Mycotoxin pollution detection method and system for pet food

Abstract

The invention relates to a mycotoxin pollution detection method and system for pet food, and the method comprises the steps: collecting a to-be-detected sample, and obtaining the to-be-detected sample; performing volatile compound analysis on the to-be-detected sample to obtain smell data of the to-be-detected sample; detecting the content of mycotoxin in the to-be-detected sample, and marking the to-be-detected sample according to the content of mycotoxin to obtain marking information; training a plurality of machine learning models through the smell data and the mark information to obtain an integrated prediction model; and performing pollution prediction on a to-be-detected product through the integrated prediction model to obtain a prediction result. The method not only has high sensitivity and specificity, but also can reduce the detection cost of mycotoxin pollution and improve the detection efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of mycotoxin detection, and in particular to a method and system for detecting mycotoxin contamination in pet food. Background Art

[0002] With the rapid development of the pet food market, the increase in the number of pets has led to increasing attention to the safety of pet food. Mycotoxins pose a potential threat to pet health and may cause a series of health problems such as growth inhibition, reproductive toxicity and immune system damage.

[0003] At present, traditional methods for detecting mycotoxin contamination levels in pet food mainly rely on chemical analysis techniques, such as high performance liquid chromatography (HPLC) and liquid chromatography-mass spectrometry (LC-MS), and usually require special equipment and laboratory environment. Not only are they not suitable for rapid detection at production sites or in various links of the supply chain, they are also time-consuming, labor-intensive, and costly. Sample pre-treatment is complex and the technical requirements for operators are high, limiting their popularity in practical applications. Summary of the invention

[0004] The present invention provides a method and system for detecting mycotoxin contamination in pet food, which are used to solve the defects of the prior art.

[0005] The present invention provides a method for detecting mycotoxin contamination in pet food, comprising: S1: Collect the sample to be tested and obtain the sample to be tested; S2: Analyze the volatile compounds of the sample to be tested to obtain odor data of the sample to be tested; S3: detecting the mycotoxin content in the sample to be tested, and marking the sample to be tested according to the mycotoxin content to obtain marking information; S4: training multiple machine learning models using the odor data and the label information to obtain an integrated prediction model; S5: Perform contamination prediction on the product to be tested by using the integrated prediction model to obtain a prediction result.

[0006] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, step S2 further comprises: S21: collecting volatile compounds of the sample to be tested by electronic nose technology; S22: performing odor analysis on the volatile compound through a sensor array to obtain an odor signal; S23: Convert the odor signal into odor data.

[0007] According to the method for detecting mycotoxin contamination in pet food provided by the present invention, the sensor array in step S22 is provided with 10 sensors.

[0008] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, the volatile compounds in step S2 include: aromatic compounds; nitrogen oxides and ozone; ammonia; hydrogen; alkanes; methane; sulfur compounds; alcohols; sulfur-containing organic compounds; and methane with a concentration greater than 100 ppm.

[0009] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, the machine learning model in step S3 includes: a linear regression model, a support vector machine, a k-nearest neighbor model, a random forest, an extreme gradient boosting model, and a multilayer perceptron.

[0010] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, the mycotoxin in step S3 is zearalenone.

[0011] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, the labeling information in step S3 includes: When the content of zearalenone is greater than When , the sample to be tested is contaminated and marked as 1; When the content of zearalenone is less than or equal to 2 When , the sample to be tested is not contaminated and is marked as 0.

[0012] According to a method for detecting mycotoxin contamination in pet food provided by the present invention, step S4 further comprises: S41: Associating the odor data with the marking information to obtain multiple sets of data including samples to be tested, odor data and corresponding marking information; S42: Inputting the multiple sets of data sets into multiple machine learning models respectively; S43: Perform 5 cross validations for the corresponding machine learning model and data set; S44: Evaluate multiple machine learning models in the same training framework to obtain evaluation results; S45: Perform weighted connection on multiple machine learning models through the evaluation results to obtain an integrated prediction model.

[0013] According to the method for detecting mycotoxin contamination in pet food provided by the present invention, the training framework in step S44 is the Scikit-Learn framework in the Jupyter Notebook environment.

