Rapid detection method for food-borne pathogenic bacteria based on interference of meat-derived symbiotic microorganisms

Abstract

The invention relates to a rapid detection method for food-borne pathogenic bacteria based on the action of meat-derived symbiotic microorganisms, belongs to the technical field of rapid detection and spectral analysis of fresh meat quality, and aims to solve the problems of tedious detection operation and high dependency on biochemical reagents caused by interference of a complex substrate and symbiotic microorganisms of fresh meat in the prior art. The method comprises the following steps: firstly, carrying out elution separation-homogeneous centrifugation-coating culture pretreatment on a fresh meat sample to eliminate the interference of a fresh meat matrix, and then preferably selecting a high-discrimination characteristic wave band from high-dimensional spectral data by applying a combined characteristic screening algorithm; and finally, detecting the food-borne pathogenic bacteria of the fresh meat by utilizing a qualitative and quantitative model established on the basis of a culture sample under the interference of the meat-borne symbiotic microorganisms in the earlier stage. The detection accuracy and sensitivity of the food-borne pathogenic bacteria in the fresh meat can be improved, meanwhile, the method has the advantages of being rapid, easy and convenient to operate and economical in cost, and technical support can be provided for detection of the food-borne pathogenic bacteria in the fresh meat.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fresh meat quality rapid detection and spectral analysis, and particularly relates to a foodborne pathogenic bacteria rapid detection method based on meat-derived symbiotic microorganism interference. BACKGROUND

[0002] Foodborne pathogenic bacteria are the main inducement of foodborne diseases, which seriously threaten public health and food safety. Fresh meat, as an important part of daily diet, is easily contaminated by foodborne pathogenic bacteria such as Salmonella, Escherichia coli and Staphylococcus aureus due to its rich nutrients such as protein and water. These pathogenic bacteria can multiply rapidly under suitable conditions and may cause diarrhea, vomiting, fever and even death if ingested by the human body. The existing detection methods of foodborne pathogenic bacteria in fresh meat (such as culture method, molecular biology method and immune analysis method) have limitations such as complicated operation, long time consumption and high dependence on biochemical reagents due to the interference of complex matrix and symbiotic microorganisms in fresh meat. Although spectral imaging technology can effectively shorten the detection time and reduce the dependence on biochemical reagents, the complex matrix of fresh meat will interfere with the spectral signal of the target pathogenic bacteria, resulting in poor sensitivity and accuracy of the model. In addition, the high data dimension and strong redundancy make it difficult for traditional feature selection methods to effectively extract key information under strong interference. Therefore, it has good application prospects to quickly obtain sample information under the interference of symbiotic microorganisms by spectral technology, extract key features by effective methods, and then realize accurate and rapid detection of target pathogenic bacteria under this interference condition. SUMMARY

[0003] The present application proposes a foodborne pathogenic bacteria rapid detection method based on meat-derived symbiotic microorganism interference, which can obtain the culture sample under the interference of meat-derived symbiotic microorganisms by pretreating the fresh meat sample to be detected through "elution separation-homogenization centrifugation-coating culture", and optimize the characteristic wavelength related to the identification of foodborne pathogenic bacteria by using the joint importance feature selection method, so as to solve the problems of complicated detection operation and high dependence on biochemical reagents caused by the interference of complex matrix and symbiotic microorganisms in fresh meat in the prior art.

[0004] The technical scheme adopted by the present application is as follows:

