Method and device for detecting pesticide residues in combination with mass spectrometry

Abstract

The invention provides a pesticide residue detection method and device combined with mass spectrometry, and relates to the technical field of pesticide residue detection.The method comprises the steps that a sample image is recognized, and a sample type and sample morphology characteristics are determined; matching and determining a pretreatment process and a sample interference matrix; detecting according to a pretreatment process; inputting the sample type, the sample morphology characteristics and the detection data set into a pesticide component analysis channel for identification, and outputting a pesticide residue component set; and performing interference correction on the plurality of pesticide concentrations according to the experimental equipment attribute characteristics, the interference matrix type set and the matrix content set, outputting a plurality of corrected pesticide concentrations, and generating a pesticide residue detection report. The technical problem that the accuracy and reliability of pesticide residue detection are poor due to the fact that matrix interference of different types of products cannot be handled in a traditional method can be solved, the interference problem of pesticide residue detection under a complex matrix can be effectively solved, and the accuracy and reliability of pesticide residue detection are remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of pesticide residue detection, and in particular to a pesticide residue detection method and device combined with mass spectrometry analysis. Background Art

[0002] Product matrix usually refers to all ingredients in the product except the target detection object. These ingredients (such as sugar in fruits, fiber in vegetables, etc.) will interfere with the detection system in pesticide residue detection, which is called matrix effect. With the increase in the variety of agricultural products and the improvement of quality and safety standards, how to effectively eliminate the interference of different matrix components in detection has become a key technical challenge in the field of pesticide residue detection.

[0003] However, when faced with samples with complex matrices (such as fruits, vegetables, grains, dairy products, etc.), traditional detection methods are difficult to control due to matrix interference, and the test results are prone to errors, affecting the accuracy and consistency of the test. This problem is particularly prominent in the field of food safety.

[0004] In summary, traditional pesticide residue detection methods have technical problems such as poor accuracy and reliability of pesticide residue detection due to their inability to cope with matrix interference from different types of products. Summary of the invention

[0005] The purpose of this application is to provide a pesticide residue detection method and device combined with mass spectrometry analysis to solve the technical problem that the traditional pesticide residue detection method is unable to cope with the matrix interference of different types of products, resulting in poor accuracy and reliability of pesticide residue detection.

[0006] In view of the above problems, the present application provides a method and device for detecting pesticide residues combined with mass spectrometry analysis.

[0007] In a first aspect, the present application provides a method for detecting pesticide residues in combination with mass spectrometry, which is implemented by a pesticide residue detection device in combination with mass spectrometry, including: acquiring images of samples to be detected, identifying sample images using a convolutional neural network, and determining sample types and sample morphological features; determining a pretreatment process and a sample interference matrix based on the matching of the sample type and sample morphological features, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; pre-treating the samples according to the pre-treatment process, and jointly detecting the treated samples by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; inputting the sample type, sample morphological features, and detection data set into a pesticide component analysis channel for identification, and outputting a pesticide residue component set, wherein the pesticide residue components include pesticide types and pesticide concentrations; performing interference correction on multiple pesticide concentrations based on the attribute characteristics of the experimental equipment, the interference matrix type set, and the matrix content set, outputting multiple corrected pesticide concentrations, and generating a pesticide residue detection report in combination with multiple pesticide types.

[0008] In a second aspect, the present application also provides a pesticide residue detection device combined with mass spectrometry analysis, which is used to perform a pesticide residue detection method combined with mass spectrometry analysis as described in the first aspect, including: a sample image recognition module, which is used to collect images of samples to be detected, use a convolutional neural network to identify sample images, and determine sample types and sample morphological characteristics; an information matching module, which is used to determine a pre-treatment process and a sample interference matrix according to the sample type and sample morphological characteristics, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; a joint detection module, which is used to match the sample according to the pre-treatment process. The sample is pre-treated, and the treated sample is jointly detected by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; a pesticide component analysis module is used to input the sample type, sample morphology characteristics and the detection data set into the pesticide component analysis channel for identification, and output a pesticide residue component set, wherein the pesticide residue component includes pesticide type and pesticide concentration; a pesticide concentration correction module is used to perform interference correction on multiple pesticide concentrations according to the attribute characteristics of the experimental equipment, the interference matrix type set and the matrix content set, output multiple corrected pesticide concentrations, and generate a pesticide residue detection report in combination with multiple pesticide types.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] By collecting images of the samples to be tested, the sample images are identified using a convolutional neural network to determine the sample type and sample morphological characteristics; the pre-treatment process and the sample interference matrix are determined according to the matching of the sample type and the sample morphological characteristics, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; the samples are pre-treated according to the pre-treatment process, and the treated samples are jointly detected by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; the sample type, sample morphological characteristics and detection data set are input into the pesticide component analysis channel for identification, and a pesticide residue component set is output, wherein the pesticide residue component includes pesticide type and pesticide concentration; interference correction is performed on multiple pesticide concentrations according to the attribute characteristics of the experimental equipment, the interference matrix type set and the matrix content set, multiple corrected pesticide concentrations are output, and a pesticide residue detection report is generated in combination with multiple pesticide types; the interference problem of pesticide residue detection under complex matrices can be effectively solved, and the accuracy and reliability of pesticide residue detection can be significantly improved. At the same time, the efficiency and consistency of pesticide residue detection can be improved through automated pre-processing and intelligent analysis of detection data.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0013] Figure 1 A schematic diagram of a method for detecting pesticide residues combined with mass spectrometry analysis in this application;

[0014] Figure 2 This is a schematic diagram of the structure of a pesticide residue detection device combined with mass spectrometry analysis in this application.