[0014] The present invention also provides a pet food mycotoxin contamination detection system, comprising: Collection module: used to collect samples to be tested and obtain samples to be tested; Analysis module: used to analyze the volatile compounds and mycotoxin content of the sample to be tested, and obtain odor data and marking information; Training module: used to train multiple machine learning models through the odor data and the label information to obtain an integrated prediction model; Detection module: used to predict the contamination of the product to be tested through the integrated prediction model to obtain the prediction result.

[0015] The present invention provides a method and system for detecting mycotoxin contamination in pet food. The electronic nose technology and sensor array are used to accurately analyze the volatile compounds in pet food, and the odor data can be quickly obtained to provide high-quality input information for the subsequent machine learning model training. Compared with the traditional chemical detection method, it is not only faster, but also can capture more subtle pollution characteristics, thereby improving the detection accuracy of mycotoxin contamination. Secondly, the present invention uses a variety of machine learning models for training, and through cross-validation and evaluation under the same training framework, finally constructs an integrated prediction model, which can combine the advantages of each model, reduce the possible deviation of a single model, and further improve the accuracy and robustness of the prediction. In addition, the present invention optimizes the detection process and model parameters, so that the detection method is more targeted and effective, and can more timely and accurately discover and deal with potential safety issues. The entire detection method of the present invention has clear steps and simple operation from sample collection to result output, and the required technology is relatively mature and easy to promote and apply. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention 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 described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of a process for detecting mycotoxin contamination in pet food provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a mycotoxin contamination detection system for pet food provided by an embodiment of the present invention; Figure 3 A schematic diagram of the difference analysis results of sample odor characteristics at different zearalenone (ZEN) pollution levels in the samples to be tested provided by an embodiment of the present invention; Figure 4A schematic diagram of ROC curves of various machine learning models provided in an embodiment of the present invention and their respective weight contributions to the integrated model; Figure 5 Schematic diagram of confusion matrix results for classifying contamination of pet food samples using various machine learning models provided in an embodiment of the present invention.

[0018] Figure numerals: 100, acquisition module; 200, analysis module; 300, training module; 400, detection module. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. 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. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0020] The embodiments of the present invention are described below with reference to the accompanying drawings.

[0021] like Figure 1 As shown, the present invention provides a method for detecting mycotoxin contamination in pet food, comprising: S1: Collect the sample to be tested and obtain the sample to be tested.

[0022] In a specific embodiment, 142 kinds of pet food samples were collected in this embodiment. The specific process of preparing the sample to be tested is: weigh 100g of the sample to be tested, put the weighed sample into a high-speed grinder, and grind it at a speed of 20,000 revolutions per minute until the sample is completely ground into a fine and uniform powder as the sample to be tested.

[0023] S2: Analyze the volatile compounds of the sample to be tested to obtain odor data of the sample to be tested.

[0024] Wherein, step S2 further comprises: S21: Collecting volatile compounds of the sample to be tested by electronic nose technology.

[0025] S22: Performing odor analysis on the volatile compounds through a sensor array to obtain an odor signal.

[0026] Wherein, the sensor array in step S22 is provided with 10 sensors.

[0027] S23: Convert the odor signal into odor data.

[0028] Wherein, the volatile compounds in step S2 include: aromatic compounds; nitrogen oxides and ozone; ammonia; hydrogen; alkanes; methane; sulfur compounds; alcohols; and sulfur-containing organic compounds.

[0029] The specific correspondence between the 10 sensors (R1~R10) and the detected substances is as follows: R1: aromatic compounds; R2: wide sensitivity range, reacts to nitrogen oxides and ozone, very sensitive, and presents a negative signal; R3: ammonia, used as a sensor for aromatic compounds; R4: mainly hydrogen, selective (breath odor); R5: alkanes, aromatic compounds, compounds with weaker polarity; R6: sensitive to methane (environment), about 10 ppm, wide range, similar to R8; R7: reacts to sulfur compounds (H2S 0.1ppm), but is sensitive to many terpenes and sulfur organic compounds (limonene, pyrazine), which are important for smell; R8: detects alcohols and some aromatic compounds; wide detection range; R9: aromatic compounds, sulfur-containing organic compounds; R10: reacts at concentrations greater than 100 ppm, sometimes very selective (methane).

[0030] In a specific embodiment, this stage prepares and electronically analyzes the pet food samples collected in step S1 and evaluates the potential association between volatile compounds and zearalenone (ZEN) contamination.

