Mutton origin identification method, system, medium and product

Abstract

The application relates to the technical field of food detection, and discloses a mutton origin identification method, a mutton origin identification system, a medium and a product. The method comprises the following steps: collecting mass spectrum data of a to-be-detected mutton sample; extracting to-be-detected feature data from the mass spectrum data according to a candidate mass-to-charge ratio feature set, wherein the candidate mass-to-charge ratio feature set is determined based on mass spectrum data of an annotated mutton sample with an origin label; and performing first-level origin identification according to the to-be-detected feature data and an origin identification model trained based on the annotated mutton sample, so as to determine that the to-be-detected mutton sample belongs to Hulun Buir mutton or non-Hulun Buir mutton.

Description

Technical Field

[0001] This application relates to the field of food testing technology, and in particular to a method, system, medium and product for identifying the origin of mutton. Background Technology

[0002] With the increasing demand for food traceability and quality control, accurate identification of the origin of meat products is of great significance. The compositional characteristics of mutton are affected by factors such as the production environment and feeding conditions. Mutton from different origins may differ in some chemical components and their characterization data. Therefore, the origin of mutton can be analyzed and identified based on the test data of mutton samples.

[0003] In existing technologies, mutton samples from different origins are typically subjected to component analysis, spectral analysis, or mass spectrometry analysis to obtain detection data that characterizes the composition of the mutton samples. Origin-related feature information is then selected from this detection data. Furthermore, an origin discrimination model can be established based on mutton samples with known origin information. This model can then be used to analyze the detection data of the mutton sample to be tested, thereby determining the corresponding origin of the mutton sample.

[0004] However, in the process of identifying the origin of mutton, the degree of difference in origin may vary depending on the identification level. For example, the differences in mutton characteristics between a large geographical area are not the same as the differences in mutton characteristics between different sub-originating areas within the same geographical area. When faced with the above-mentioned different granularity of origin identification needs, existing origin identification methods have difficulty in simultaneously addressing both origin identification over a large geographical area and more granular origin identification within a region, thus affecting the precision of mutton origin identification. Summary of the Invention

[0005] One objective of this application is to provide a method, system, medium, and product for identifying the origin of mutton, at least to address the problem that existing technologies struggle to simultaneously meet the needs for identifying the origin of mutton at different granularities.

[0006] To achieve the above objectives, some embodiments of this application provide the following aspects: In a first aspect, some embodiments of this application also provide a method for identifying the origin of mutton, the method comprising: collecting mass spectrometry data of a mutton sample to be tested; extracting test feature data from the mass spectrometry data of the mutton sample to be tested according to a candidate mass-to-charge ratio feature set; wherein the candidate mass-to-charge ratio feature set is determined based on the mass spectrometry data of multiple labeled mutton samples with origin labels; performing a first-level origin identification on the mutton sample to be tested according to the test feature data and an origin identification model, determining whether the mutton sample to be tested belongs to Hulunbuir mutton or non-Hulunbuir mutton; the origin identification model is trained based on training feature data extracted from the mass spectrometry data of multiple labeled mutton samples according to the candidate mass-to-charge ratio feature set and the corresponding origin labels; The mass spectrometry data is mass spectrometry data collected under the first ion polarity. The target feature data is extracted from the mass spectrometry data according to the first candidate mass-to-charge ratio feature subset. The first candidate mass-to-charge ratio feature subset is the mass-to-charge ratio feature in the candidate mass-to-charge ratio feature set that corresponds to the first ion polarity. The origin identification model adopts a supervised classification model that can output the origin category based on the input features; The method for determining the candidate mass-to-charge ratio feature set includes: Based on the origin label, the mass spectrometry data of multiple labeled mutton samples are grouped. Based on the mass-to-charge ratio characteristic response of each labeled mutton sample in the same group, determine the intra-group fluctuation parameter corresponding to each mass-to-charge ratio characteristic. Based on the differences in the corresponding mass-to-charge ratio characteristic responses between different groups, determine the inter-group discrimination parameters corresponding to each mass-to-charge ratio characteristic; Based on the intra-group fluctuation parameters and the inter-group differentiation parameters, determine the stability differentiation parameters corresponding to each of the mass-to-charge ratio characteristics; Based on the stability differentiation parameter, the candidate mass-to-charge ratio feature set is obtained by screening from multiple mass-to-charge ratio features.

[0007] Secondly, some embodiments of this application also provide a mutton origin identification system, including a mass spectrometry detection device and an electronic device; the mass spectrometry detection device is used to collect mass spectrometry data of mutton samples; the electronic device includes one or more processors and a memory storing a computer program, wherein when the computer program is executed by the processor, the electronic device performs the data processing and origin identification steps as described in any of the above methods based on the mass spectrometry data collected by the mass spectrometry detection device.

[0008] Thirdly, some embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes an electronic device to perform the method described above based on mass spectrometry data collected by a mass spectrometry detection device.

[0009] Fourthly, some embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, causes an electronic device to perform the steps of the method described above based on mass spectrometry data acquired by a mass spectrometry detection device.

[0010] Compared with related technologies, the solution provided in this application embodiment is to characterize the mutton sample to be tested and the training sample based on the corresponding mass-to-charge ratio features, and to process the test feature data using the training feature data and the origin identification model trained with the corresponding origin label, thereby determining whether the mutton sample to be tested belongs to Hulunbuir mutton or not, thus realizing mutton origin identification based on mass spectrometry features. Attached Figure Description

[0011] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0012] Figure 1 An exemplary flowchart of a method for identifying the origin of mutton provided in some embodiments of this application; Figure 2 This is a schematic diagram of an electronic device structure provided for some embodiments of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] Specifically, this application provides a method for identifying the origin of mutton, which can be executed by a mutton origin identification system. The mutton origin identification system may include a mass spectrometry detection device and an electronic device. The mass spectrometry detection device can be used to collect mass spectrometry data from mutton samples, and the electronic device can be used to process the mass spectrometry data and complete the mutton origin identification. Figure 1 As shown, the method may include: Step S101: Collect mass spectrometry data of the mutton sample to be tested; Step S102: Extract the target feature data from the mass spectrometry data of the mutton sample to be tested according to the candidate mass-to-charge ratio feature set; wherein, the candidate mass-to-charge ratio feature set is determined based on the mass spectrometry data of multiple labeled mutton samples with origin labels. Step S103: Based on the test feature data and the origin identification model, perform first-level origin identification on the test mutton sample to determine whether the test mutton sample belongs to Hulunbuir mutton or non-Hulunbuir mutton; the origin identification model is trained based on training feature data extracted from the mass spectrometry data of multiple labeled mutton samples according to the candidate mass-to-charge ratio feature set and the corresponding origin labels. Specifically, in step S101, the mutton sample to be tested can be a mutton sample for which the origin needs to be determined. The mass spectrometry detection equipment can detect the mutton sample to be tested and generate mass spectrometry data corresponding to the mutton sample. The mass spectrometry data can include the mass spectrometry response corresponding to different mass-to-charge ratio positions, and the mass spectrometry response can be represented by peak intensity, peak area, or other data that can characterize the degree of ion response at the corresponding mass-to-charge ratio position.

[0015] In one implementation, the mass spectrometry data output by the mass spectrometry detection device can be recorded according to a preset data format, so that each mutton sample to be tested corresponds to a set of mass-to-charge ratio positions and the mass spectrometry response corresponding to each mass-to-charge ratio position. For different mutton samples to be tested, the mass spectrometry data can be saved in a consistent data format so that the corresponding feature data can be extracted later based on the candidate mass-to-charge ratio feature set.