[0015] Description of reference numerals:

[0016] Sample image recognition module 11, information matching module 12, joint detection module 13, pesticide component analysis module 14, pesticide concentration correction module 15. DETAILED DESCRIPTION

[0017] This application provides a pesticide residue detection method and device combined with mass spectrometry analysis to solve the technical problem that the accuracy and reliability of pesticide residue detection are poor due to the inability to cope with matrix interference of different types of products in traditional pesticide residue detection methods. It can effectively solve the interference problem of pesticide residue detection under complex matrices, significantly improve the accuracy and reliability of pesticide residue detection, and at the same time, through automated pre-processing and intelligent analysis of detection data, it can improve the efficiency and consistency of pesticide residue detection.

[0018] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0019] For example, please refer to the attached Figure 1 The present application provides a method for detecting pesticide residues in combination with mass spectrometry, which is applied to a pesticide residue detection device in combination with mass spectrometry, and specifically comprises the following steps:

[0020] Step 1: Capture images of the samples to be tested, use convolutional neural networks to identify sample images, and determine sample types and sample morphological features.

[0021] Specifically, a high-definition camera is used to collect images of samples to be tested (vegetables, fruits, etc.) to obtain sample images; a feature extraction model is constructed based on a convolutional neural network to extract sample features, wherein the feature extraction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer, the input data of the output layer is the product image, and the output data of the output layer is the product morphological features (shape, size, structure, color, texture) and product type (apple, orange), etc., and then a sample product image set, a sample product type set and a sample morphological feature set are collected as training data, and the feature extraction The model is trained until convergence. First, the image, type label and morphological feature label are input into the feature extraction model. Then the cross entropy loss function is used to calculate the error between the model prediction result and the true label, and the Adam or SGD optimization algorithm is used to optimize the model parameters to reduce the loss value. Through multiple rounds of iterative training, the model learns the correspondence between image features and labels, so that the feature extraction capability is gradually enhanced. When the loss function value of the model stabilizes in a lower range and the accuracy on the validation set meets the standard, it means that the model has converged and the training can be stopped to obtain a trained feature extraction model.

[0022] Finally, the sample image is input into the feature extraction model for feature recognition, and the sample type and sample morphology characteristics are output. By constructing a feature extraction model based on a convolutional neural network for sample image recognition, the accuracy of sample classification can be significantly improved, providing a basis for subsequent matching pre-processing procedures and sample interference matrices.

[0023] Step 2: Determine the pre-treatment process and sample interference matrix according to the sample type and sample morphological feature matching, wherein the sample interference matrix includes an interference matrix type set and a matrix content set.

[0024] In one embodiment, step 2 of the present application includes:

[0025] Based on big data and guided by pesticide residue detection, a sample product type set is retrieved and obtained, and multiple sample product morphological feature sets of the sample product type set are retrieved again; a first sample product type is randomly selected, and a first sample product morphological feature of the first sample product type is randomly selected, and multiple first sample interference matrix type sets and multiple first sample matrix content sets are retrieved and obtained according to the first sample product type and the first sample product morphological feature; the occurrence frequency of the same matrix type in the multiple first sample interference matrix type sets is counted, and screening is performed according to a predetermined frequency threshold to determine the first high-frequency interference matrix type set; based on the multiple first sample matrix content sets, multiple first high-frequency sample matrix content sets corresponding to the first high-frequency interference matrix type set are obtained, and the first high-frequency matrix content set is obtained by mean calculation; a first mapping relationship between the first sample product type, the first sample product morphological feature and the first high-frequency interference matrix type set and the first high-frequency matrix content set is established, and a product feature-interference matrix comparison table is constructed according to the first mapping relationship; the sample type and the sample morphological feature are input into the product feature-interference matrix comparison table to obtain the sample interference matrix by matching.

[0026] Specifically, first, the data sources are determined based on big data, including public data sets, industry databases, scientific research papers, agricultural product testing reports, and food safety data released by the government. These data sources usually contain pesticide residue information of multiple products; then, information retrieval is performed with pesticide residue detection as the guide word, that is, according to the needs of pesticide residue detection, relevant data is retrieved and the type information of sample products is extracted. The product types usually involved include different categories of agricultural products, such as fruits (apples, oranges, strawberries, etc.), vegetables (spinach, tomatoes, carrots, etc.) and grains (wheat, rice, etc.); the retrieved product types are further summarized to obtain a sample product type set. Then, the typical morphological features of each sample product type are further retrieved. The morphological features mainly include the shape, size, structure, color, and texture of the product, and multiple sample product morphological feature sets of the sample product type set are obtained, wherein each sample product type corresponds to a sample product morphological feature set.

[0027] Then, a sample product type is randomly selected from the constructed sample product type set. For example, assuming that the sample product types include apples, oranges, spinach, wheat, etc., one of them (such as apple) is randomly selected as the first sample product type; for the selected first sample product type (such as apple), a feature (such as color, size or texture, etc.) is randomly selected from the corresponding morphological feature set, and it is assumed that the selected color is the first sample product morphological feature. Further based on the first sample product type (such as apple) and morphological features (such as color), the possible interference matrix types in different scenarios are retrieved. For example, the matrix interference components of apples include sugars, organic acids, cellulose, etc., and multiple first sample interference matrix type sets are obtained; at the same time, for the obtained interference matrix type set, the content range of each matrix component in the sample product (such as apple) is further retrieved and recorded to form a first sample matrix content set. For example, sugar may account for 5-15%, cellulose content is 1-3%, etc., and multiple first sample matrix content sets are obtained.

