CN-119880863-B - Method for detecting polluted vegetable oil by deep learning auxiliary three-dimensional fluorescence spectrum

CN119880863BCN 119880863 BCN119880863 BCN 119880863BCN-119880863-B

Abstract

The invention belongs to the technical field of food safety detection, and particularly relates to a detection method of deep learning auxiliary three-dimensional fluorescence spectrum on polluted vegetable oil. According to the invention, firstly, the optimal adsorption condition of magnesium silicate on vegetable oil is predicted based on a linear regression model and an L-BFGS-B algorithm, and the targeted separation of mineral oil in polluted vegetable oil is realized through the adsorption of magnesium silicate on the vegetable oil under the optimal condition. Then, based on the separated and obtained mineral oil pollutant samples, various pre-training network models are applied to extract characteristic fingerprints of different mineral oil categories from the collected three-dimensional fluorescence spectrum, so that several common mineral oil pollutant categories are distinguished. And then, a parallel factor data dimension reduction algorithm is combined to decompose and judge fluorescent components of corresponding pollutants in the mineral oil. And finally, constructing a support vector regression model of the corresponding components based on the corresponding relation between the fluorescent signals and the concentrations of the pollution components, so as to realize quantitative detection of various pollution components in the mineral oil.

Inventors

ZHANG JIAN
JIN ZHU
ZHANG KUI

Assignees

安徽工业大学

Dates

Publication Date: 20260508
Application Date: 20250113

Claims (5)

1. The detection method of the deep learning auxiliary three-dimensional fluorescence spectrum on the polluted vegetable oil is characterized by comprising the following steps of: firstly, predicting optimal adsorption conditions of magnesium silicate on different vegetable oils; step two, sample preparation and data pretreatment; Step three, the type of mineral oil pollutants is qualitative; Step four, qualitative and quantitative determination of mineral oil pollution components; firstly, utilizing magnesium silicate to adsorb amount data of different vegetable oils under different adsorption conditions, taking the adsorption amount as a dependent variable and different adsorption conditions as independent variables, establishing a linear regression model, fitting a regression equation and calculating a determination coefficient of the model; the regression equation is in the form of: ; wherein y is the adsorption quantity, , , , Respectively represents the dosage, time, water content and temperature of magnesium silicate, Is a regression coefficient; In order to find the optimal experimental condition, solving a regression equation by using an L-BFGS-B optimization algorithm, setting an adsorption target, and minimizing the deviation of the predicted adsorption and the target value; The optimized variable range is limited in the actual experimental condition, and the optimal condition meeting the adsorption target is gradually approximated through the optimization process; After optimization is completed, the optimal experimental conditions of each vegetable oil comprise four variables of magnesium silicate consumption, time, water content and temperature; In the second step, mineral oil with different concentration gradients is respectively added into different vegetable oils, then a certain amount of normal hexane is added into each mixture, effective adsorption of the vegetable oil is realized under the predicted optimal adsorption condition of magnesium silicate on different vegetable oils, the magnesium silicate is filtered after the adsorption is finished to obtain a corresponding mineral oil to-be-detected solution, the to-be-detected solution is transferred into a quartz cuvette, a fluorescence spectrophotometer is used for collecting spectral data, firstly, the collected fluorescence spectral data is used for removing background signals of a solvent to eliminate interference of the solvent to the spectral data, based on the characteristic that the excitation wavelength and the emission wavelength are close to each other and are easy to scatter, scattering noise is removed by utilizing a specific wavelength region clipping method to obtain higher signal-to-noise ratio, then interpolation method is adopted to supplement missing data to ensure the integrity of the data, and the sample data of the mineral oil with the same kind and concentration is subjected to average treatment to improve the stability and representativeness of the data, and the generalization capability of a model is enhanced by a data enhancement technology.
2. The method for detecting the polluted vegetable oil by using the deep learning-aided three-dimensional fluorescence spectrum according to claim 1, wherein in the third step, based on the separated and acquired mineral oil pollutant samples, a plurality of pre-training network models are applied to extract characteristic fingerprints of different mineral oil categories from the acquired three-dimensional fluorescence spectrum, the performance of each model is optimized through super parameters, and finally, an optimal model is selected to distinguish several common mineral oil pollutant categories.
3. The method for detecting the polluted vegetable oil by using the deep learning-assisted three-dimensional fluorescence spectrum according to claim 2, wherein a plurality of pre-training network models including an artificial neural network ANN, a convolutional neural network CNN and a cyclic neural network RNN are used for extracting characteristic fingerprints of different mineral oil categories from the acquired three-dimensional fluorescence spectrum, the super parameters including learning rate, batch size, training round number, optimizer type and activation function are optimized through grid search, random search and Bayesian optimization technology, the performances of the models are compared, and finally the optimal model is selected to distinguish several common mineral oil pollutant categories.
4. The method for detecting the polluted vegetable oil by using the deep learning-aided three-dimensional fluorescence spectrum according to claim 3 is characterized by combining a parallel factor data dimension reduction algorithm to decompose and judge fluorescent components of corresponding pollutants in the mineral oil, and constructing a support vector regression model of the corresponding components based on the corresponding relation between fluorescent signals and concentrations of the various pollution components so as to realize quantitative detection of various pollution components in the mineral oil.
5. The method for detecting the polluted vegetable oil by using the deep learning auxiliary three-dimensional fluorescence spectrum according to claim 4 is characterized by detecting abnormal values of the preprocessed three-dimensional fluorescence spectrum by using parallel factor analysis, obtaining the optimal component number of related fluorescent components through screening and verification by a core consistency, variance interpretation rate, factor matching scoring and split-half analysis method, qualitatively judging main mixed pollution component types in the mineral oil based on the corresponding relation between each standard substance of the mineral oil and excitation and emission wavelengths of each fluorescent component obtained by the parallel factor analysis, and further constructing a support vector regression model of the corresponding components based on the corresponding relation between fluorescent signals and concentrations of each pollution component, thereby realizing quantitative detection of various pollution components in the mineral oil.