[0016] In one implementation, probe electrospray ionization mass spectrometry (PSI) can be used to acquire mass spectrometry data of the mutton sample to be tested. A probe is used to sample the mutton sample to form a test sample, and the mass spectrometry detection equipment obtains the mass spectrometry response corresponding to different mass-to-charge ratio positions. Probe electrospray ionization mass spectrometry, as one method of mass spectrometry data acquisition, does not constitute a limitation on the method of mass spectrometry data acquisition.

[0017] In one implementation, a labeled mutton sample library can be established by selecting multiple mutton samples from different production areas, and the production area information corresponding to each mutton sample can be recorded. For Hulunbuir mutton samples, they can be further labeled according to different pastoral banners. Multiple labeled mutton samples can cover mutton samples obtained in different seasons and under different pasture conditions, so as to make the sample sources have a certain degree of diversity, and can be used to determine the candidate mass-to-charge ratio feature set and train the production area identification model.

[0018] When using probe electrospray ionization mass spectrometry for detection, the corresponding sample pretreatment method and mass spectrometry detection conditions can be determined according to the matrix characteristics of the mutton sample. After determining the detection conditions, the labeled mutton sample and the mutton sample to be tested are subjected to mass spectrometry detection according to the same detection procedure.

[0019] Specifically, for step S102, for the mass spectrometry data corresponding to multiple labeled mutton samples, the mass spectrometry data can be organized, features filtered, and features expressed according to a preset data processing flow. Training data for origin identification can then be constructed based on the processed mass spectrometry data. The labeled mutton samples can be those whose origin information has been determined. The origin label is used to characterize the known origin of the corresponding labeled mutton sample. The origin label can include a first-level label to distinguish between Hulunbuir mutton and non-Hulunbuir mutton. For labeled mutton samples belonging to Hulunbuir mutton, the origin label can also include a second-level label to characterize the corresponding pastoral banner.

[0020] The candidate mass-to-charge ratio (M / C ratio) feature set can be understood as a set of M / C ratio features pre-determined from the mass spectrometry data of multiple labeled mutton samples to characterize the differences in mutton origin. Each M / C ratio feature can correspond to a single M / C ratio location, or it can correspond to a preset M / C ratio range centered on the target M / C ratio location.

[0021] Before identifying the mutton samples to be tested, a training sample set can be established based on multiple labeled mutton samples with origin tags, and a candidate mass-to-charge ratio feature set can be determined based on the mass spectrometry data corresponding to the training sample set. After the candidate mass-to-charge ratio feature set is determined, it can serve as the basis for feature extraction used in the subsequent construction of training data and the construction of test data.

[0022] For a mutton sample to be tested, the mass spectrometry response corresponding to each mass-to-charge ratio feature in the candidate mass-to-charge ratio feature set can be found in the mass spectrometry data. The corresponding mass spectrometry responses are then organized according to a preset arrangement order of the candidate mass-to-charge ratio features to form the feature data to be tested. For example, when the candidate mass-to-charge ratio feature set includes multiple mass-to-charge ratio features arranged sequentially, the corresponding response of each mass-to-charge ratio feature in the mass spectrometry data of the mutton sample to be tested can be determined sequentially, and the multiple corresponding responses can be arranged in the same order to form a feature vector to be tested. The feature data to be tested can be represented using the aforementioned feature vector.

[0023] By using the same candidate mass-to-charge ratio feature set for feature representation of the mutton samples to be tested and the labeled mutton samples, it can be ensured that the features to be tested received by the origin identification model correspond to the features used in the model training process.

[0024] In step S103, specifically, the mass spectrometry responses corresponding to the candidate mass-to-charge ratio feature sets can be extracted from the mass spectrometry data of each labeled mutton sample according to the same feature extraction rules as the feature data to be tested, thus obtaining the training feature data corresponding to each labeled mutton sample. The training feature data and the origin labels of the corresponding labeled mutton samples can together constitute the training data of the origin identification model.

[0025] In this embodiment, the origin identification model employs a multilayer perceptron classification model. The multilayer perceptron classification model includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer receives training feature data or test feature data; the first hidden layer includes 32 neurons, and the second hidden layer includes 16 neurons, both employing the ReLU activation function; the output layer includes 2 neurons and employs the Softmax function, outputting the category probabilities corresponding to Hulunbuir mutton and non-Hulunbuir mutton, respectively.

[0026] During model training, the training feature data corresponding to each labeled mutton sample was used as the model input, and the corresponding origin label was used as supervision information. The cross-entropy loss function was used to determine the loss between the output origin category probability and the origin label, and the Adam optimization algorithm was used to update the model parameters. The learning rate was set to 0.001, and the maximum number of training epochs was set to 500. The training samples were stratified according to the origin label, with 80% of the labeled mutton samples used for model parameter training and 20% used for model validation. Training ended when the validation set loss did not decrease for 20 consecutive epochs, and the model parameters corresponding to the lowest validation set loss were saved.

[0027] After training, the trained model can be validated using the validation set accuracy. In this embodiment, the preset usable threshold is set to 85%, and the origin identification model obtained using the above model structure and training method achieves an accuracy of 87.5% on the validation set. When the validation set accuracy reaches the preset usable threshold, the corresponding model is determined as the origin identification model for identifying the mutton samples to be tested; when the validation set accuracy does not reach the preset usable threshold, model training is re-executed. Therefore, before the model is used to identify mutton samples to be tested, its ability to identify the correspondence between mutton mass spectrometry features and origin categories can be verified using samples that were not used in the current model parameter training.

[0028] Since the training feature data is formed by the mass spectrometry responses corresponding to multiple candidate mass-to-charge ratio features, it belongs to continuous numerical features of fixed dimensions. The first-level origin identification is used to determine whether a mutton sample belongs to Hulunbuir mutton or not. Therefore, the multilayer perceptron classification model establishes a mapping relationship between multiple mass-to-charge ratio features and origin categories through hidden layers, and forms the category probabilities corresponding to the two origin categories through the output layer. When identifying the mutton sample to be tested, the feature data to be tested is input into the trained multilayer perceptron classification model, and the category with the higher category probability is determined as the corresponding first-level identification result.

[0029] When performing the first-level origin identification, the feature data to be tested can be input into the origin identification model. Based on the feature data and the feature distribution relationship corresponding to different origin categories learned during the training phase, the origin identification model outputs the first-level identification result for the mutton sample to be tested. The first-level identification result includes either Hulunbuir mutton or non-Hulunbuir mutton.

[0030] Compared with related technologies, in this embodiment, by characterizing the mutton sample to be tested and the training sample based on the corresponding mass-to-charge ratio features, and using the origin identification model trained with the training feature data and the corresponding origin label to process the feature data to be tested, it is possible to determine whether the mutton sample to be tested belongs to Hulunbuir mutton or not, thereby realizing mutton origin identification based on mass spectrometry features.

[0031] Optionally, in some embodiments, the method for determining the candidate mass-to-charge ratio feature set may include: Step S201: Group the mass spectrometry data of multiple labeled mutton samples according to the origin label.

[0032] Specifically, based on the origin labels of the labeled mutton samples, mutton samples with the same origin label can be grouped into the same group. Each group can include multiple mutton samples with the same origin label.