[0028] In the obtained multiple first sample interference matrix type sets (such as the interference matrix type set in the apple sample), the names of all interference matrix types are listed, such as sugar, protein, fat, organic acid, cellulose, etc., and the frequency statistics of each matrix type are performed at the same time, and the number of times each matrix type appears in different data sources or documents is recorded. For example, sugar appears 5 times, organic acid appears 3 times, cellulose appears 4 times, etc. The frequency threshold is set according to the detection requirements. The predetermined frequency threshold can be flexibly adjusted according to experience or the distribution of matrix types in the data set. For example, a higher threshold is used to screen matrices with more significant interference effects, and a lower threshold is suitable for a wide range of potential interference factors. A higher frequency of occurrence is usually selected to ensure that the screened matrix types are representative and significant. For example, the frequency threshold is set to 3 times, indicating that only matrix types that appear 3 times or more in the data source are considered to be high-frequency interference matrix types. Further, according to the predetermined frequency threshold, matrix types with a frequency greater than or equal to the threshold are screened out from the frequency statistics results. For example, if the frequency of sugar is 5, the frequency of organic acid is 3, and the frequency of fat is 2, then only sugar and organic acid are retained and fat is excluded when the threshold is 3. The screened high-frequency matrix types are grouped into a first high-frequency interference matrix type set, which includes major matrix interference components with significant effects and can better represent the main interference characteristics of the sample.

[0029] Based on the previously screened first high-frequency interference matrix type set, multiple first sample matrix content sets corresponding thereto are screened out. For example, assuming that the first high-frequency interference matrix type set includes sugars, organic acids and cellulose, the content data of these matrix types are extracted from multiple first sample matrix content sets. Then the content data of each high-frequency interference matrix type is averaged to obtain the average content value of the matrix type, and the first high-frequency matrix content set is obtained. Further establish the first mapping relationship between the first sample product type, the first sample product morphological characteristics and the first high-frequency interference matrix type set, and the first high-frequency matrix content set, that is, the corresponding interference matrix type and content relationship for a specific product type and its morphological characteristics; and use the same method to determine multiple mapping relationships, and construct a product feature-interference matrix comparison table based on multiple mapping relationships, wherein each row of the product feature-interference matrix comparison table represents a product type and its main morphological characteristics, and is associated with the corresponding interference matrix type and content.

[0030] Finally, the sample type and sample morphological features are input into the product feature-interference matrix comparison table for matching. In the comparison table, the corresponding entry is searched according to the input sample type and morphological features. This entry records the interference matrix type and content information of the sample type and specific morphological features. After finding the matching entry, the high-frequency interference matrix type set under the entry is read, that is, the matrix type that has a significant interference effect on the sample type. For example, for the red color of apples and the characteristics of 6-8 cm in diameter, the interference matrix type set may include sugar, organic acid and cellulose, and the sample interference matrix is ​​output. The sample interference matrix includes an interference matrix type set and a matrix content set, such as a sugar content of 6.6%, an organic acid content of 1.33%, and a cellulose content of 2.5%. By inputting the sample type and morphological features and matching the product feature-interference matrix comparison table, the type and content information of the sample interference matrix can be obtained quickly and accurately, providing accurate data support for matrix interference correction in pesticide residue detection. This method significantly improves the automation and standardization level of the detection process, making the detection results more reliable.

[0031] On the other hand, based on the same method of constructing the product feature-interference matrix comparison table, a pretreatment process matching table is constructed, which associates each sample type and its specific morphological characteristics with the optimal pretreatment method to effectively reduce matrix interference and improve the accuracy of the test results.

[0032] Finally, the sample type and sample morphological characteristics are input into the pre-processing process matching table, and the pre-processing process is output. The optimal pre-processing process is found and determined through the pre-processing process matching table, which can effectively improve the matrix removal effect and detection accuracy in pesticide residue detection. This matching process ensures that the processing flow of different samples is accurate and standardized, making the detection process more efficient and reliable.

[0033] Step 3: Pre-treat the sample according to the pre-treatment process, and perform joint detection on the treated sample by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set.

[0034] In one embodiment, step three of the present application includes:

[0035] Pre-treat the sample according to the pre-treatment process to obtain a polar component sample and a volatile component sample, wherein the pre-treatment process includes liquid-liquid distribution, polar separation, volatile separation, solid phase extraction purification and enrichment and concentration; quantitatively analyze the polar component sample based on liquid chromatography tandem mass spectrometry and output liquid phase detection data; quantitatively analyze the volatile component sample based on gas chromatography tandem mass spectrometry and output gas phase detection data; use the liquid phase detection data and the gas phase detection data as the detection data set, wherein the detection data set at least includes chromatographic peak retention time, mass spectrometry signal intensity and mass-to-charge ratio.