[0005] First, a spectral detection model of foodborne pathogenic bacteria based on meat-derived symbiotic microorganism interference is established, and then the prediction model is used for rapid detection of foodborne pathogenic bacteria in fresh meat. The establishment of the spectral detection model of foodborne pathogenic bacteria based on meat-derived symbiotic microorganism interference specifically comprises the following steps: Step1: Preparation of culture sample under the interference of meat-derived symbiotic microorganisms: First, the fresh meat is pretreated by "washing, separating, and homogenizing centrifugation" to obtain meat-derived symbiotic microorganisms, and then the meat-derived symbiotic microorganisms are confirmed to be free of target pathogenic bacteria. Then, the meat-derived symbiotic microorganisms free of target pathogenic bacteria are mixed with target foodborne pathogenic bacteria and inoculated into a universal culture medium for constant temperature culture to prepare a culture sample under the interference of meat-derived symbiotic microorganisms. Step2: Collecting the spectral image of the culture sample under the interference of meat-derived symbiotic microorganisms and performing black and white correction; Step3: Using the generated mask and morphological method to obtain the region of interest (ROI) of the culture sample under the interference of meat-derived symbiotic microorganisms, and then extracting the one-dimensional average spectral data of the ROI; Step4: Using the concentration gradient stratified sampling method, the one-dimensional spectral data is divided into a training set and a validation set; Step5: Performing SG+2D preprocessing on the divided one-dimensional spectral data of the culture sample; Step6: Extracting effective feature bands from the preprocessed one-dimensional spectral data using a joint importance feature selection method. The joint importance feature selection method includes mutual information entropy (MIE) calculation between features and targets, and feature importance score (FIS) calculation based on the gradient boosting framework. Then, a joint feature importance map is drawn based on MIE and FIS, and features are optimized according to the joint feature importance map; Step7: Establishing a fresh meat foodborne pathogenic bacteria qualitative and quantitative detection model based on the joint importance feature selection method optimized spectral feature signal, using the training set sample to train the model, and the validation set sample to verify the model. The detection model includes a qualitative discrimination model for foodborne pathogenic bacteria contamination and identification of foodborne pathogenic bacteria species, and a quantitative regression model for foodborne pathogenic bacteria contamination concentration prediction. The qualitative model evaluation index is classification accuracy (Acc), and the quantitative model evaluation index is determination coefficient (R 2 and root mean square error (RMSE). The model includes but is not limited to machine learning models such as SVM, RF, and PLS, and deep learning models such as CNN, RNN, and LSTM; The use of the prediction model for rapid detection of foodborne pathogenic bacteria in fresh meat specifically includes the following steps: Step A: Pretreating the fresh meat sample to be detected by "washing, separating, and homogenizing centrifugation" to obtain a culture sample under the interference of meat-derived symbiotic microorganisms to be detected; StepB: Obtaining the spectral image of the sample to be detected and performing black and white correction; StepC: Extracting one-dimensional average spectral data based on the ROI of the sample to be detected; Step D: SG+2D preprocessing is performed on the spectrum data of the to-be-detected sample, and effective characteristic bands are extracted; Step E: the extracted effective characteristic bands are input into the established fresh meat foodborne pathogenic bacteria qualitative detection model to obtain the category of the culture sample under the interference of the to-be-detected meat symbiotic microorganism: no pathogenic bacteria pollution, Escherichia coli pollution, Salmonella pollution, and Staphylococcus aureus pollution; the extracted effective characteristic bands are input into the established fresh meat foodborne pathogenic bacteria quantitative detection model to obtain the concentration of the foodborne pathogenic bacteria in the culture sample under the interference of the to-be-detected meat symbiotic microorganism: 10 1 CFU / g, 10 2 CFU / g, 10 3 CFU / g, 10 4 CFU / g, 10 5 CFU / g, 10 6 CFU / g, 10 7 CFU / g; in order, until all samples are detected.

[0006] The Step 1 specifically comprises the following steps: Step 1.1: First, weigh 25 g ± 2 g of fresh meat sample into a sterile homogenization sampling bag, and then add 225 mL of diluent to mix with the fresh meat sample, wherein the diluent comprises: phosphate buffered saline solution PBS, 0.85%-0.90% sodium chloride solution, 0.85%-0.90% physiological saline, and preferably phosphate buffered saline solution PBS; Step 1.2: use a patting homogenizer to pat the mixture of the fresh meat sample and the diluent for 1-2 min to prepare a fresh meat sample homogenate; Step 1.3: take 50 mL ± 1 mL of the homogenate into a centrifuge, and centrifuge (8000 rpm, 2 min) the homogenate at low speed to separate meat microorganisms and meat fat and tissue impurities, and take the supernatant after centrifugation to prepare a meat symbiotic microorganism suspension, and the suspension is divided into test tubes; Step 1.4: after plate counting of the meat symbiotic microorganism suspension and the foodborne pathogenic bacteria suspension, 10-fold serial dilution suspensions are prepared, and then the meat symbiotic microorganism suspension and the foodborne pathogenic bacteria suspension with different concentrations are mixed at a ratio of 1:1 to prepare a mixed bacteria suspension; Step 1.5: adopt coating plate method to coat different concentrations of mixed bacteria suspension on solid medium, the solid medium includes: Luria-Bertani medium LB, nutrient agar medium NA, tryptone soya agar medium TSA, plate count agar medium PCA, preferably nutrient agar medium NA; Step 1.6: after the agar medium is solidified, the plate is inverted and placed in a constant temperature incubator for 36℃±1℃ culture for 24h±2h to obtain the culture sample under the interference of meat-derived symbiotic microorganisms.