[0033] For each labeled mutton sample, the mass spectrometry data can be organized according to a uniform mass-to-charge ratio (M / C ratio) position, allowing for comparison of the corresponding M / C ratio characteristics among different labeled mutton samples. For any given M / C ratio characteristic, the M / C ratio characteristic responses corresponding to multiple labeled mutton samples within the same group can be obtained.

[0034] Step S202: Based on the mass-to-charge ratio characteristic response of each labeled mutton sample within the same group, determine the intra-group fluctuation parameter corresponding to each mass-to-charge ratio characteristic.

[0035] Specifically, the intragroup variability parameter is used to characterize the degree of dispersion in the response of multiple labeled mutton samples from the same origin to the same mass-to-charge ratio feature. The smaller the intragroup variability, the more stable the response of the corresponding mass-to-charge ratio feature is among samples from the same origin.

[0036] In this embodiment, variance is used to determine the within-group fluctuation parameter. Specifically, it is assumed that the fluctuation parameter is formed based on the country of origin label. Group 1, the 1st group The groups include The first labeled mutton sample, the [number]th The first group The labeled mutton sample was in the first... The mass-charge ratio characteristic response on each mass-charge ratio characteristic is The total number of labeled mutton samples is ,in, This represents the sum of the sample sizes in each group.

[0037] No. The first group Mean of grouped response corresponding to each mass-to-charge ratio feature Determined according to the following formula: ; No. Intragroup fluctuation parameters corresponding to individual mass-to-charge ratio characteristics Determined according to the following formula: ; Therefore, the intragroup fluctuation parameters Labeled mutton samples used to characterize the same place of origin were in the first... The degree of dispersion of the response on the mass-to-charge ratio characteristic The smaller the value, the more stable the response of the corresponding mass-to-charge ratio characteristic is in samples from the same origin.

[0038] Step S203: Determine the inter-group distinction parameter corresponding to each mass-to-charge ratio characteristic based on the differences in the corresponding mass-to-charge ratio characteristic responses between different groups.

[0039] Specifically, the intergroup discrimination parameter is used to characterize the degree of difference in the response of mutton samples from different origins to the same mass-to-charge ratio feature.

[0040] In this embodiment, the first The mean of the overall response corresponding to the individual mass-to-charge ratio characteristics Determined according to the following formula: ; Furthermore, the first Between-group distinguishing parameters corresponding to individual mass-to-charge ratio characteristics Determined according to the following formula: ; Between-group distinguishing parameters The larger the value, the better the mutton sample from different origins is in the 1st rank. The more pronounced the difference in response characteristics on the mass-to-charge ratio, the more significant the difference.

[0041] Step S204: Determine the stability differentiation parameter corresponding to each of the mass-to-charge ratio characteristics based on the intra-group fluctuation parameter and the inter-group differentiation parameter.

[0042] Specifically, the stability discrimination parameter is used to comprehensively characterize the intra-group stability and inter-group discrimination of mass-to-charge ratio characteristics. For mass-to-charge ratio characteristics that can be used for origin identification, it is desirable that the mass spectrometry response within the same origin is relatively stable, and that there are significant differences in response between different origins.

[0043] In this embodiment, the first Stability distinguishing parameter corresponding to each mass-to-charge ratio characteristic Determined according to the following formula: ; in, This is a preset positive number used to avoid fluctuations within the group. When the value is zero or close to zero, a division-by-zero operation or numerical anomaly occurs. Therefore, when the inter-group differentiation is large and the intra-group fluctuation is small, a large stable differentiation parameter can be obtained.

[0044] Step S205: Based on the stability differentiation parameter, the candidate mass-to-charge ratio feature set is obtained by filtering from multiple mass-to-charge ratio features.

[0045] Specifically, multiple mass-to-charge ratio features can be sorted according to stability discrimination parameters, and mass-to-charge ratio features whose stability discrimination parameters meet preset screening conditions can be selected. In this embodiment, the stability discrimination parameters corresponding to each mass-to-charge ratio feature are used... The mass-to-charge ratio (MMR) features are sorted from largest to smallest, and the top 20 MMR features are selected to form a candidate MMR feature set. When the total number of MMR features participating in the screening is less than 20, all MMR features are combined into a candidate MMR feature set. Therefore, given that the labeled mutton samples and their mass spectrometry data are determined, the corresponding candidate MMR feature sets can be repeatedly obtained according to the above-determined calculation relationships and screening rules.

[0046] The selected mass-to-charge ratio features can form a candidate mass-to-charge ratio feature set. The candidate mass-to-charge ratio feature set can record the mass-to-charge ratio position corresponding to each candidate mass-to-charge ratio feature and the arrangement order among the candidate mass-to-charge ratio features, so that the training feature data and the test feature data can be constructed using a unified feature position.

[0047] In this embodiment, candidate mass-to-charge ratio (M / C ratio) features are not selected solely based on the existence of response differences between different origins, but rather determined by combining the degree of differentiation between different origins and the degree of response fluctuation within the same origin. M / C ratio features with large inter-group differences but also large intra-group fluctuations can be suppressed, while M / C ratio features with significant inter-group differences and relatively stable intra-group responses are more likely to be retained, thus enabling the candidate M / C ratio feature set to stably characterize mass spectrometry differences between origins.

[0048] Optionally, in some embodiments, determining the stable differentiation parameter corresponding to each mass-to-charge ratio feature based on the intra-group fluctuation parameter and the inter-group differentiation parameter, i.e., step S204 may include: dividing the inter-group differentiation parameter corresponding to each mass-to-charge ratio feature by the sum of the corresponding intra-group fluctuation parameter and a preset positive number to obtain the stable differentiation parameter corresponding to each mass-to-charge ratio feature.

[0049] Specifically, for the first Individual mass-to-charge ratio characteristics, and their corresponding intra-group fluctuation parameters With preset positive numbers Add them together to get the denominator, and then use the corresponding inter-group discrimination parameter. The ratio is calculated using the numerator to obtain the first... Stability distinguishing parameter corresponding to each mass-to-charge ratio characteristic For multiple mass-to-charge ratio characteristics, the corresponding stability distinguishing parameters are determined according to the same calculation relationship.

[0050] Optionally, in some embodiments, step S205, which involves filtering the candidate mass-to-charge ratio feature set from multiple mass-to-charge ratio features based on the stability differentiation parameter, may include: Step S2051: Detect the flavor compounds in the multiple labeled mutton samples using gas chromatography-mass spectrometry (GC-MS) to determine the flavor compound response of each flavor compound in the multiple labeled mutton samples.

[0051] In practical applications, gas chromatography-mass spectrometry can be used to detect multiple labeled mutton samples separately, and the responses of different flavor substances in the multiple labeled mutton samples can be determined based on the detection results.

[0052] Flavor compound response is used to characterize the detection response of corresponding flavor compounds in labeled mutton samples. Flavor compound response can be expressed as the peak area, peak intensity, or standardized response data of the corresponding chromatographic peak.

[0053] To establish the correspondence between different detection results, mass spectrometry data and gas chromatography-mass spectrometry (GC-MS) detection data participating in cross-detection analysis can be mapped to the same labeled mutton sample. Thus, for any labeled mutton sample, the mass-to-charge ratio characteristic response and the flavor compound responses corresponding to multiple flavor compounds can be obtained.