[0036] Specifically, a pretreatment process is obtained, wherein the pretreatment process includes liquid-liquid distribution, polarity separation, volatile separation, solid phase extraction purification, and enrichment and concentration. Liquid-liquid distribution refers to liquid-liquid distribution of the sample in two solvents of different polarities. For example, acetonitrile and n-hexane are used to separate polar and non-polar components for further detection; polarity separation refers to further separation of polar components after liquid-liquid distribution, and extraction of polar substances in the sample to ensure that it is suitable for liquid chromatography analysis, such as using salting-out method (such as adding MgSO and NaCl) to separate the extract into layers. The upper polar solvent layer usually contains water-soluble or highly polar pesticides, which are suitable for subsequent liquid chromatography detection, and the lower solvent layer containing less polar compounds is suitable for gas chromatography analysis; volatile separation refers to pretreatment of volatile components, For example, volatile low molecular weight compounds are obtained through distillation or extraction to make them suitable for gas chromatography analysis; solid phase extraction purification refers to the use of solid phase extraction (SPE) methods, such as using C18 columns or NH2 columns, to purify sample extracts, remove interfering matrices, and improve detection selectivity. Solid phase extraction enriches the target pesticide components in the sample and reduces background interference; enrichment and concentration refers to the volatilization of the purified sample solvent by rotary evaporation or nitrogen blowing, so that the target components are enriched in a small volume of solvent to improve detection sensitivity.

[0037] Then, the sample is pretreated according to the pretreatment process to obtain a polar component sample and a volatile component sample, and then the pretreated polar component sample is injected into a liquid chromatography system, and the liquid chromatography system separates the sample components according to the difference in polarity of the compounds, and the separated components are sequentially entered into a tandem mass spectrometer for analysis; the tandem mass spectrometer generates a detection signal by fragmenting ions, performs quantitative analysis on the target polar pesticide component, and obtains liquid phase detection data; the detection data, including chromatographic peak retention time, mass spectrometer signal intensity and mass-to-charge ratio, are recorded as liquid phase detection data output; on the other hand, the pretreated volatile component sample is injected into a gas chromatography system; the gas chromatography separates according to the difference in volatility of different components in the sample, and the separated low molecular weight, volatile pesticide components are sent to a mass spectrometer system for analysis; the quantitative analysis of the volatile components is achieved through multi-stage fragmentation and selective ion detection (such as SIM or MRM mode) to generate gas phase detection data; the gas phase detection data, including chromatographic peak retention time, mass spectrometer signal intensity and mass-to-charge ratio, are recorded.

[0038] Finally, the liquid phase detection data and gas phase detection data are integrated into a complete detection data set, in which the detection data set at least includes chromatographic peak retention time, mass spectrometry signal intensity and mass-to-charge ratio. Through a series of pre-treatment steps, the samples are successfully separated into polar component and volatile component samples, and quantitatively analyzed by liquid chromatography-mass spectrometry and gas chromatography-mass spectrometry respectively. The constructed detection data set contains key parameters such as chromatographic peak retention time, mass spectrometry signal intensity and mass-to-charge ratio, which provides comprehensive and accurate data support for subsequent pesticide residue detection and analysis.

[0039] Step 4: Input the sample type, sample morphological characteristics and detection data set into the pesticide component analysis channel for identification, and output a pesticide residue component set, wherein the pesticide residue component includes pesticide type and pesticide concentration.

[0040] In one embodiment, step 4 of the present application includes:

[0041] Taking the sample type and sample morphological features as retrieval constraints, multiple sample detection data sets, multiple sample pesticide type sets and multiple sample pesticide concentration sets are retrieved and acquired; taking the sample detection data set as input, and taking the sample pesticide type set and the sample pesticide concentration set as supervision, the random decision forest is trained and verified according to the multiple sample detection data sets, multiple sample pesticide type sets and multiple sample pesticide concentration sets until convergence, thereby generating the pesticide component analysis channel.

[0042] Specifically, using the sample type and sample morphological features as retrieval constraints, multiple sample detection data sets that meet the conditions in the database are retrieved and collected, each detection data set should contain detailed data of the sample after pre-processing and detection, and multiple sample pesticide type sets and multiple sample pesticide concentration sets corresponding to the multiple sample detection data sets are obtained, wherein the sample detection data, sample pesticide type set and sample pesticide concentration set correspond to each other one by one.

[0043] Then, a pesticide component analysis channel is constructed based on random decision forests, where random forests are an integrated learning method that improves the accuracy and robustness of the model by integrating multiple decision trees. Then, the sample detection data set is used as input, and the sample pesticide type set and sample pesticide concentration set are used as supervision. The pesticide component analysis channel is supervised and trained according to the multiple sample detection data sets, multiple sample pesticide type sets, and multiple sample pesticide concentration sets. That is, the detection data set of the training set is used as input, and the pesticide type set and concentration set are used as supervision data to train the random forest model. The model continuously optimizes the node splitting rules of each tree by learning the correspondence between the sample detection data and the pesticide type and concentration; at the same time, the mean square error (MSE) or classification cross entropy is used as the loss function to evaluate the accuracy of predicting the pesticide type and concentration and guide the model optimization; the random forest model gradually reduces the loss function value through iterative training. As the number of trees increases, the model continuously corrects the wrong predictions and improves the overall accuracy. After the iterative training is completed, the validation set is input into the model to evaluate the model's ability to predict pesticide types and concentrations. Based on the performance of the validation set, the model parameters, such as the number of trees and maximum depth, are further adjusted to improve model performance. When the loss function value on the validation set reaches a stable state and the accuracy reaches the expected standard, the model can be considered to have converged and the trained pesticide component analysis channel is output.

[0044] Finally, the test data set is input into the pesticide component analysis channel for identification, and the pesticide residue component set is output, where the pesticide residue components include pesticide type and pesticide concentration. Through the use of random decision forests for training and verification, the pesticide component analysis channel constructed based on test data, pesticide type and concentration can efficiently and accurately perform qualitative and quantitative analysis of pesticide residues, providing an intelligent and automated solution for pesticide residue detection.