[0007] The Step 3 and Step C specifically comprise the following steps: Step 3.1: remove the waveband with large noise at the first and last end waveband position; Step 3.2: use the generated mask and morphological method to extract the region of interest, mainly including waveband subtraction, threshold segmentation, corrosion, expansion and other steps to remove the medium background and retain the microbial community; Step 3.3: by statistically averaging the spectral information of the pixels in the ROI, a curve representing the overall spectral characteristics of the region is obtained;

[0008] The Step 6 specifically comprises the following steps: Step 6.1: calculate the mutual information entropy MIE of all characteristic wavelengths according to the "mutual information" criterion; Step 6.2: based on the gradient boosting framework, the feature importance score FIS is calculated by iterative optimization of the objective function; Step 6.3: based on the mutual information entropy MIE and the feature importance score FIS, a feature joint importance map is drawn, which can be divided into four quadrants according to the characteristics of the features: quadrant I is a double-high feature, that is, a key feature consistently identified by the comprehensive method, quadrant III is a double-low feature, that is, a secondary feature consistently identified by the comprehensive method, and quadrants II and IV are divided features with inconsistent evaluation conclusions between the "mutual information" criterion and the gradient boosting framework; Step 6.4: retain the key features in the joint importance map that are consistently identified by the "mutual information" criterion and the gradient boosting framework. Figure I

[0009] The mutual information entropy MIE includes: ① Maximum correlation mutual information entropy: calculate the maximum correlation mutual information entropy MIE that measures the dependence of a single feature (wavelength) x i on the target variable Y, the target variable Y is whether contaminated and the category of contaminated foodborne pathogenic bacteria in the classification task of foodborne pathogenic bacteria species classification, and the actual concentration of contaminated foodborne pathogenic bacteria in the regression task of foodborne pathogenic bacteria content prediction, the maximum correlation mutual information entropy calculation formula is as follows: ​where P(x i ,y) is the joint probability distribution of the feature (wavelength) x i and the target variable Y, P(x i ) is the marginal probability distribution of the feature (wavelength) x i , and P(y) is the marginal probability distribution of the target variable Y; ② Maximum Relevance Minimum Redundancy Mutual Information Entropy: the mutual information entropy MIE of each feature (wavelength) is calculated by balancing the maximum relevance I(x i ; Y) and the minimum redundancy I(x i ,x j ), so as to obtain a feature combination highly relevant to the target variable Y but with the least information overlap between each other, and the greater the value of the mutual information entropy MIE, the more important the feature. The calculation formula of the maximum relevance minimum redundancy mutual information entropy is as follows: where S is the feature set, |S| is the number of features, I(x i ; Y) is the maximum relevance information entropy, which is calculated as in ①, and I(x i ,x j ) is the minimum redundancy, and the calculation formula is as follows: where P(x i ,x j ) is the joint probability distribution of the feature (wavelength) x i and the feature (wavelength) x j , P(x i ) is the marginal probability distribution of the feature (wavelength) x i , and P(x j ) is the marginal probability distribution of the feature (wavelength) x j ; ③ Conditional Maximum Mutual Information Entropy: the feature (wavelength) x i that is still relevant to the target variable Y under the condition of the selected features is preferentially selected, and the calculation formula of the conditional maximum mutual information entropy is as follows: MIE(x i ; Y | x j ) = H(x i ; x j ) + H(Y; x j ) - H(x i ; Y; x j ) - H(x j ) where H(x i ; x j ) is the joint entropy of the candidate wavelength x i and the selected wavelength x j , and H(Y; x j) is the joint entropy of the target variable Y and the selected waveband x j , H(x i ; Y; x j ) is the joint entropy of the candidate waveband x i , the target variable Y and the selected waveband x j , and H(x j ) is the entropy of the selected waveband x j .

[0010] The gradient boosting framework-based evaluation of feature importance is: by calculating the contribution of each feature (wavelength) x i in tree splitting, a feature importance score FIS is generated, FIS reflects the total contribution of the feature to reducing the loss in all trees, and the higher the score, the more important the feature, and the gradient boosting framework includes: LightGBM and GradientBoosting, and the FIS calculation formula is as follows: Wherein, m is the number of tree splitting, and Delta L k,p is the loss reduction amount brought by the pth splitting in the kth tree.

[0011] Compared with the prior art, the present application has the following advantages: First, it can accurately identify and quantify foodborne pathogenic bacteria under the interference of meat-derived symbiotic microorganisms, significantly improve the efficiency and sensitivity of detection, and has the advantages of rapidity, simplicity and economy, thereby providing a new strategy and idea for rapid detection of foodborne pathogenic bacteria in fresh meat. Second, the joint importance feature selection algorithm is combined to efficiently extract spectral features with low redundancy, high discrimination and global representation, avoiding the limitations of single index evaluation of feature importance, and significantly improving the usability of spectral data. Third, the fresh meat sample is pretreated by "washing separation-homogenization centrifugation-coating culture", which effectively eliminates the interference of fresh meat matrix and physical impurities, significantly improves the detection efficiency of foodborne pathogenic bacteria, and simplifies the overall detection process. Fourth, the present application provides theoretical support and technical support for improving the intelligent detection technology level of fresh meat, and has important practical significance for ensuring the quality and safety of fresh meat and protecting the legitimate rights and interests of consumers. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is a schematic diagram of the method implementation process of the present application

[0013] Figure 2 is a schematic diagram of the preparation of the culture sample under the interference of meat-derived symbiotic microorganisms according to the embodiment of the present application

[0014] Figure 3This is a spectral curve of a culture sample subjected to interference from meat-derived symbiotic microorganisms, as described in Part a of this invention, after SG+2D pretreatment.