[0054] In one implementation, gas chromatography-mass spectrometry (GC-MS) can be used to detect flavor compounds in labeled mutton samples, and the GC-MS detection results can be used as an independent detection dimension separate from the aforementioned mass spectrometry data. The detected flavor compounds may include aldehydes, esters, and other flavor compounds in the mutton samples. For mutton samples participating in cross-detection analysis, correspondences between the mutton samples and mass spectrometry data, as well as flavor compound responses, can be established separately to compare the two detection results at the same sample scale.

[0055] Step S2052: Based on the mass spectrometry response of each mass-to-charge ratio feature in the multiple labeled mutton samples and the response of each flavor substance, determine the cross-detection correlation parameter corresponding to each mass-to-charge ratio feature.

[0056] Specifically, the cross-detection correlation parameter is used to characterize the degree of correlation between the mass spectrometry response change of a certain mass-to-charge ratio feature and the response changes of one or more flavor substances in the same batch of labeled mutton samples.

[0057] For any mass-to-charge ratio characteristic, the mass spectrometry responses of multiple labeled mutton samples can be combined into a mass spectrometry response sequence according to the order of the labeled mutton samples. For any flavor compound, the flavor compound responses of multiple labeled mutton samples can be combined into a flavor compound response sequence according to the same sample order. Cross-detection correlation parameters can be determined based on the statistical correlation between the mass spectrometry response sequences and the flavor compound response sequences.

[0058] For example, the correlation coefficients between a mass-to-charge ratio feature and multiple flavor substance response sequences can be determined separately, and the maximum value of the absolute value of the correlation coefficient, the statistical value of multiple higher correlation coefficients, or the degree of correlation after combination processing can be used as the cross-detection association parameter of the corresponding mass-to-charge ratio feature.

[0059] It should be noted that cross-detection correlation parameters are used to reflect the correlation of response changes of different detection data on multiple samples, and do not require a one-to-one chemical relationship between a mass-to-charge ratio feature and a flavor substance.

[0060] Step S2053: Based on the stable differentiation parameter and the cross-detection correlation parameter, the candidate mass-to-charge ratio feature set is obtained by filtering from multiple mass-to-charge ratio features.

[0061] Specifically, the mass-to-charge ratio characteristics can be evaluated by comprehensively considering the stable distinguishing parameters and cross-detection correlation parameters.

[0062] In one implementation, the stable differentiation parameter and the cross-detection correlation parameter can be normalized separately, and then combined according to preset weights to form a comprehensive screening parameter corresponding to the mass-to-charge ratio feature. The comprehensive screening parameter can then be sorted or threshold-screened.

[0063] In another implementation, a first mass-to-charge ratio feature set can be formed first based on stable distinguishing parameters, and then mass-to-charge ratio features that satisfy preset correlation conditions across detection correlation parameters can be selected from the first mass-to-charge ratio feature set to form a candidate mass-to-charge ratio feature set.

[0064] In this embodiment, by introducing the flavor compound responses of multiple labeled mutton samples in another detection dimension, the mass-to-charge ratio (MMR) feature not only needs to meet the requirements of intra-group stability and inter-group distinguishability, but can also be further screened based on cross-detection response correlations across multiple samples. Therefore, the possibility of candidate features being formed solely by accidental differences in a single mass spectrometry dataset can be reduced, making the final retained candidate MMR features more clearly supported by cross-detection data.

[0065] Optionally, in some embodiments, step S102, which involves extracting the target feature data from the mass spectrometry data of the mutton sample based on the candidate mass-to-charge ratio feature set, may include: Step S1021: Based on the candidate mass-to-charge ratio feature set, determine the mass spectrometry response corresponding to each candidate mass-to-charge ratio feature in the mass spectrometry data of the mutton sample to be tested.

[0066] Specifically, the corresponding mass spectrometry response can be found in the mass spectrometry data of the mutton sample to be tested based on the position of each candidate mass-to-charge ratio recorded in the candidate mass-to-charge ratio feature set.

[0067] Considering that the actual peak positions corresponding to the same mass-to-charge ratio feature may vary in different mass spectrometry data, a corresponding mass-to-charge ratio matching range can be set for each candidate mass-to-charge ratio feature. For any candidate mass-to-charge ratio feature, the target mass spectrum peak can be determined within the corresponding mass-to-charge ratio matching range, and the peak intensity, peak area, or response value obtained by summarizing the target mass spectrum peak according to preset rules can be used as the mass spectrometry response of the corresponding mass-to-charge ratio feature.

[0068] The mass spectrometry responses corresponding to multiple candidate mass-to-charge ratio features can be arranged in the order of the candidate mass-to-charge ratio feature set to form the feature data to be processed.

[0069] Step S1022: Based on the mass spectrometry correction parameters and response normalization parameters determined from the mass spectrometry data of multiple labeled mutton samples, the mass spectrometry response is corrected and normalized to obtain correction feature data.

[0070] Specifically, mass spectrometry correction parameters can be used to correct the mass spectrometry response of the corresponding mass-to-charge ratio feature in different mass spectrometry data, and response normalization parameters can be used to transform the mass spectrometry response corresponding to different candidate mass-to-charge ratio features to a consistent data scale.

[0071] In one implementation, response baseline parameters corresponding to each candidate mass-to-charge ratio feature can be determined based on the mass spectrometry responses of multiple labeled mutton samples, and mass spectrometry correction parameters can be determined based on the response baseline parameters. For the mutton sample to be tested, baseline correction or response scaling correction can be performed on the mass spectrometry responses corresponding to each candidate mass-to-charge ratio feature based on the mass spectrometry correction parameters.

[0072] For response normalization, the response center parameter and response scale parameter for each candidate mass-to-charge ratio feature can be determined based on the training data corresponding to multiple labeled mutton samples. The response center parameter can be, for example, the mean or median of the responses corresponding to the training samples, while the response scale parameter can be, for example, the standard deviation, interquartile range, or other statistical parameters used to characterize the range of response variation. The mass spectrometry response of the mutton sample to be tested can then be normalized based on the corresponding response center parameter and response scale parameter.

[0073] Step S1023: Based on the dimension reduction mapping parameters determined by the correction feature data corresponding to the multiple labeled mutton samples, the correction feature data is mapped to the target feature space to obtain the feature data to be tested.

[0074] Specifically, the dimensionality reduction mapping parameters are used to transform the corrected feature data, which contains multiple candidate mass-to-charge ratio feature responses, into a lower-dimensional feature representation. The target feature space can be understood as a low-dimensional feature space established based on labeled mutton samples and used by the origin identification model.

[0075] In one implementation, a training feature matrix can be formed from the calibration feature data corresponding to multiple labeled mutton samples. Principal component analysis is then performed on the training feature matrix, and a dimensionality reduction mapping matrix is ​​determined based on multiple principal components that meet preset information retention conditions. This dimensionality reduction mapping matrix is ​​used as the dimensionality reduction mapping parameter. For the mutton sample to be tested, the same dimensionality reduction mapping matrix can be used to perform matrix mapping on the calibration feature data to obtain the feature data to be tested.

[0076] Dimensionality reduction mapping can also be achieved using other methods that can determine a fixed mapping relationship based on training samples and transform high-dimensional features to the target feature space.

[0077] The training feature data is obtained by processing the mass spectrometry response corresponding to the candidate mass-to-charge ratio feature set in the mass spectrometry data of each labeled mutton sample using the mass spectrometry correction parameters, the response normalization parameters, and the dimensionality reduction mapping parameters.