Description

Method for detecting polluted vegetable oil by deep learning auxiliary three-dimensional fluorescence spectrum Technical Field The invention belongs to the technical field of food safety detection, and particularly relates to a detection method of deep learning auxiliary three-dimensional fluorescence spectrum on polluted vegetable oil. Background The edible vegetable oil is polluted by mineral oil, seriously violates food safety and endangers public health. Typically, gas chromatography mass spectrometry (GC-MS) is used as the first choice for analysis of mineral oil contaminants in vegetable oils. The method has high sensitivity and accuracy in a laboratory, and can effectively identify and quantify mineral oil pollutants and components thereof. However, vegetable oils themselves are complex matrices containing large amounts of fatty acids and other organic compounds, which are similar to the components of mineral oils, tend to produce matrix effects that affect the accuracy of GC-MS detection results, and the process is complex and long, which hampers the food safety detection process requiring rapid response. It is therefore extremely important to develop an efficient detection method. Highly alkylated aromatic compounds in mineral oils have a fluorescent response due to the large pi bonds in their molecular structure. Excitation-emission matrix fluorescence (EEMF) technology has been increasingly used in food analysis because of its high sensitivity and the advantage of obtaining information about the fluorescence excitation wavelength, emission wavelength, and fluorescence intensity of fluorescent compounds in a short time. However, the collected fluorescence data are large and complex, and simple methods such as linear regression and multivariate statistical analysis cannot fully mine and effectively utilize the fluorescence information, so that a more excellent high-dimensional data processing and complex pattern recognition method are required. The rapid development of Artificial Intelligence (AI) will remodel almost every field of work. As a sub-field of artificial intelligence, deep learning has a strong information mining capability, and in recent years, application research thereof in the field of food safety detection is endless. The three-dimensional fluorescence spectrum is combined with a Convolutional Neural Network (CNN) to achieve 97.5% classification accuracy for different pesticide residue types on the surfaces of vegetables and fruits. The Zhang research team successfully realizes the classification of heavy metals such as lead, cadmium, arsenic and the like in different food samples based on the same method, and the classification accuracy is as high as 98.2 percent. On the other hand, the optimal condition of magnesium silicate adsorption vegetable oil is predicted through machine learning, the targeted separation of mineral oil can be realized, the matrix effect is eliminated to a great extent, and more effective spectrum data are acquired, so that the accuracy of classification and quantification is improved. At present, deep learning has achieved better results in studying classification problems. However, for consumers, it is of greater concern which components of mineral oil contaminants are harmful to the human body, whether the content of these components is below a safety threshold. Therefore, it is extremely important how to analyze the harmful components and the contents thereof in each mineral oil pollutant. Parallel factor analysis (PARAFAC) is a multivariate data analysis technique capable of decomposing fluorescent signals into individual fluorescence phenomena. This technique allows the tracking of fluorescent components in different parts of the mineral oil and the separation of fluorescent signals with specific excitation and emission spectra. However, the relationship between the fluorescence intensity and the concentration obtained by the analysis is not always linearly dependent on factors such as self-absorption effect and fluorescence quenching. Support vector regression (Support Vector Regression, SVR) is a regression model extended by a support vector machine (Support Vector Machine, SVM), and can well process nonlinear relations with a large range based on a kernel function, and is frequently applied to nonlinear relation environments by researchers at present. Although the problem of mineral oil contamination of vegetable oils has been a growing concern in the field of food safety, the current research has been focused mainly on the detection of the source and total amount of mineral oils, and, upon searching, no related research and report has been found on the classification of mineral oils in contaminated vegetable oils and the qualitative and quantitative analysis of their contaminating components. In view of the above, the inventors expect to provide a method for detecting contaminated vegetable oil by deep learning-assisted three-dimensional fluoresce