[0015] Figure 4 This is a full-band distribution diagram of the mutual information entropy (MIE) involved in Part a of Embodiment a of the present invention.

[0016] Figure 5 The Feature Importance Score (FIS) distribution map across all bands involved in Part a of Embodiment a of this invention.

[0017] Figure 6 This is a diagram illustrating the joint importance of embodiments a and b.

[0018] Figure 7 The mutual information entropy (MIE) and feature importance score (FIS) of the feature wavelengths preferred by the joint importance feature selection method involved in Embodiment a of this invention are described in Part a.

[0019] Figure 8 This is a schematic diagram showing the distribution of the characteristic wavelengths selected by the joint importance feature selection method in the entire wavelength band, as described in Part a of Embodiment a of the present invention.

[0020] Figure 9 This is a comparison chart of the detection performance of PLS-DA and SVM models involved in Part b of this embodiment of the invention. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this does not limit the scope of the present invention.

[0022] In this embodiment of the invention, ATCC series standard strains are used to prepare culture samples under the interference of meat-derived symbiotic microorganisms. Using standard strains can ensure that the experiment can be reproduced, and the biological characteristics of the strains are known, which facilitates comparison with existing methods.

[0023] Specifically, in this embodiment, the names and types of the foodborne pathogens used are: Escherichia coli ATCC 25922, Salmonella Typhimurium ATCC 14028, and Staphylococcus aureus ATCC 25923; the fresh meat used is mutton.

[0024] like Figure 1 As shown, a rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms includes the following steps: first, establishing a spectral detection model for foodborne pathogens based on interference from meat-derived symbiotic microorganisms, and then using the prediction model to rapidly detect foodborne pathogens in fresh meat;

[0025] Part a: Establish a spectrum detection model of foodborne pathogenic bacteria based on interference of meat-derived symbiotic microorganisms

[0026] a1: First, the sheep meat is pretreated by "washing and separating-homogenizing centrifugation" to obtain sheep-derived symbiotic microorganisms, and then the sheep-derived symbiotic microorganisms are subjected to non-target pathogenic bacteria confirmation. Then, the sheep-derived symbiotic microorganisms without target pathogenic bacteria are mixed with target foodborne pathogenic bacteria and inoculated into universal culture medium to prepare culture samples under the interference of sheep-derived symbiotic microorganisms after constant temperature culture.

[0027] Specifically, in this embodiment, the non-target pathogenic bacteria verification of sheep-derived symbiotic microorganisms is to confirm that the sheep samples are not contaminated with Escherichia coli, Salmonella and Staphylococcus aureus by polymerase chain reaction, so as to ensure the accuracy of the target variable Y.

[0028] Specifically, in this embodiment, the culture sample preparation step under the interference of sheep-derived symbiotic microorganisms is as shown in Figure 2 , and specifically as follows:

[0029] a1.1: First, 25g±2g of sheep sample is weighed and placed in a sterile homogenization sampling bag, and then 225mL of diluent PBS is added to mix with the sheep sample.

[0030] a1.2: The mixture of sheep sample and PBS is beaten for 1min~2min by using a beating homogenizer to prepare a sheep sample homogenate.

[0031] a1.3: 50mL±1mL of homogenate is taken and placed in a centrifuge, and the homogenate is centrifuged at low speed (8000rpm, 2min) to separate the sheep-derived microorganisms from meat fat and tissue impurities. After centrifugation, the supernatant is taken to prepare a sheep-derived symbiotic microorganism suspension, and the suspension is divided into test tubes.

[0032] a1.4: After plate counting of the sheep-derived symbiotic microorganism suspension and the foodborne pathogenic bacteria suspension, 10-fold serial dilution suspensions are prepared, and then the sheep-derived symbiotic microorganism suspension and the foodborne pathogenic bacteria suspension of different concentrations are mixed at a ratio of 1:1 to prepare a mixed bacteria suspension.

[0033] Specifically, in this embodiment, the mixed bacteria suspension is composed of three pathogenic bacteria, Escherichia coli, Salmonella and Staphylococcus aureus, and the sheep-derived symbiotic microorganisms.

[0034] a1.5: Different concentrations of mixed bacteria suspension are uniformly coated on NA culture medium.

[0035] a1.6: After the coating of the medium is solidified, the plate is inverted and placed in an incubator at 36°C ± 1°C for 24h ± 2h to obtain the culture sample under the interference of the symbiotic microorganisms from mutton.

[0036] a2: Collect the spectral image of the culture sample under the interference of the symbiotic microorganisms from mutton and perform black and white correction.