[0078] In other words, the same mass spectrometry correction parameters, response normalization parameters, and dimensionality reduction mapping parameters are used in both the training and testing phases. Therefore, the training feature data and the test feature data can reside in the same target feature space.

[0079] In this embodiment, the candidate mass-to-charge ratio characteristic responses in the mass spectrometry data are uniformly corrected, normalized, and spatially mapped before being input into the origin identification model. The training samples and the test samples use the same data processing benchmark, which can reduce the feature space deviation caused by different data processing scales, so that the origin discrimination relationship established in the training stage can be directly used for the mutton samples to be tested.

[0080] Optionally, in some embodiments, the mass spectrometry data is mass spectrometry data collected under the first ion polarity, and the analyte feature data is extracted from the mass spectrometry data according to a first candidate mass-to-charge ratio feature subset, wherein the first candidate mass-to-charge ratio feature subset is the mass-to-charge ratio feature in the candidate mass-to-charge ratio feature set that corresponds to the first ion polarity.

[0081] The step S103, which involves performing a first-level origin identification on the mutton sample based on the test feature data and the origin identification model, to determine whether the mutton sample belongs to Hulunbuir mutton or not, can include: Step S1031: Based on the feature data to be tested and the first identification sub-model in the origin identification model, determine the first-level candidate identification result and the corresponding identification confidence level of the mutton sample to be tested.

[0082] Specifically, the first identification sub-model can be a classification model trained based on training feature data corresponding to the polarity of the first ion. After inputting the feature data to be tested into the first identification sub-model, the first identification sub-model can output the category score or category probability corresponding to the Hulunbuir mutton category and the non-Hulunbuir mutton category, respectively.

[0083] The first-level candidate identification result can be a category score or a category with a high probability. The identification confidence score is used to characterize the degree of certainty of the first identification sub-model regarding the first-level candidate identification result.

[0084] In one implementation, the category probability corresponding to the first-level candidate identification result can be used as the identification confidence level. In another implementation, the identification confidence level can be determined based on the difference between the category scores corresponding to two categories.

[0085] Step S1032: If the identification confidence level meets the preset confidence conditions, determine whether the mutton sample to be tested belongs to Hulunbuir mutton or non-Hulunbuir mutton based on the first-level candidate identification results.

[0086] Specifically, the pre-set confidence conditions can be determined in advance based on the recognition results of the first recognition sub-model on the verification data.

[0087] For example, when using candidate class probabilities as the recognition confidence level, a preset confidence threshold can be set as a preset confidence condition. Similarly, when using the difference between scores corresponding to two classes as the recognition confidence level, a preset difference threshold can be set as a preset confidence condition.

[0088] If the confidence level meets the preset confidence conditions, it means that the test feature data corresponding to the polarity of the first ion has been able to form a first-level identification result that meets the preset requirements. The first-level candidate identification result can be directly determined as the first-level origin of the mutton sample to be tested.

[0089] Step S1033: If the identification confidence does not meet the preset confidence condition, supplementary mass spectrometry data of the mutton sample to be tested is collected using a second ion polarity different from the first ion polarity. Supplementary feature data is extracted from the supplementary mass spectrometry data according to the second candidate mass-to-charge ratio feature subset corresponding to the second ion polarity in the candidate mass-to-charge ratio feature set.

[0090] Specifically, when the recognition confidence corresponding to the first ion polarity is insufficient, mass spectrometry detection under the second ion polarity can be performed on the mutton sample to be tested to form supplementary mass spectrometry data.

[0091] For the supplementary mass spectrometry data under the second ion polarity, the corresponding mass spectrometry response can be found based on the second candidate mass-to-charge ratio feature subset, and supplementary feature data can be formed according to the data processing rules consistent with the model training phase.

[0092] The first candidate mass-to-charge ratio feature subset and the second candidate mass-to-charge ratio feature subset correspond to different ion polarities. Therefore, the target feature data and the supplementary feature data can respectively characterize the mass spectrometry features of the mutton sample under the two ion polarities.

[0093] Step S1034: Obtain bipolar joint feature data based on the feature data to be tested and the supplementary feature data; determine whether the mutton sample to be tested belongs to Hulunbuir mutton or non-Hulunbuir mutton based on the bipolar joint feature data and the supplementary identification sub-model in the origin identification model.

[0094] Specifically, the target feature data and supplementary feature data can be combined according to a preset data arrangement relationship to obtain bipolar joint feature data. For example, they can be spliced ​​in the order of the feature corresponding to the first ion polarity first and the feature corresponding to the second ion polarity last. The bipolar joint feature data can also be combined after mapping the data corresponding to the two ion polarities through a predetermined feature mapping method.

[0095] The supplementary identification sub-model can take bipolar joint feature data as input and output the identification results corresponding to Hulunbuir mutton or non-Hulunbuir mutton.

[0096] The first identification sub-model is trained based on the first polarity training feature data corresponding to multiple labeled mutton samples and the corresponding place of origin labels. The supplementary identification sub-model is trained based on the bipolar joint training feature data corresponding to multiple labeled mutton samples and the corresponding place of origin labels. The bipolar joint training feature data is obtained based on the corresponding first polarity training feature data and second polarity training feature data.

[0097] Specifically, when training the first identification sub-model, the first polarity training feature data can be extracted from the mass spectrometry data of multiple labeled mutton samples collected under the first ion polarity based on the first candidate mass-to-charge ratio feature subset, and then combined with the corresponding place of origin label to train the first identification sub-model.

[0098] When training the supplementary identification sub-model, the first polarity training feature data and the second polarity training feature data corresponding to the first ion polarity and the second ion polarity of multiple labeled mutton samples can be obtained respectively. The bipolar joint training feature data can be obtained by using the same data combination method as in step S1034. Then, the supplementary identification sub-model is trained based on the bipolar joint training feature data and the place of origin label.

[0099] In this embodiment, by making the first ion polarity detection the initial data source for the first level of identification, a second ion polarity detection is added only when the identification result corresponding to the first ion polarity cannot meet the preset confidence conditions. The second ion polarity can provide ion response information different from that of the first ion polarity. Low-confidence samples can be further identified by utilizing the combined features of the two ion polarities, while samples for which the first ion polarity has already formed a clear identification result do not require additional second ion polarity data. Thus, mass spectrometry information for origin determination can be configured according to the actual identification status of the mutton sample to be tested.

[0100] Optionally, in some embodiments, the origin identification model is trained based on the training feature data and the corresponding origin labels; the training of the origin identification model may include: Step S301: Determine the ecological characteristics of the place of origin corresponding to each of the labeled mutton samples based on the ecological data of the production area corresponding to each of the place of origin labels.

[0101] Specifically, the ecological data of the production area can be environmental and breeding background data related to the production area of ​​the mutton sample, which may include one or more of the following: pasture, soil, climate and grazing cycle.

[0102] The ecological data for the production areas can be organized according to the production areas corresponding to the mutton samples. Pasture-related data can be used to characterize the pasture conditions of the corresponding production area, soil data can be used to characterize the soil conditions of the corresponding production area, climate data can be used to characterize the climate conditions of the corresponding production area, and grazing cycle data can be used to characterize the grazing time characteristics of the production area where the corresponding mutton sample is located. Based on the origin labels of each labeled mutton sample, the above-mentioned ecological data for the corresponding production area can be linked with the labeled mutton samples, and then the resulting multi-category ecological data can be converted into corresponding production area ecological characteristics.