[0045] Step 5: Perform interference correction on multiple pesticide concentrations according to the property characteristics of the experimental equipment, the interference matrix type set and the matrix content set, output multiple corrected pesticide concentrations, and generate a pesticide residue detection report in combination with multiple pesticide types.

[0046] In one embodiment, step five of the present application includes:

[0047] Based on big data, guided by the multiple pesticide types in the pesticide residue component set, a sample equipment attribute feature set, multiple sample interference matrix type sets, multiple sample matrix content sets, multiple sample pesticide concentration sets, and multiple sample correction pesticide concentration sets are retrieved and obtained as a sample data set; the sample data set is divided according to the pesticide type to obtain multiple training data sets corresponding to the multiple pesticide types;

[0048] Specifically, in the detection of pesticide residues, in order to improve the accuracy and reliability of the detection, it is necessary to correct the experimental data for matrix interference. By combining the attribute characteristics of the experimental equipment, the interference matrix type set and the matrix content set, the pesticide concentration in the test results is corrected, and finally a complete pesticide residue detection report is generated.

[0049] Firstly, guided by the multiple pesticide types (such as imidacloprid, chlorpyrifos, cypermethrin, etc.) included in the pesticide residue component set, the relevant data sets in the database are retrieved to construct a sample data set covering multiple pesticide types, wherein the retrieved data content includes a sample equipment attribute feature set, a sample interference matrix type set, a sample matrix content set, a sample pesticide concentration set, and a sample corrected pesticide concentration set after matrix interference correction. The sample equipment attribute feature set includes information such as equipment type, detection sensitivity, and resolution, reflecting the influence of different detection equipment on sample detection; the sample interference matrix type set includes sugar, cellulose, fat, etc., which affects the accuracy of pesticide detection; the sample pesticide concentration set refers to the original concentration value of the detected pesticide, which provides input data for the correction model; the sample corrected pesticide concentration set refers to the corrected pesticide concentration value, which serves as the target output data of the correction model, and multiple sample interference matrix type sets, multiple sample matrix content sets, multiple sample pesticide concentration sets, and multiple sample corrected pesticide concentration sets are obtained as sample data sets. The sample data set is further divided according to the pesticide type, that is, the sample data set is divided into multiple subsets according to different pesticide types, so as to build an independent calibration model for each pesticide type, and obtain multiple training data sets corresponding to multiple pesticide types. By dividing the sample data set based on the pesticide type, a more targeted calibration model can be built, the accuracy and adaptability of the model can be improved, and standardized and accurate calibration support can be provided for pesticide residue detection.

[0050] Using the multiple training data sets, supervised training is performed based on machine learning to construct multiple interference correction branches, and an interference corrector is constructed based on the fusion of the multiple interference correction branches;

[0051] In one embodiment, the present application further comprises the following steps:

[0052] N interference correction operators are configured based on machine learning, where N is an integer greater than 2; a first training data set of a first pesticide type is randomly selected from the multiple training data sets, the first training data set is divided into 3N equal parts, and the first training data set is selected 3N times with replacement to construct a first training set, and the first training sets are selected iteratively N times to obtain N training sets; using the N training sets, the N interference correction operators are respectively subjected to supervised training and cross-validation to obtain N interference correction plug-ins that meet the convergence conditions; a first interference correction unit is integrated and constructed based on the N interference correction plug-ins, a first interference correction branch is generated in combination with the first pesticide type, and the first interference correction branch is added to the multiple interference correction branches.

[0053] Specifically, first, N interference correction operators are configured based on machine learning, wherein N is an integer greater than 2, such as BP neural network, support vector machine, random decision forest, etc., and a correction operator adapted can be selected according to actual needs. Then the first training data set of the first pesticide type (such as imidacloprid) is randomly selected in the multiple training data sets. The data set includes relevant information such as detection characteristics, matrix interference characteristics, equipment attributes and correction concentration of the first pesticide type, and the first training data set is divided into 3N parts, 3N times are selected with replacement, and the first training set is composed, and the same method is used to iteratively select N times to obtain N training sets, wherein the selection method with replacement can ensure that some data blocks are repeatedly extracted, and the weight of a specific sample is increased, thereby improving the diversity of data.

[0054] Then, the N interference correction operators are supervised and cross-validated using the N training sets to obtain N interference correction plug-ins that meet the convergence conditions. For example, a first interference correction plug-in is constructed based on the first correction operator (BP neural network), wherein the first interference correction plug-in includes an input layer, a hidden layer, and an output layer, and the input data of the first interference correction plug-in are equipment attribute characteristics, interference matrix type set, matrix content set, and pesticide concentration set, and the output data is a corrected pesticide concentration set; then, the first interference correction plug-in is supervised and validated through the first training set. First, the training sample is input into the network, and the output correction concentration is calculated, and then the difference between the predicted correction concentration and the target correction concentration is calculated, and then the BP algorithm is used to calculate the gradient of the error relative to each parameter, and the network weight is updated, and the above steps are repeated until the model reaches the convergence condition on the training set, that is, the loss function value tends to be stable. After the supervised training is completed, K-fold cross validation (such as 5-fold cross validation) is used for validation training. During each training, the data is divided into K parts, one of which is used as a validation set and the remaining part is used as a training set. Alternate verification, calculate the average correction error of the model, when the verification error reaches the preset threshold and is stable, the model is judged to have converged, and the first interference correction plug-in that has been trained is obtained. Through supervised training and verification of the first training set by BP neural network, the first interference correction plug-in that meets the convergence conditions is constructed. This plug-in effectively corrects the interference of equipment attributes, matrix type and concentration data, and provides accurate concentration correction function for pesticide residue detection system.