[0037] Specifically, in this embodiment, the spectral image acquisition system is a push-broom short-wave-near-infrared hyperspectral imaging system SWIR-HSI, which mainly consists of a spectrometer (ImSpectorN25E 2 / 3, Specim Company, Finland), an electric displacement platform, a surface array light source, a dark box, a computer, and the like. The SWIR-HSI system covers 288 wavebands in the range of 907.92-2711.59 nm, with a spectral resolution of 12 nm and a sampling interval of 5.6 nm.

[0038] Specifically, in this embodiment, the SWIR-HSI system acquisition parameters are as follows: displacement table moving distance: 140 mm, moving speed: 54.8189 mm / s, exposure time: 4.5 ms, sample-to-camera lens distance: 28.3 cm.

[0039] Specifically, in this embodiment, the hyperspectral image black and white correction file is obtained by collecting the hyperspectral image with the lens closed, and the white correction file is obtained by collecting the hyperspectral image of a standard white plate. The black and white correction calculation formula is as follows: wherein, I R is the hyperspectral image after black and white correction, I is the original hyperspectral image, I W is the white correction hyperspectral image, I D is the black correction hyperspectral image.

[0040] a3: Select the region of interest ROI of the culture under the interference of the symbiotic microorganisms from mutton using the generated mask and morphological method, and then extract the one-dimensional spectral data of the ROI;

[0041] Specifically, in this embodiment, the a3 specifically comprises the following steps:

[0042] a3.1: Remove the wavebands with relatively large noise at the first and last ends. Specifically, in this embodiment, 64 wavebands with relatively large noise at the first and last ends are removed, and 224 wavebands in the range of 1002.15-2472.69 nm are retained

[0043] a3.2: Extract the ROI using the generated mask and morphological method, which mainly includes the steps of waveband subtraction, threshold segmentation, corrosion, and expansion to remove the medium background and retain the microbial community.

[0044] a3.3: Obtain the curve representing the overall spectral characteristics of the region by statistically averaging the spectral information of the pixels within the ROI;

[0045] Specifically, in this embodiment, 1308 hyperspectral images of culture samples under the interference of meat-derived symbiotic microorganisms were collected, and 1308 one-dimensional spectral data were extracted in total for establishing a detection model of foodborne pathogenic bacteria in fresh mutton based on spectral data.

[0046] a4: The one-dimensional spectral data were divided into a training set (n=916) and a validation set (n=392) using the concentration gradient stratified sampling method, and the specific distribution is shown in Table 1:

[0047] Table 1 Data set division

[0048] a5: The one-dimensional spectral data of the divided culture samples under the interference of meat-derived symbiotic microorganisms were subjected to SG+2D pretreatment, and the one-dimensional spectral curve after pretreatment is shown in Figure 3 .

[0049] a6: Effective feature bands were extracted using a joint importance feature selection method;

[0050] Specifically, in this embodiment, the joint importance feature selection method is based on the maximum correlation minimum redundancy mutual information entropy and the LightGBM gradient boosting framework, and the a6 specifically comprises the following steps:

[0051] a6.1: The mutual information entropy MIE of the features was calculated according to the "maximum correlation-minimum redundancy" criterion, and in this embodiment, the mutual information entropy MIE of all feature wavelengths in the full wavelength distribution is shown in Figure 4 .

[0052] Specifically, in this embodiment, the mutual information entropy MIE calculation includes the following steps:

[0053] a6.1.1: Calculate the maximum correlation I(x i ; Y) that measures the degree of dependence of a single feature (wavelength) x i on the target variable Y, which is the class label of the foodborne pathogenic bacteria in the classification task of foodborne pathogenic bacteria species differentiation, and the actual concentration of the pathogenic bacteria in the regression task of foodborne pathogenic bacteria content prediction, and the calculation formula of the maximum correlation is as follows: where P(x i , y) is the joint probability distribution of the feature (wavelength) x i and the target variable Y, P(x i) is the marginal probability distribution of feature (wavelength) x i P(y) is the marginal probability distribution of target variable Y;

[0054] a6.1.2: Calculate the minimum redundancy I(x i ,x j ) which measures the degree of information overlap between feature (wavelength) x i and feature (wavelength) x j , the formula is as follows: Where P(x i ,x j ) is the joint probability distribution of feature (wavelength) x i and feature (wavelength) x j , P(x i ) is the marginal probability distribution of feature (wavelength) x i , P(x j ) is the marginal probability distribution of feature (wavelength) x j ;

[0055] a6.1.3: Calculate the mutual information entropy MIE of each feature (wavelength) by balancing the maximum correlation I(x i ; Y) and the minimum redundancy I(x i ,x j ), get the feature combination which is highly related to the target variable Y but has the least information overlap with each other, and the greater the MIE value, the more important the feature, the formula is as follows: Where S is the feature set, and |S| is the number of features.