[0103] Based on the origin labels of the labeled mutton samples, the ecological data of the corresponding production areas can be linked to the labeled mutton samples. For numerical ecological data, normalization or standardization can be performed; for categorical ecological data, it can be converted into category codes or feature vectors; for data with time variations, the mean, range of variation, or other statistical characteristics can be determined based on the corresponding statistical period. Multiple processed ecological data can then form the ecological characteristics of the origin corresponding to the labeled mutton samples.

[0104] Origin ecological characteristics are mainly used to provide auxiliary information related to the origin of the samples during the model training phase.

[0105] Step S302: Train a multimodal teacher model based on the training feature data, the ecological features of the place of origin, and the corresponding place of origin labels.

[0106] Specifically, the multimodal teacher model is a supervised classification model that receives mass spectrometry dimension information and production area ecological dimension information during the training phase.

[0107] In this embodiment, the multimodal teacher model includes a mass spectrometry feature branch, a production area ecological feature branch, a feature fusion layer, and a teacher output layer. The mass spectrometry feature branch takes training feature data as input and includes 32 neurons using the ReLU activation function; the production area ecological feature branch takes production area ecological features as input and includes 16 neurons using the ReLU activation function. The mass spectrometry mapping features output by the mass spectrometry feature branch and the ecological mapping features output by the production area ecological feature branch are concatenated according to feature dimensions to obtain the fused input features. The feature fusion layer includes 16 neurons using the ReLU activation function to perform feature mapping on the fused input features. The teacher output layer includes 2 neurons using the Softmax function, outputting the category probabilities corresponding to Hulunbuir mutton and non-Hulunbuir mutton categories, respectively.

[0108] When training the multimodal teacher model, the corresponding origin labels are used as supervision information. The cross-entropy loss function is used to determine the loss between the origin category probability of the teacher output layer and the origin label. The Adam optimization algorithm is used to update the model parameters of the multimodal teacher model, and the learning rate is set to 0.001.

[0109] For any labeled mutton sample, the training feature data and corresponding origin ecological features can be input into the multimodal teacher model, with the corresponding origin label used as supervision information. During training, the multimodal teacher model can establish a mapping relationship between mass spectrometry features, origin ecological features, and origin categories.

[0110] After training, for each labeled mutton sample, the multimodal teacher model can output the category probabilities corresponding to multiple candidate origin categories.

[0111] Step S303: Input the training feature data into the student model, and train the student model according to the probability distribution of the origin category and the corresponding origin label output by the multimodal teacher model for the corresponding labeled mutton sample.

[0112] Specifically, the student model uses only the training feature data as input. For the same labeled mutton sample, it can obtain the true origin label and the origin category probability distribution output by the multimodal teacher model.

[0113] The probability distribution of origin categories can be understood as the data composed of the probability values ​​given by the multimodal teacher model for each candidate origin category. For example, in the case of two candidate origin categories, the probability distribution of origin categories can include the probabilities corresponding to the two candidate origin categories respectively.

[0114] In this embodiment, the student model includes an input layer, a first hidden layer, a second hidden layer, and a student output layer. The input layer receives training feature data. The first hidden layer includes 32 neurons using the ReLU activation function, the second hidden layer includes 16 neurons using the ReLU activation function, and the student output layer includes 2 neurons that use the Softmax function to output the probability distribution of the origin category corresponding to Hulunbuir mutton and non-Hulunbuir mutton.

[0115] For any labeled mutton sample, let the probability distribution of the origin category output by the student model be... The multimodal teacher model outputs the probability distribution of origin category for the same labeled mutton sample as follows: Label classification loss of the student model Cross-entropy loss is used to characterize the difference between the probability distribution of origin categories output by the student model and the corresponding origin labels; distribution matching loss is also employed. KL divergence is used to characterize the probability distribution of origin category output by the student model. The probability distribution of origin category output by the multimodal teacher model The differences between them.

[0116] Training loss of the student model Determined according to the following formula: ; Among them, label classification loss The weight is 0.7, and the distribution matching loss is... The weight is 0.3. During training, the Adam optimization algorithm is used based on the training loss. Update the student model parameters, set the learning rate to 0.001, and the maximum number of training rounds to 500. End training when the validation set loss does not decrease for 20 consecutive rounds, and save the student model parameters corresponding to the lowest validation set loss.

[0117] Therefore, the multimodal teacher model performs feature mapping on the training feature data and the origin ecological features through a mass spectrometry feature branch and an origin ecological feature branch, respectively, and then concatenates the mapped features output from the two feature branches before inputting them into the feature fusion layer. The student model only receives the training feature data and determines the training loss based on the label classification loss and distribution matching loss, and updates the student model parameters with the training loss.

[0118] Step S304: The trained student model is determined as the origin identification model.

[0119] Specifically, after the student model meets the preset training termination conditions, the model structure and model parameters corresponding to the student model can be saved, and the trained student model can be used as the origin identification model in the identification stage of the mutton sample to be tested.

[0120] Since the student model only receives training feature data during the training phase, the origin identification model after training can also only receive the test feature data obtained from the mass spectrometry data of the mutton sample to be tested during the testing phase, without needing to provide the corresponding ecological data of the origin for each mutton sample to be tested.

[0121] In this embodiment, the ecological information of the production area can participate in the establishment of the relationship between production area categories during the model training stage. The category probability distribution formed by the multimodal teacher model can also be passed as additional training information to the student model that only receives mass spectrometry features. Therefore, the ecological information of the production area can assist the production area identification model during the training stage, while the identification can still be completed based on the mass spectrometry features of the mutton sample during the actual testing stage, avoiding the need to use the ecological data corresponding to the test sample as a necessary input for the actual application stage of the model.

[0122] Optionally, in some embodiments, the step of inputting the training feature data into the student model and training the student model based on the probability distribution of the origin category and the corresponding origin label output by the multimodal teacher model for the corresponding labeled mutton sample, i.e., step S303, may include: Step S3031: Determine the label classification loss based on the probability distribution of origin categories and the corresponding origin labels output by the student model; Step S3032: Determine the distribution matching loss based on the probability distribution of origin categories output by the student model and the probability distribution of origin categories output by the multimodal teacher model; Step S3033: Determine the training loss of the student model based on the label classification loss and the corresponding first weight, and the distribution matching loss and the corresponding second weight; Step S3034: Update the model parameters of the student model according to the training loss.

[0123] Specifically, for any labeled mutton sample, let the corresponding place of origin label be... The student model is for the first The output category probability for each origin category is: The multimodal teacher model is aimed at the first The output category probability for each origin category is: ,in, =1, 2, corresponding to Hulunbuir mutton category and non-Hulunbuir mutton category respectively. Take 0 or 1.

[0124] Label classification loss Determined according to the following formula: ; Distributed matching loss Determined according to the following formula: ; In this embodiment, the first weight is set to 0.7, the second weight is set to 0.3, and the training loss of the student model is... Determined according to the following formula: ; In step S3034, the training loss is used. As the optimization objective of the student model, the aforementioned Adam optimization algorithm is used to update the student model parameters until the aforementioned training termination condition is met.