[0055] Finally, a first interference correction unit is integrated and constructed according to the N interference correction plug-ins, and a first interference correction branch is generated in combination with the first pesticide type, that is, the first interference correction branch is called by matching the first pesticide type, and the first interference correction branch is added to the multiple interference correction branches.

[0056] In one embodiment, the present application further comprises the following steps:

[0057] Read the N convergence precisions of the N interference correction plug-ins; configure N weight ratios for the N convergence precisions based on the coefficient of variation method; and integrate and construct a first interference correction unit based on the N interference correction plug-ins and the N weight ratios, wherein the output of the first interference correction unit is the weighted output result of the N interference correction plug-ins.

[0058] Specifically, the convergence accuracy of the N interference correction plug-ins is read, where each interference correction plug-in has a convergence accuracy during the training and verification process, which is usually expressed as the average error of the model (such as mean square error MSE), characterizing the accuracy of the model output, and obtaining N convergence accuracy. Then, based on the coefficient of variation method, weights are assigned according to the relative differences in the convergence accuracy of each plug-in, where the convergence accuracy and weight are positively correlated, that is, the greater the convergence accuracy, the greater the corresponding weight of the plug-in, and N weight ratios are obtained. Through this positive correlation weight allocation method, the correction accuracy differences of each plug-in can be effectively utilized, the influence of high convergence accuracy plug-ins can be maximized, and accurate weighted output can be provided for the correction unit.

[0059] Then, based on the N interference correction plug-ins and the N weight ratios, the first interference correction unit is integrated and constructed, wherein the output of the first interference correction unit is the weighted output result of the N interference correction plug-ins, that is, the output of the first interference correction unit is the weighted correction concentration value of all plug-ins, which can integrate the advantages of multiple correction plug-ins and improve the overall correction accuracy and reliability. By configuring weights through the coefficient of variation method and integrating the weighted outputs of N interference correction plug-ins, the first interference correction unit can provide accurate and comprehensive pesticide concentration correction, providing strong support for the high-precision analysis of the detection system.

[0060] The multiple pesticide types, multiple pesticide concentrations, experimental equipment attribute characteristics, interference matrix type set, and matrix content set are input into the interference corrector, and multiple corrected pesticide concentrations are output.

[0061] Specifically, the multiple pesticide types are finally input into the interference corrector for matching, the corresponding interference correction branches are called, and multiple matching interference correction branches are determined; then multiple pesticide concentrations, experimental equipment attribute characteristics, interference matrix type sets, and matrix content sets are input into multiple matching interference correction branches for analysis, and multiple corrected pesticide concentrations are output.

[0062] In summary, the pesticide residue detection method combined with mass spectrometry provided in this application has the following technical effects:

[0063] By collecting images of the samples to be tested, the sample images are identified using a convolutional neural network to determine the sample type and sample morphological characteristics; the pre-treatment process and the sample interference matrix are determined according to the matching of the sample type and the sample morphological characteristics, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; the samples are pre-treated according to the pre-treatment process, and the treated samples are jointly detected by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; the sample type, sample morphological characteristics and detection data set are input into the pesticide component analysis channel for identification, and a pesticide residue component set is output, wherein the pesticide residue component includes pesticide type and pesticide concentration; interference correction is performed on multiple pesticide concentrations according to the attribute characteristics of the experimental equipment, the interference matrix type set and the matrix content set, multiple corrected pesticide concentrations are output, and a pesticide residue detection report is generated in combination with multiple pesticide types; the interference problem of pesticide residue detection under complex matrices can be effectively solved, and the accuracy and reliability of pesticide residue detection can be significantly improved. At the same time, the efficiency and consistency of pesticide residue detection can be improved through automated pre-processing and intelligent analysis of detection data.

[0064] Embodiment 2: Based on the pesticide residue detection method combined with mass spectrometry analysis in the above embodiment, the present application also provides a pesticide residue detection device combined with mass spectrometry analysis, please refer to the attached Figure 2 ,include:

[0065] The sample image recognition module 11 is used to collect images of the samples to be detected, identify the sample images using a convolutional neural network, and determine the sample type and sample morphological characteristics; the information matching module 12 is used to determine the pre-treatment process and the sample interference matrix according to the sample type and sample morphological characteristics, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; the joint detection module 13 is used to pre-treat the sample according to the pre-treatment process, and perform joint detection on the treated sample by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; the pesticide component analysis module 14 is used to input the sample type, sample morphological characteristics and detection data set into the pesticide component analysis channel for identification, and output a pesticide residue component set, wherein the pesticide residue component includes a pesticide type and a pesticide concentration; the pesticide concentration correction module 15 is used to perform interference correction on multiple pesticide concentrations according to the attribute characteristics of the experimental equipment, the interference matrix type set and the matrix content set, output multiple corrected pesticide concentrations, and generate a pesticide residue detection report in combination with multiple pesticide types.

[0066] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0067] Based on big data and guided by pesticide residue detection, a sample product type set is retrieved and obtained, and multiple sample product morphological feature sets of the sample product type set are retrieved again; a first sample product type is randomly selected, and a first sample product morphological feature of the first sample product type is randomly selected, and multiple first sample interference matrix type sets and multiple first sample matrix content sets are retrieved and obtained according to the first sample product type and the first sample product morphological feature; the occurrence frequency of the same matrix type in the multiple first sample interference matrix type sets is counted, and screening is performed according to a predetermined frequency threshold to determine the first high-frequency interference matrix type set; based on the multiple first sample matrix content sets, multiple first high-frequency sample matrix content sets corresponding to the first high-frequency interference matrix type set are obtained, and the first high-frequency matrix content set is obtained by mean calculation; a first mapping relationship between the first sample product type, the first sample product morphological feature and the first high-frequency interference matrix type set and the first high-frequency matrix content set is established, and a product feature-interference matrix comparison table is constructed according to the first mapping relationship; the sample type and the sample morphological feature are input into the product feature-interference matrix comparison table to obtain the sample interference matrix by matching.