[0031] First, 142 pet food samples from different brands were crushed into powder. Three subsamples (10 g each) were taken from each sample and placed in a 50-ml round-bottom flask to equilibrate the volatile compounds at 25°C for 12 hours before analysis.

[0032] A portable AIRPEN3 electronic nose equipped with 10 different metal oxide sensors was used to analyze the volatile compounds in the sample. First, the ambient air filtered by activated carbon was introduced into the sensor array at a flow rate of 400 mL / min until the response curve of each sensor was stable (about 60 seconds), and the signal of each sensor was recorded at a frequency of 1 Hz. The output signal of the electronic nose was calculated as The ratio of represents the electronic conductivity of each sensor when detecting the sample odor, Represents the electronic conductivity in clean air. After recording the sample odor signal, air filtered through activated carbon was drawn into the odor channel and chamber again at a flow rate of 400 mL / min for 60 seconds to clean the system and restore the sensor signal to the baseline.

[0033] The specific test results are as follows: 10 volatile compounds were detected in each sample, among which R7 (responsive to sulfur compounds) detected the strongest signal, followed by R9 sensor (aromatic compounds) and R6 sensor (sensitive to methane).

[0034] S3: Detecting the mycotoxin content in the sample to be tested, and marking the sample to be tested according to the mycotoxin content to obtain marking information.

[0035] After step S2, in step S3, the present embodiment analyzes the odor characteristics of 142 pet food samples and divides the samples into two categories: samples labeled 1 (ZEN contamination level exceeds 250 μg / kg) and samples labeled 0 (ZEN contamination level is below this limit).

[0036] Subsequently, the odor data of sensors R1 to R10 obtained by the electronic nose were used to compare the two types of samples in order to explore the relationship between odor characteristics and ZEN pollution levels. The t-test statistical analysis method was used to test the significant differences in the odor characteristics of the two groups of samples in order to identify odor characteristics that may be related to ZEN pollution.

[0037] The specific process of sample treatment was as follows: 25 g of sample was mixed with 100 mL of 50% acetonitrile solution and extracted in an oscillator at 250 rpm for 1 hour; after the extract was filtered, 100 µL of acetic acid was added to 2 mL of the filtrate, and then 750 µL of the mixture was purified using a MycoSpin 400 multifunctional purification column (Romer Labs, Getzersdorf, Austria), and the analytes were quantified by ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS / MS), followed by derivatization; the limit of detection (LOD) was 3 µg / kg, and the limit of quantification (LOQ) was 10 µg / kg.

[0038] For chromatographic separation, a Phenomenex Gemini C18 column (4.6 × 150 mm, 5 μm) was used. The column temperature was set at 40 °C, the flow rate was 1 mL / min, the injection volume was 20 μL, and the mobile phase consisted of 2 mmol / L ammonium acetate and 0.5% acetic acid in water (phase A) and 2 mmol / L ammonium acetate and 0.5% acetic acid in methanol (phase B). The elution conditions were: 90% A and 10% B from 0 to 1 min; 3% A and 97% B from 14 to 15 min; and 90% A and 10% B from 15.1 to 20 min.

[0039] Mass spectrometry detection used electrospray ionization in positive and negative ion modes, with the ion source temperature set at 650 °C in positive ion mode and 600 °C in negative ion mode. The air curtain pressure was maintained at 35 psi, and the collision odor pressure was moderate. The electrospray voltage was set to 5000 V in positive ion mode and -4500 V in negative ion mode, and the pressures of Gas1 and Gas2 were 60 psi and 65 psi, respectively.

[0040] The results are shown in Table 1. Among the 142 pet food samples, ZEN concentrations ranged from undetectable to a maximum of 4133 μg / kg, with a median concentration of 177.5 μg / kg and an average concentration of 392.1 μg / kg (standard deviation = 626.8 μg / kg). It is worth noting that 54 samples (38.0%) exceeded the prescribed limit of 250 μg / kg, while the remaining 88 samples (62.0%) had concentrations below this limit.

[0041]

[0042] Finally, the ZEN contamination level of the sample to be tested was marked. A content greater than 250 μg / kg was marked as "1" to represent contamination, and a content less than or equal to 250 μg / kg was marked as "0" to represent non-contamination.