[0125] In one implementation, after the origin identification model is trained, labeled mutton samples that were not involved in the model training can be used as independent verification samples to perform blind testing on the origin identification model. During blind testing, the feature data corresponding to the independent verification samples can be input into the origin identification model. Without providing the model with the actual origin labels of the independent verification samples, the corresponding origin identification results are obtained, and then compared with the actual origin labels of the independent verification samples. Based on the comparison results corresponding to multiple independent verification samples, the model training parameters can be adjusted or the model training can be re-executed, and the adjusted origin identification model can be used for subsequent mutton origin identification.

[0126] In some embodiments, this application also provides a mutton origin identification system. The mutton origin identification system may include mass spectrometry detection equipment and electronic equipment.

[0127] Mass spectrometry equipment can be used to acquire mass spectrometry data from mutton samples and provide the acquired data to electronic devices. The equipment outputs mass-to-charge ratio position and the corresponding mass spectrometric response. Data transmission between the mass spectrometry equipment and electronic devices can be achieved via wired or wireless communication interfaces, or by importing data files.

[0128] The electronic device may include one or more processors and a memory storing a computer program. When the computer program is executed by the processor, it can enable the electronic device to perform the data processing and origin identification steps in any of the foregoing embodiments based on the mass spectrometry data collected by the mass spectrometry detection device.

[0129] Specifically, the electronic device can store a set of candidate mass-to-charge ratio (MMR) features and an origin identification model. After receiving the mass spectrometry data of the mutton sample to be tested, the electronic device can form the target feature data based on the set of candidate MMR features, and complete the first-level origin identification based on the target feature data and the origin identification model. For the mutton origin identification system using the aforementioned optional implementation method, the electronic device can also save the mass spectrometry correction parameters, response normalization parameters, dimensionality reduction mapping parameters, identification confidence conditions, and different identification sub-models required by the corresponding implementation method, and perform data processing according to the corresponding implementation method.

[0130] By integrating mass spectrometry detection of mutton samples with feature processing and origin identification functions in electronic devices within the same mutton origin identification system, the data output by the mass spectrometry detection device can be directly used as the data basis for subsequent origin identification. The electronic device can complete the mutton origin identification according to unified data processing rules, thus forming a complete processing link from mutton sample mass spectrometry data acquisition to origin identification result output.

[0131] Electronic devices can be various forms of digital computing devices, such as desktop computers, workstations, servers, or other computing devices capable of performing the aforementioned data processing programs.

[0132] Figure 2 An exemplary structural diagram of an electronic device is disclosed. The electronic device may include one or more processors 1101, memory 1102, input device 1103, and output device 1104. The processor 1101, memory 1102, input device 1103, and output device 1104 may be connected via a bus 1105 or other communication connection methods.

[0133] The processor 1101 can execute the computer program stored in the memory 1102 and perform mass spectrometry data processing and mutton origin identification according to the computer program. The processor 1101 may include a central processing unit, a graphics processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other processing devices capable of executing computer programs.

[0134] The memory 1102 can be used to store computer programs and data used in the process of identifying the origin of mutton. The data may include mass spectrometry data, candidate mass-to-charge ratio feature sets, training feature data, feature data to be tested, origin identification models, and model parameters and data processing parameters used in the corresponding embodiments.

[0135] Input device 1103 can be used to receive information input by a user or data input by an external device. Output device 1104 can be used to output the mutton origin identification result. Output device 1104 may include a display device, and may also include a data interface for sending the identification result to other information systems.

[0136] With the aforementioned system structure, the mass spectrometry detection equipment is responsible for generating detection data corresponding to the mutton sample, and the electronic equipment is responsible for extracting features and determining the place of origin according to the predetermined data processing relationship. There is a clear data connection relationship between the two types of functions, thus enabling the aforementioned mutton place of origin identification method to be implemented in a system form.

[0137] In some embodiments, this application also provides a computer-readable storage medium. The computer-readable storage medium may store a computer program, which, when executed by a processor, enables an electronic device to perform the data processing and origin identification steps described in any of the foregoing embodiments based on mass spectrometry data acquired by a mass spectrometry detection device.

[0138] Computer-readable storage media can be non-transitory storage media capable of storing computer programs and being read by a processor, such as magnetic storage media, optical storage media, semiconductor storage media, or other storage media capable of persistently storing computer programs.

[0139] The computer program may include program instructions for reading mass spectrometry data of mutton samples, program instructions for forming test feature data based on a set of candidate mass-to-charge ratio features, and program instructions for determining the origin of mutton samples based on an origin identification model. Depending on the implementation method, the computer program may also include program instructions for implementing corresponding data processing steps.

[0140] By storing the data processing and origin identification logic in mutton origin identification on a computer-readable storage medium, the same data processing logic can be loaded and executed on different electronic devices, enabling the software implementation of the aforementioned mutton origin identification method to be stored and deployed.

[0141] In some embodiments, this application also provides a computer program product. The computer program product may include a computer program, which, when executed by a processor, enables an electronic device to perform the data processing and origin identification steps described in any of the foregoing embodiments based on mass spectrometry data acquired by a mass spectrometry detection device.

[0142] Computer program products can be implemented in the form of programs that can be acquired and executed by electronic devices. Computer programs can be stored on local storage media or provided by servers or other computing devices to electronic devices that perform lamb origin identification.

[0143] The computer program can organize multiple program modules or program instructions according to the logical relationship corresponding to the aforementioned mutton origin identification method. Each program module or program instruction can respectively realize one or more functions in mass spectrometry data reading, formation of target feature data, and origin identification.

[0144] The computer program product enables the aforementioned data processing and origin identification logic to be configured into electronic devices in the form of a program. After the processor executes the corresponding computer program, it can complete the identification of the origin of mutton according to the same data processing relationship. Therefore, a software implementation form corresponding to the aforementioned method can be formed.

[0145] The aforementioned step divisions are primarily used to clearly illustrate the logical relationships between different processing procedures. Without changing the data sources, processing relationships, or the purpose of the processing results among the steps, some steps can be combined for execution, or a single step can be divided into multiple sub-steps for execution.

[0146] The models, parameters, feature sets, and feature data involved in the embodiments of this application can all be stored in software form, or implemented by hardware, software, or a combination of hardware and software according to actual computing needs. Different functions can be deployed in the same electronic device, or they can be performed separately by multiple computing devices with data communication relationships.

[0147] It should be noted that the terms "first" and "second" used in the embodiments of this application are mainly used to distinguish objects with different functions, different levels, or different data sources, and do not indicate that there is necessarily a difference in time order or importance between the objects. Unless the logical relationship of the method steps explicitly requires a specific execution order, the actual execution order of each data processing process can be determined according to the data dependency relationship.

[0148] The computer program described in this application embodiment can be implemented using one or more programming languages. The computer program can be executed entirely on the local electronic device, or it can be executed in conjunction with a remote computing device. The remote computing device can establish a data communication connection with the local electronic device through a local area network, a wide area network, or other communication networks.

[0149] The foregoing embodiments are merely illustrative of the technical solutions of this application and are not intended to limit the scope of protection of this application. Without altering the technical solutions defined in the claims, those skilled in the art can select the corresponding data processing method based on the actual testing equipment, data scale, and computing environment.