[0068] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0069] Pre-treat the sample according to the pre-treatment process to obtain a polar component sample and a volatile component sample, wherein the pre-treatment process includes liquid-liquid distribution, polar separation, volatile separation, solid phase extraction purification and enrichment and concentration; quantitatively analyze the polar component sample based on liquid chromatography tandem mass spectrometry and output liquid phase detection data; quantitatively analyze the volatile component sample based on gas chromatography tandem mass spectrometry and output gas phase detection data; use the liquid phase detection data and the gas phase detection data as the detection data set, wherein the detection data set at least includes chromatographic peak retention time, mass spectrometry signal intensity and mass-to-charge ratio.

[0070] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0071] Taking the sample type and sample morphological features as retrieval constraints, multiple sample detection data sets, multiple sample pesticide type sets and multiple sample pesticide concentration sets are retrieved and acquired; taking the sample detection data set as input, and taking the sample pesticide type set and the sample pesticide concentration set as supervision, the random decision forest is trained and verified according to the multiple sample detection data sets, multiple sample pesticide type sets and multiple sample pesticide concentration sets until convergence, thereby generating the pesticide component analysis channel.

[0072] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0073] Based on big data, guided by the multiple pesticide types in the pesticide residue component set, sample equipment attribute feature sets, multiple sample interference matrix type sets, multiple sample matrix content sets, multiple sample pesticide concentration sets and multiple sample correction pesticide concentration sets are retrieved and obtained as sample data sets; the sample data sets are divided according to the pesticide types to obtain multiple training data sets corresponding to the multiple pesticide types; using the multiple training data sets, supervised training is performed based on machine learning to construct multiple interference correction branches, and an interference corrector is constructed based on the fusion of the multiple interference correction branches; the multiple pesticide types, multiple pesticide concentrations, experimental equipment attribute characteristics, interference matrix type sets, and matrix content sets are input into the interference corrector, and multiple corrected pesticide concentrations are output.

[0074] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0075] N interference correction operators are configured based on machine learning, where N is an integer greater than 2; a first training data set of a first pesticide type is randomly selected from the multiple training data sets, the first training data set is divided into 3N equal parts, and the first training data set is selected 3N times with replacement to construct a first training set, and the first training sets are selected iteratively N times to obtain N training sets; using the N training sets, the N interference correction operators are respectively subjected to supervised training and cross-validation to obtain N interference correction plug-ins that meet the convergence conditions; a first interference correction unit is integrated and constructed based on the N interference correction plug-ins, a first interference correction branch is generated in combination with the first pesticide type, and the first interference correction branch is added to the multiple interference correction branches.

[0076] Furthermore, the pesticide residue detection device combined with mass spectrometry is also used for:

[0077] Read the N convergence precisions of the N interference correction plug-ins; configure N weight ratios for the N convergence precisions based on the coefficient of variation method; and integrate and construct a first interference correction unit based on the N interference correction plug-ins and the N weight ratios, wherein the output of the first interference correction unit is the weighted output result of the N interference correction plug-ins.

[0078] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The pesticide residue detection method combined with mass spectrometry analysis and the specific examples in the aforementioned embodiment 1 are also applicable to a pesticide residue detection device combined with mass spectrometry analysis in this embodiment. Through the aforementioned detailed description of a pesticide residue detection method combined with mass spectrometry analysis, those skilled in the art can clearly know a pesticide residue detection device combined with mass spectrometry analysis in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0079] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0080] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A method for detecting pesticide residues in combination with mass spectrometry, characterized in that: include: Capture images of samples to be tested, use convolutional neural networks to identify sample images, and determine sample types and sample morphological features; Determining the pre-treatment process and the sample interference matrix according to the sample type and the sample morphological feature matching, wherein the sample interference matrix includes an interference matrix type set and a matrix content set; Pre-process the sample according to the pre-processing process, and jointly detect the processed sample by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; Input the sample type, sample morphological characteristics and detection data set into the pesticide component analysis channel for identification, and output a pesticide residue component set, wherein the pesticide residue component includes pesticide type and pesticide concentration; According to the property characteristics of the experimental equipment, the interference matrix type set and the matrix content set, interference correction is performed on multiple pesticide concentrations, multiple corrected pesticide concentrations are output, and a pesticide residue detection report is generated in combination with multiple pesticide types.

2. The method for detecting pesticide residues combined with mass spectrometry according to claim 1, characterized in that: Determining the sample interference matrix according to the sample type and sample morphological characteristics matching includes: Based on big data and guided by pesticide residue detection, a sample product type set is retrieved and obtained, and multiple sample product morphology feature sets of the sample product type set are retrieved again; Randomly select a first sample product type, and randomly select a first sample product morphological feature of the first sample product type, and retrieve and obtain a plurality of first sample interference matrix type sets and a plurality of first sample matrix content sets according to the first sample product type and the first sample product morphological feature; Counting the occurrence frequencies of the same matrix type in the plurality of first sample interfering matrix type sets, and screening them according to a predetermined frequency threshold to determine a first high-frequency interfering matrix type set; Based on the multiple first sample matrix content sets, multiple first high-frequency sample matrix content sets corresponding to the first high-frequency interference matrix type set are obtained, and the first high-frequency matrix content set is obtained by average calculation; Establishing a first mapping relationship between the first sample product type, the first sample product morphological feature, the first high-frequency interference matrix type set, and the first high-frequency matrix content set, and constructing a product feature-interference matrix comparison table according to the first mapping relationship; The sample type and sample morphological characteristics are input into the product characteristics-interference matrix comparison table, and the sample interference matrix is ​​obtained by matching.