[0056] a6.2: Evaluate feature importance based on the LightGBM gradient boosting framework, calculate feature importance score FIS by iterative optimization of objective function, in this embodiment, the feature importance score FIS of all feature wavelengths in the full wavelength distribution is shown in Figure 5 .

[0057] Specifically, in this embodiment, the evaluation of feature importance based on the LightGBM gradient boosting framework is: by calculating the contribution of each feature (wavelength) x i in tree splitting, generate feature importance score FIS, FIS reflects the total contribution of the feature to reducing the loss in all trees, and the higher the score, the more important the feature, the formula is as follows: Where m is the number of tree splits, and ΔL k,pis the loss reduction amount brought by the pth split in the kth tree.

[0058] a6.3: Draw feature joint importance map based on MIE and FIS Figure 6

[0059] a6.4: Retain joint importance Figure I key features within the quadrant;

[0060] Specifically, in this embodiment, the number of preferred feature wavelengths based on the joint importance feature selection method is 12, which are: 1190.63 nm, 1310.01 nm, 1372.84 nm, 1536.22 nm, 1737.31 nm, 1837.86 nm, 1875.57 nm, 1988.70 nm, 2001.27 nm, 2020.1 nm, 2057.83 nm, and 2290.40 nm. The MIE and FIS of the 12 key features are shown in Figure 7 , and the full-band distribution is shown in Figure 8 .

[0061] a7: Train and verify the detection model of food-borne pathogenic bacteria in mutton based on the joint importance feature selection method preferred spectral features;

[0062] Specifically, in this embodiment, PLS-DA and SVM models are used, and the model evaluation index is classification accuracy ACC, wherein the number of principal components n_components of PLS-DA is 7, the penalty factor c of SVM is 26.1364, and the kernel function gamma is 0.3236. The classification accuracy of the classification model in the training set and the validation set is shown in Table 2:

[0063] Table 2 Model classification results

[0064] Part b: The detection process of food-borne pathogenic bacteria in mutton using the established PLS-DA and SVM prediction model is performed in the order of steps b1-b6;

[0065] b1: The mutton sample to be detected is pretreated by "elution separation-homogenization centrifugation-coating culture" to obtain a culture sample under the interference of mutton-derived symbiotic microorganisms to be detected;

[0066] b2: Obtain the spectral image of the sample to be detected and perform black and white correction;

[0067] b3: Extract one-dimensional average spectral data based on the ROI of the sample to be detected;

[0068] b4: Perform SG+2D preprocessing on the spectral data of the sample to be detected and extract effective feature bands;

[0069] ​b6: input 1190.63 nm, 1310.01 nm, 1372.84 nm, 1536.22 nm, 1737.31 nm, 1837.86 nm, 1875.57 nm, 1988.70 nm, 2001.27 nm, 2020.1 nm, 2057.83 nm, 2290.40 nm, 12 key characteristic bands into the established PLS-DA and SVM prediction model detection model, and obtain the sample categories of the culture sample under the interference of the sheep meat source symbiotic microorganism to be detected: no pathogenic bacteria pollution, Escherichia coli pollution, Salmonella pollution and Staphylococcus aureus pollution, and sequentially detect until all samples are detected.

[0070] At the same time, in order to compare the detection effect of the model, the sample category label determined based on the PCR method is used to evaluate the feature selection method and the model detection effect. A total of 132 samples were prepared for model detection, including 33 samples of non-pathogenic bacteria pollution, 33 samples of Escherichia coli pollution, 33 samples of Salmonella pollution and 33 samples of Staphylococcus aureus pollution. The classification model accuracy of the established PLS-DA and SVM model is 0.9469 and 0.9621 respectively, and the detection effect comparison is shown in the following table. Figure 9 According to the above results, the combined importance feature selection algorithm can effectively screen the key spectral features, the extracted features have low redundancy, high discrimination and global representativeness, and significantly improve the reliability and usability of the spectral data; at the same time, under the interference of meat source symbiotic microorganisms, efficient, accurate and economical identification and detection of target pathogenic bacteria in fresh meat is realized, which effectively overcomes the limitations of the prior art and shows good application prospect.

[0071] When performing qualitative and quantitative detection of other microorganisms in fresh meat based on spectral image technology, the detection method and detection process proposed in the present application can be referred to for operation.

[0072] The above embodiments are only used to illustrate the present application, and do not limit the present application. Any modification, equivalent replacement, improvement made within the spirit and principles of the present application on the basis of the technical essence of the present application shall be included in the scope of the present application, and the patent protection scope of the present application is defined by the claims.