Claims

1. A method for identifying the origin of mutton, characterized in that, The method includes: Collect mass spectrometry data of the mutton sample to be tested; Based on the candidate mass-to-charge ratio feature set, the target feature data are extracted from the mass spectrometry data of the mutton sample to be tested; wherein, the candidate mass-to-charge ratio feature set is determined based on the mass spectrometry data of multiple labeled mutton samples with origin labels; Based on the test feature data and the origin identification model, the test mutton sample is subjected to first-level origin identification to determine whether the test mutton sample belongs to Hulunbuir mutton or not Hulunbuir mutton; the origin identification model is trained based on training feature data extracted from the mass spectrometry data of multiple labeled mutton samples according to the candidate mass-to-charge ratio feature set and the corresponding origin labels; the mass spectrometry data is mass spectrometry data collected under the first ion polarity, and the test feature data is extracted from the mass spectrometry data according to the first candidate mass-to-charge ratio feature subset, which is the mass-to-charge ratio feature in the candidate mass-to-charge ratio feature set corresponding to the first ion polarity; The origin identification model adopts a supervised classification model that can output the origin category based on the input features; The method for determining the candidate mass-to-charge ratio feature set includes: Based on the origin label, the mass spectrometry data of multiple labeled mutton samples are grouped. Based on the mass-to-charge ratio characteristic response of each labeled mutton sample in the same group, determine the intra-group fluctuation parameter corresponding to each mass-to-charge ratio characteristic. Based on the differences in the corresponding mass-to-charge ratio characteristic responses between different groups, determine the inter-group discrimination parameters corresponding to each mass-to-charge ratio characteristic; Based on the intra-group fluctuation parameters and the inter-group differentiation parameters, determine the stability differentiation parameters corresponding to each of the mass-to-charge ratio characteristics; Based on the stability differentiation parameter, the candidate mass-to-charge ratio feature set is obtained by screening from multiple mass-to-charge ratio features.

2. The method according to claim 1, characterized in that, The step of determining the stability differentiation parameter corresponding to each mass-to-charge ratio characteristic based on the intra-group fluctuation parameter and the inter-group differentiation parameter includes: Divide the inter-group discrimination parameter corresponding to each of the mass-to-charge ratio characteristics by the sum of the corresponding intra-group fluctuation parameter and a preset positive number to obtain the stable discrimination parameter corresponding to each of the mass-to-charge ratio characteristics.

3. The method according to claim 1, characterized in that, The step of selecting the candidate mass-to-charge ratio feature set from multiple mass-to-charge ratio features based on the stability differentiation parameter includes: The flavor compounds in multiple labeled mutton samples were detected using gas chromatography-mass spectrometry to determine the flavor compound response of each flavor compound in the multiple labeled mutton samples. Based on the mass spectrometry response of each mass-to-charge ratio feature in multiple labeled mutton samples and the response of each flavor substance, determine the cross-detection correlation parameter corresponding to each mass-to-charge ratio feature; Based on the stability differentiation parameter and the cross-detection correlation parameter, the candidate mass-to-charge ratio feature set is obtained by screening from multiple mass-to-charge ratio features.

4. The method according to claim 1, characterized in that, The step of extracting the target feature data from the mass spectrometry data of the mutton sample to be tested based on the candidate mass-to-charge ratio feature set includes: Based on the set of candidate mass-to-charge ratio features, determine the mass spectrometry response corresponding to each candidate mass-to-charge ratio feature in the mass spectrometry data of the mutton sample to be tested; Based on the mass spectrometry correction parameters and response normalization parameters determined from the mass spectrometry data of multiple labeled mutton samples, the mass spectrometry response is corrected and normalized to obtain correction feature data. Based on the dimension reduction mapping parameters determined by the correction feature data corresponding to multiple labeled mutton samples, the correction feature data is mapped to the target feature space to obtain the feature data to be tested. The training feature data is obtained by processing the mass spectrometry response corresponding to the candidate mass-to-charge ratio feature set in the mass spectrometry data of each labeled mutton sample using the mass spectrometry correction parameters, the response normalization parameters, and the dimensionality reduction mapping parameters.

5. The method according to claim 1, characterized in that, The step of performing first-level origin identification on the mutton sample based on the test feature data and the origin identification model to determine whether the mutton sample belongs to Hulunbuir mutton or not includes: Based on the feature data to be tested and the first identification sub-model in the origin identification model, the first-level candidate identification result and the corresponding identification confidence level of the mutton sample to be tested are determined. If the identification confidence level meets the preset confidence conditions, the mutton sample to be tested is determined to be either Hulunbuir mutton or non-Hulunbuir mutton based on the first-level candidate identification results. If the identification confidence level does not meet the preset confidence conditions, supplementary mass spectrometry data of the mutton sample to be tested is collected using a second ion polarity different from the first ion polarity. Supplementary feature data is extracted from the supplementary mass spectrometry data based on the second candidate mass-to-charge ratio feature subset corresponding to the second ion polarity in the candidate mass-to-charge ratio feature set. Based on the test feature data and the supplementary feature data, bipolar joint feature data is obtained. Based on the bipolar joint feature data and the supplementary identification sub-model in the origin identification model, it is determined whether the test mutton sample belongs to Hulunbuir mutton or non-Hulunbuir mutton. The first identification sub-model is trained based on the first polarity training feature data corresponding to multiple labeled mutton samples and the corresponding place of origin labels. The supplementary identification sub-model is trained based on the bipolar joint training feature data corresponding to multiple labeled mutton samples and the corresponding place of origin labels. The bipolar joint training feature data is obtained based on the corresponding first polarity training feature data and second polarity training feature data.

6. The method according to claim 1, characterized in that, The origin identification model is trained based on the training feature data and the corresponding origin labels; the training of the origin identification model includes: Based on the ecological data of the production area corresponding to each of the aforementioned origin labels, the ecological characteristics of the origin corresponding to each of the labeled mutton samples are determined. A multimodal teacher model is trained based on the training feature data, the ecological features of the place of origin, and the corresponding place of origin labels; The training feature data is input into the student model, and the student model is trained based on the probability distribution of the place of origin category and the corresponding place of origin label output by the multimodal teacher model for the corresponding labeled mutton sample. The trained student model is then identified as the origin identification model.

7. The method according to claim 6, characterized in that, The step of inputting the training feature data into the student model and training the student model based on the probability distribution of the origin category and the corresponding origin label output by the multimodal teacher model for the corresponding labeled mutton sample includes: Based on the probability distribution of origin categories and the corresponding origin labels output by the student model, determine the label classification loss; Based on the probability distribution of origin categories output by the student model and the probability distribution of origin categories output by the multimodal teacher model, determine the distribution matching loss; The training loss of the student model is determined based on the label classification loss and its corresponding first weight, and the distribution matching loss and its corresponding second weight. The model parameters of the student model are updated based on the training loss.

8. A system for identifying the origin of mutton, characterized in that, This includes mass spectrometry detection equipment and electronic equipment; The mass spectrometry detection device is used to collect mass spectrometry data of mutton samples; The electronic device includes one or more processors and a memory storing a computer program. When the computer program is executed by the processor, the electronic device performs the data processing and origin identification steps in the method as described in any one of claims 1 to 7 based on the mass spectrometry data acquired by the mass spectrometry detection device.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it causes the electronic device to perform the data processing and origin identification steps in the method as described in any one of claims 1 to 7 based on the mass spectrometry data collected by the mass spectrometry detection device.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it causes the electronic device to perform the data processing and origin identification steps in the method as described in any one of claims 1 to 7 based on the mass spectrometry data collected by the mass spectrometry detection device.

Images