3. The method for detecting pesticide residues combined with mass spectrometry according to claim 1, characterized in that: The samples are pre-treated according to the pre-treatment process, and the processed samples are jointly detected by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set, including: Pre-treating the sample according to the pre-treatment process to obtain a polar component sample and a volatile component sample, wherein the pre-treatment process includes liquid-liquid distribution, polarity separation, volatile separation, solid phase extraction purification, and enrichment and concentration; Performing quantitative analysis on the polar component sample based on liquid chromatography tandem mass spectrometry, and outputting liquid phase detection data; Performing quantitative analysis on the volatile component sample based on gas chromatography tandem mass spectrometry and outputting gas phase detection data; The liquid phase detection data and the gas phase detection data are used as the detection data set, wherein the detection data set at least includes chromatographic peak retention time, mass spectrum signal intensity and mass-to-charge ratio.

4. The method for detecting pesticide residues combined with mass spectrometry according to claim 1, characterized in that: Construct a pesticide component analysis channel, including: Using the sample type and sample morphological features as search constraints, searching and acquiring multiple sample detection data sets, multiple sample pesticide type sets, and multiple sample pesticide concentration sets; Taking a sample detection data set as input and a sample pesticide type set and a sample pesticide concentration set as supervision, a random decision forest is trained and verified according to the multiple sample detection data sets, the multiple sample pesticide type sets and the multiple sample pesticide concentration sets until convergence, thereby generating the pesticide component analysis channel.

5. The method for detecting pesticide residues combined with mass spectrometry according to claim 1, characterized in that: According to the experimental equipment attribute characteristics, the interference matrix type set and the matrix content set, interference correction is performed on multiple pesticide concentrations, and multiple corrected pesticide concentrations are output, including: Based on big data, guided by multiple pesticide types in the pesticide residue component set, a sample equipment attribute feature set, multiple sample interference matrix type sets, multiple sample matrix content sets, multiple sample pesticide concentration sets, and multiple sample correction pesticide concentration sets are retrieved and obtained as a sample data set; Dividing the sample data set according to pesticide types to obtain multiple training data sets corresponding to multiple pesticide types; Using the multiple training data sets, supervised training is performed based on machine learning to construct multiple interference correction branches, and an interference corrector is constructed based on the fusion of the multiple interference correction branches; The multiple pesticide types, multiple pesticide concentrations, experimental equipment attribute characteristics, interference matrix type set, and matrix content set are input into the interference corrector, and multiple corrected pesticide concentrations are output.

6. The method for detecting pesticide residues combined with mass spectrometry according to claim 5, characterized in that: Construct multiple interference correction branches, including: configuring N interference correction operators based on machine learning, where N is an integer greater than 2; Randomly selecting a first training data set of a first pesticide type from the multiple training data sets, dividing the first training data set into 3N equal parts, selecting 3N parts with replacement to construct a first training set, iterating and selecting N times to obtain N training sets; Using the N training sets, respectively performing supervised training and cross-validation on the N interference correction operators to obtain N interference correction plug-ins that meet convergence conditions; A first interference correction unit is integrated and constructed based on the N interference correction plug-ins, and a first interference correction branch is generated in combination with the first pesticide type and added to the multiple interference correction branches.

7. The method for detecting pesticide residues combined with mass spectrometry according to claim 6, characterized in that: A first interference correction unit is constructed based on the N interference correction plug-ins, including: Reading N convergence precisions of the N interference correction plug-ins; Based on the coefficient of variation method, N weight proportions are configured for the N convergence precisions; A first interference correction unit is integrated and constructed based on the N interference correction plug-ins and the N weight ratios, wherein the output of the first interference correction unit is a weighted output result of the N interference correction plug-ins.

8. A pesticide residue detection device combined with mass spectrometry, characterized in that: The steps for implementing the method for detecting pesticide residues combined with mass spectrometry according to any one of claims 1 to 7 include: The sample image recognition module is used to collect images of the samples to be tested, identify the sample images using a convolutional neural network, and determine the sample type and sample morphological characteristics; An information matching module, used to determine the pre-treatment process and the sample interfering matrix according to the sample type and the sample morphological characteristics, wherein the sample interfering matrix includes an interfering matrix type set and a matrix content set; A joint detection module, used to pre-process the samples according to the pre-processing process, and perform joint detection on the processed samples by liquid chromatography tandem mass spectrometry and gas chromatography tandem mass spectrometry to obtain a detection data set; A pesticide component analysis module, used to input the sample type, sample morphological characteristics and detection data set into the pesticide component analysis channel for identification, and output a pesticide residue component set, wherein the pesticide residue component includes the pesticide type and the pesticide concentration; The pesticide concentration correction module is used to perform interference correction on multiple pesticide concentrations according to the property characteristics of the experimental equipment, the interference matrix type set and the matrix content set, output multiple corrected pesticide concentrations, and generate a pesticide residue detection report in combination with multiple pesticide types.

Images