[0043] The results are as follows Figure 3 As shown, Figure 3 This is a differential analysis of the odor characteristics of samples at different ZEN pollution levels (1: >250μg / kg, 0: ≤250μg / kg). As can be seen from the figure, except for R8, all volatile compounds between the ZEN excessive group and the non-exceeding group were significantly different ( p <0.05), indicating that the volatile compound data captured by the electronic nose can serve as an important indicator to distinguish pet food samples with ZEN contamination above and below the legal threshold.

[0044] S4: Detecting the mycotoxin content in the sample to be tested, and marking the sample to be tested according to the mycotoxin content to obtain marking information.

[0045] Among them, the machine learning model in step S4 includes: linear regression model, support vector machine, k-nearest neighbor model, random forest, extreme gradient boosting model, and multilayer perceptron.

[0046] In order to develop and evaluate a prediction model for pet food sample classification based on volatile compound data and its association with ZEN contamination, six machine learning algorithms were implemented: linear regression (LR), support vector machine (SVM), k-nearest neighbor (KNN), random forest (RF), extreme gradient boosting (XGB) and multi-layer perceptron (MLP). The models were selected based on their robustness and effectiveness in handling complex and multivariate classification tasks.

[0047] Wherein, step S4 further comprises: S41: Associating the odor data with the marking information to obtain multiple data sets including samples to be tested, odor data and corresponding marking information.

[0048] S42: Input the multiple sets of data sets into multiple machine learning models respectively.

[0049] S43: Perform 5 cross validations for the corresponding machine learning model and data set.

[0050] In model training, odor data containing ten variables were used as input to the machine learning model, and each continuous variable was standardized by the "MinMaxScaler" function in the Scikit-Learn library to ensure feature consistency. Specifically, each pet food sample was analyzed three times independently using an electronic nose, and each measurement corresponded to the same class label (1 or 0).

[0051] In order to avoid the electronic nose measurement data of the same pet food sample appearing in both the training set and the validation set, the present invention adopts a cross-validation strategy. In addition, in order to ensure the rigor of model evaluation and reduce the result deviation, a random splitting strategy is adopted. Specifically, the data set is randomly split 5 times, each time into 10 equal parts to ensure the variability between splits. Each time, 90% of the data is used as the training set, and the remaining 10% is used as the test set, that is, n_splits=5, test_size=0.1.

[0052] S44: Evaluate multiple machine learning models within the same training framework to obtain evaluation results.

[0053] The training framework in step S44 is the Scikit-Learn framework in the Jupyter Notebook environment.

[0054] In addition, in order to realize an integrated model combining predictions of multiple algorithms, in order to further improve the classification accuracy by leveraging the advantages of different models, the overall performance is improved by reducing the variance and bias inherent in a single model. All models in step S4 of the present invention are implemented in the Scikit-Learn framework in the Jupyter Notebook environment, and the hyperparameters of each model are adjusted by cross-validation to optimize the performance and ensure the best balance between bias and variance. The evaluation of model performance is based on standard classification indicators, including accuracy, precision, recall, and F1 value. A comprehensive evaluation is also provided through the receiver operating characteristic curve and the area under the curve (AUC-ROC) to evaluate their predictive ability in classifying ZEN contaminated pet food samples.

[0055] The results showed that as shown in Table 2 , pet food samples were classified according to the ZEN contamination level (250 µg / kg) using six machine learning algorithms, including logistic regression (LR), k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and multi-layer perceptron (MLP).

[0056] Each model performed well on the volatile compound data, with classification accuracy ranging from 77.1% to 86.6%. Among them, the MLP classifier showed the highest individual accuracy of 86.6%, F1 score of 79.8%, recall of 83.8%, and precision of 77.9%. The ensemble model combining the predictions of all classifiers further enhanced the classification performance, reaching the highest accuracy of 90.1% and precision of 92.3%.

[0057]

[0058] S45: Perform weighted connection on multiple machine learning models through the evaluation results to obtain an integrated prediction model.

[0059] in addition, Figure 4 The ROC curves in the figure show the performance of six individual models, including logistic regression (LR), k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost, XGB) and multi-layer perceptron (MLP), as well as the integrated model. The integrated model is Figure 4 The part marked with Mean Ensemble indicates the ensemble mean. Figure 4 The ROC curve results of each model are shown in the figure. The vertical axis of each figure is True positive rate, indicating the true positive rate, and the horizontal axis of each figure is False positive rate, indicating the false positive rate.