Claims

1. A rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms, characterized in that, First, a spectral detection model for foodborne pathogens based on interference from meat-derived symbiotic microorganisms was established, and then the prediction model was used to rapidly detect foodborne pathogens in fresh meat. The establishment of a spectral detection model for foodborne pathogens based on interference from meat-derived symbiotic microorganisms specifically includes the following steps: Step 1: Preparation of culture samples under interference from meat-derived symbiotic microorganisms: First, the fresh meat is pretreated by "elution-homogenization centrifugation" to obtain meat-derived symbiotic microorganisms. Then, the meat-derived symbiotic microorganisms without target pathogens are identified. Then, the meat-derived symbiotic microorganisms without target pathogens are mixed with the target foodborne pathogens and inoculated into a general culture medium. After incubation at constant temperature, culture samples under interference from meat-derived symbiotic microorganisms are prepared. Step 2: Acquire spectral images of culture samples under interference from meat-derived symbiotic microorganisms and perform black-and-white correction; Step 3: Use generation mask and morphological methods to obtain the region of interest (ROI) of the culture samples under interference from meat-derived symbiotic microorganisms, and then extract the one-dimensional average spectral data of the ROI; Step 4: Use the concentration gradient stratified sampling method to divide the one-dimensional spectral data into a training set and a validation set; Step 5: Preprocess the one-dimensional spectral data of the divided culture samples to reduce noise, correct drift, eliminate scattering effects, improve the usability of spectral data and the reliability of analysis results. The preprocessing is a combination spectral preprocessing method of smoothing the second derivative SG+2D. Step 6: Extract effective feature bands from the preprocessed one-dimensional spectral data using the joint importance feature selection method. The joint importance feature selection method consists of two parts: one is to calculate the mutual information entropy (MIE) between the feature and the target, and the other is to calculate the feature importance score (FIS) based on the gradient boosting framework. Then, a joint feature importance map is drawn based on the MIE and FIS, and features are selected according to the joint feature importance map. Step 7: Establish a qualitative and quantitative detection model for foodborne pathogens in fresh meat based on a joint importance feature selection method to optimize spectral feature signals. The model is trained using a training set and validated using a validation set. The detection model involved in this invention includes: a qualitative discriminant model for identifying foodborne pathogen contamination and the types of contaminated foodborne pathogens, and a quantitative regression model for predicting foodborne pathogen contamination concentrations. The evaluation index for the qualitative model is classification accuracy (Acc), and the evaluation index for the quantitative model is the coefficient of determination (R²). 2 The root mean square error (RMSE) and the models involved in this invention include, but are not limited to, machine learning models SVM, RF and PLS, and deep learning models CNN, RNN and LSTM. The method of using a predictive model for rapid detection of foodborne pathogens in fresh meat: Step A: The fresh meat sample to be tested is pretreated by "elution-homogenization-spreading culture" to obtain the culture sample under the interference of meat-derived symbiotic microorganisms to be tested; Step B: Acquire the spectral image of the sample to be detected and perform black and white correction; Step C: Extract one-dimensional average spectral data based on the ROI of the sample to be tested; Step D: Perform SG+2D preprocessing on the spectral data of the sample to be tested and extract effective feature bands; Step E: Input the effective characteristic bands into the established qualitative detection model for foodborne pathogens in fresh meat to obtain the categories of culture samples under interference from meat-derived symbiotic microorganisms: no pathogenic bacteria contamination, Escherichia coli contamination, Salmonella contamination, and Staphylococcus aureus contamination; input the effective characteristic bands into the established quantitative detection model for foodborne pathogens in fresh meat to obtain the concentration of foodborne pathogens contaminating the culture samples under interference from meat-derived symbiotic microorganisms: 10 1 CFU / g, 10 2 CFU / g, 10 3 CFU / g, 10 4 CFU / g, 10 5 CFU / g, 10 6 CFU / g, 10 7 CFU / g; test sequentially until all samples have been tested.

2. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1.1: First, weigh 25g ± 2g of fresh meat sample and place it in a sterile homogenization sampling bag. Then, add 225mL of diluent and mix it with the fresh meat sample. The diluent includes: phosphate buffered saline (PBS), 0.85%-0.90% sodium chloride solution, and 0.85%-0.90% physiological saline, preferably phosphate buffered saline (PBS). Step 1.2: Use a tapping homogenizer to tap the mixture of fresh meat sample and diluent for 1 to 2 minutes to prepare a fresh meat sample homogenate; Step 1.3: Take 50mL±1mL of homogenate and place it in a centrifuge. Centrifuge at low speed (8000rpm, 2min) to separate meat-derived microorganisms from meat fat and tissue impurities. After centrifugation, take the supernatant to prepare a meat-derived symbiotic microbial suspension and dispense the suspension into test tubes. Step 1.4: After plate counting of meat-derived microbial suspension and foodborne pathogenic bacteria suspension, prepare 10-fold serial dilution suspensions. Then, mix meat-derived symbiotic microbial suspensions and foodborne pathogenic bacteria suspensions of different concentrations at a 1:1 ratio to prepare mixed bacterial suspensions. Step 1.5: Using the plating method, the mixed bacterial suspensions of different concentrations are evenly spread on solid culture media, including: Luria-Bertani medium LB, nutrient agar medium NA, tryptone soybean agar medium TSA, and plate counting agar medium PCA, preferably nutrient agar medium NA; Step 1.6: After the agar medium has solidified, the plates are inverted and placed in a constant temperature incubator at 36℃±1℃ for 24h±2h to obtain culture samples under the interference of meat-derived symbiotic microorganisms.

3. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 1, characterized in that, Step 6 specifically includes the following steps: Step 6.1: Calculate the mutual information entropy (MIE) for all characteristic wavelengths based on the "mutual information" criterion; Step 6.2: Based on the gradient boosting framework, calculate the feature importance score (FIS) by iteratively optimizing the objective function; Step 6.3: Draw a joint importance graph of features based on mutual information entropy (MIE) and feature importance score (FIS). The joint importance graph can be divided into four quadrants according to the characteristics of the features: Quadrant I represents key features consistently recognized by the comprehensive method; Quadrant III represents secondary features consistently judged by the comprehensive method; Quadrants II and IV contain discrepancies between the "mutual information" criterion and the evaluation conclusion of the gradient boosting framework. Step 6.4: Retain the key features identified in the joint importance graph quadrant I that are consistent with the gradient boosting framework.

4. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 3, characterized in that, The mutual information entropy MIE includes: ① Maximum correlation mutual information entropy: Calculates the feature (wavelength) x i The maximum correlation mutual information entropy (MIE) is the degree of association between the target variable Y and the target variable Y. In the classification task of distinguishing foodborne pathogens, Y represents whether the food is contaminated and the type of contaminated foodborne pathogens. In the regression task of predicting the content of foodborne pathogens, Y represents the actual concentration of contaminated foodborne pathogens. The formula for calculating the maximum correlation mutual information entropy is as follows: Wherein, P(x i ,y) is the characteristic (wavelength) x i The joint probability distribution of the target variable Y with the target variable, P(x) i ) is the characteristic (wavelength) x i The marginal probability distribution of the target variable Y, P(y) is the marginal probability distribution of the target variable Y. ② Maximum correlation and minimum redundancy mutual information entropy: By balancing the maximum correlation I(x) i ;Y) and minimum redundancy I(x) i ,x j Calculate the mutual information entropy (MIE) for each feature (wavelength) to obtain the feature combination that is highly correlated with the target variable Y but has the least information overlap with each other. The larger the MIE value, the more important the feature. The formula for calculating the maximum correlation and minimum redundancy MIE is as follows: Where S is the feature set, |S| is the number of features, and I(x) i ;Y) is the maximum relevant entropy, calculated in the same way as ①, I(x) i ,x j The minimum redundancy is calculated using the following formula: Wherein, P(x i ,x j ) is the characteristic (wavelength) x i With characteristic (wavelength) x j The joint probability distribution, P(x) i ) is the characteristic (wavelength) x i Marginal probability distribution, P(x) j ) is the characteristic (wavelength) x j The marginal probability distribution; ③Conditionally maximize mutual information entropy: Prioritize features (wavelengths) x that are still relevant to the target variable Y under the selected feature conditions. i The formula for calculating conditionally maximizing mutual information entropy is as follows: MIE(x i ;Y|x j )=H(x i ;x j )+H(Y;x j )-H(x i ;Y;x j )-H(x j ) Wherein, H(x) i ;x j ) represents candidate band x i With the selected band x j The joint entropy, H(Y; x) j Let Y be the target variable and x be the selected band. j The joint entropy, H(x) i ;Y;x j ) represents candidate band x i Target variable Y, selected band x j The joint entropy of the three, H(x) j (x) represents the selected band. j Its own entropy.

5. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 3, characterized in that, The gradient boosting framework evaluates feature importance by calculating the x-value of each feature (wavelength). i The contribution to tree splitting is used to generate a Feature Importance Score (FIS). The FIS reflects the total contribution of a feature to loss reduction across all trees, with higher scores indicating more important features. The gradient boosting framework includes LightGBM and GradientBoosting. The FIS calculation formula is as follows: Where m is the number of tree splits, ΔL k,p It represents the decrease in loss caused by the p-th split in the k-th tree.

6. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 1, characterized in that, The fresh meat includes: lamb, beef, pork and chicken.

7. The rapid detection method for foodborne pathogens based on interference from meat-derived symbiotic microorganisms according to claim 1, characterized in that, Step 3 and Step C specifically include the following steps: Step 3.1: Remove the bands with high noise at the beginning and end of the band position; Step 3.2: Use mask generation and morphological methods to extract the region of interest, mainly including steps such as band subtraction, thresholding, erosion, and dilation to remove the culture medium background and retain the microbial community; Step 3.3: Obtain a curve representing the overall spectral characteristics of the region by statistically averaging the spectral information of the pixels within the ROI.

Images