

8. The method for detecting contaminated vegetable oil using deep learning-assisted three-dimensional fluorescence spectroscopy according to claim 7, characterized in that: Parallel factor analysis was used to detect outliers in the preprocessed three-dimensional fluorescence spectra, and the optimal number of components of the relevant fluorescence components was screened and verified through core consistency, variance explanation rate, factor matching score and split-half analysis. Based on the correspondence between the excitation and emission wavelengths of each fluorescent component obtained by parallel factor analysis and the standard substances of mineral oil, the categories of the main mixed polluting components in the mineral oil are qualitatively determined. Furthermore, based on the correspondence between the fluorescence signals and concentrations of each polluting component, a support vector regression model of the corresponding component is established to achieve quantitative detection of various polluting components in mineral oil.

Images