[0060] pass Figure 4 It can be seen that although the LR model has a lower accuracy (77.1%), its AUC is not the lowest, which indicates that LR still provides important information in distinguishing ZEN-contaminated and uncontaminated samples, contributing a weight of 0.3; in contrast, the best-performing MLP model achieved an accuracy of 86.6% and an AUC of 0.89, but only contributed a weight of 0.15 to the integrated model; other models, such as SVM, k-NN, and XGBoost, contributed weights between 0.15 and 0.2, reflecting their moderate influence. Although RF had moderate individual performance (accuracy of 84.8% and AUC of 0.85 ± 0.09), it did not contribute any weight to the integrated model, indicating that RF's decision path is redundant with other models (such as XGBoost or k-NN) or has limited help in improving the overall discrimination ability of the integrated model. Finally, the ROC curve of the integrated model shows (AUC = 0.90 ± 0.08), and combining these models can achieve the best overall performance, with an accuracy of 90.1% and a precision of 92.3%.

[0061] in addition, Figure 5 The confusion matrix of different machine learning models in pet food sample contamination classification is shown. Figure 5 A in the middle is logistic regression (LR), Figure 5 B is support vector machine (SVM), Figure 5 Where C is k-nearest neighbor (k-NN), Figure 5 D in the figure is random forest (RF), Figure 5 E in the middle stands for extreme gradient boosting (XGB), Figure 5 Where F is a multi-layer perceptron (MLP) and Figure 5 Where G is the integrated model. Figure 5 The Mean Confusion Matrix of the prediction results of each model is shown in the figure. The vertical axis on the right represents the number of samples in the cell (that is, the intersection of a true label and a predicted label), the vertical axis on the left represents Truelable, which represents the true label (including 0, 1), and the horizontal axis represents Predicted lable, which represents the predicted label (including 0, 1).

[0062] In the confusion matrix, positive samples (contaminated, class 1) are above the critical value (250 μg / kg), while negative samples (uncontaminated, class 0) are below the critical value. Each row of the matrix represents the actual class of the sample (true label), and the column corresponds to the predicted class (predicted label). LR ( Figure 5 A) and SVM ( Figure 5 B) shows a tendency to correctly predict negative samples, but poor prediction results for positive samples. k-NN ( Figure 5 C) Moderate performance, with high misclassification rates for both positive and negative classes, RF ( Figure 5 D) XGB Figure 5 E) and MLP ( Figure 5 F) performs better in both categories with relatively high precision and recall, the ensemble model ( Figure 5 G) shows the highest classification accuracy among all models, indicating that the ensemble model correctly predicted 96.7% of negative samples (true negatives) and 76.4% of positive samples (true positives), indicating that its classification performance is balanced and robust.

[0063] S5: Perform contamination prediction on the product to be tested by using the integrated prediction model to obtain a prediction result.

[0064] In step S5, as in the training process, pollution prediction is performed on the product to be tested, including collecting volatile compounds of the product to be tested, performing volatile compound analysis, obtaining odor data of the product to be tested, and then using the integrated prediction model to perform pollution prediction.

[0065] Wherein, the mycotoxin is zearalenone.

[0066] The prediction results in step S5 include: Contamination exists, that is, the predicted binary classification result label is "1", and the content of zearalenone in the contaminated product is greater than ; There is no contamination, that is, the predicted binary classification result label is "0", and the content of zearalenone in the tested product without contamination is less than .

[0067] After the sample collection, odor analysis, and machine learning single model establishment in steps S1 to S4, the integrated prediction model construction can be obtained, and then we use the model for the detection of the product to be tested to complete the mycotoxin contamination of pet food.

[0068] like Figure 2 As shown, the present invention also provides a mycotoxin contamination detection system for pet food, comprising: Collection module 100: used to collect samples to be tested and obtain samples to be tested; Analysis module 200: used to analyze the volatile compounds and mycotoxin content of the sample to be tested to obtain odor data and marking information; Training module 300: used to train multiple machine learning models through the odor data and the label information to obtain an integrated prediction model; Detection module 400: used to predict the contamination of the product to be tested by using the integrated prediction model to obtain a prediction result.

[0069] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0070] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0071] The present invention provides a method and system for detecting mycotoxin contamination in pet food. By analyzing the volatile compounds of the sample and combining with advanced machine learning algorithms, the method can achieve rapid classification and detection of zearalenone (ZEN) contamination. The method not only has high sensitivity and specificity, but also can reduce detection costs and improve detection efficiency. The present invention provides a method for detecting zearalenone (ZEN) contamination in pet food based on machine learning and electronic nose technology. The electronic nose system is used to collect volatile compounds in pet food samples. By analyzing the odor data, the method can achieve rapid identification and classification of ZEN contamination. The trained machine learning model can accurately analyze the collected sample data, so as to quickly determine whether the sample exceeds the safety threshold (250µg / kg). The method not only achieves high sensitivity and high specificity detection of ZEN contamination, but also can complete the analysis in a short time, significantly improving the detection efficiency. Furthermore, the present invention also improves the classification accuracy by integrating the model and combining the performance of multiple machine learning classification models. The method can achieve high-precision classification of samples, optimize the detection process, and reduce the dependence on traditional chemical analysis methods.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting mycotoxin contamination in pet food, characterized in that: include: S1: Collect the sample to be tested and obtain the sample to be tested; S2: Analyze the volatile compounds of the sample to be tested to obtain odor data of the sample to be tested; S3: detecting the mycotoxin content in the sample to be tested, and marking the sample to be tested according to the mycotoxin content to obtain marking information; S4: training multiple machine learning models using the odor data and the label information to obtain an integrated prediction model; S5: Perform contamination prediction on the product to be tested by using the integrated prediction model to obtain a prediction result.

2. The method for detecting mycotoxin contamination in pet food according to claim 1, characterized in that: Step S2 further comprises: S21: collecting volatile compounds of the sample to be tested by electronic nose technology; S22: performing odor analysis on the volatile compound through a sensor array to obtain an odor signal; S23: Convert the odor signal into odor data.

3. The method for detecting mycotoxin contamination in pet food according to claim 2, characterized in that: The sensor array in step S22 is provided with 10 sensors.

4. The method for detecting mycotoxin contamination in pet food according to claim 1, characterized in that: The volatile compounds in step S2 include: aromatic compounds; nitrogen oxides and ozone; ammonia; hydrogen; alkanes; methane; sulfur compounds; alcohols; sulfur-containing organic compounds; and methane with a concentration greater than 100 ppm.

5. The method for detecting mycotoxin contamination in pet food according to claim 1, characterized in that: The machine learning models in step S3 include: linear regression model, support vector machine, k-nearest neighbor model, random forest, extreme gradient boosting model, and multilayer perceptron.

6. The method for detecting mycotoxin contamination in pet food according to claim 1, characterized in that: The mycotoxin in step S3 is zearalenone.

7. The method for detecting mycotoxin contamination in pet food according to claim 6, characterized in that: The marking information in step S3 includes: When the content of zearalenone is greater than When , the sample to be tested is contaminated and marked as 1; When the content of zearalenone is less than or equal to 2 When , the sample to be tested is not contaminated and is marked as 0.

8. The method for detecting mycotoxin contamination in pet food according to claim 1, characterized in that: Step S4 further comprises: S41: Associating the odor data with the marking information to obtain multiple sets of data including samples to be tested, odor data and corresponding marking information; S42: Inputting the multiple sets of data sets into multiple machine learning models respectively; S43: Perform 5 cross validations for the corresponding machine learning model and data set; S44: Evaluate multiple machine learning models in the same training framework to obtain evaluation results; S45: Perform weighted connection on multiple machine learning models through the evaluation results to obtain an integrated prediction model.

9. The method for detecting mycotoxin contamination in pet food according to claim 8, characterized in that: The training framework in step S44 is the Scikit-Learn framework in the Jupyter Notebook environment.

10. A system for detecting mycotoxin contamination in pet food, characterized in that: include: Collection module: used to collect samples to be tested and obtain samples to be tested; Analysis module: used to analyze the volatile compounds and mycotoxin content of the sample to be tested, and obtain odor data and marking information; Training module: used to train multiple machine learning models through the odor data and the label information to obtain an integrated prediction model; Detection module: used to predict the contamination of the product to be tested through the integrated prediction model to obtain the prediction result.

Images