Nuclear magnetic resonance detection method for adulterated non-dairy cream in single cream

Abstract

The invention relates to the technical field of food adulteration detection, and discloses a nuclear magnetic resonance detection method for adulterated non-dairy cream in single cream. According to the method, a nuclear magnetic resonance spectrogram fingerprint database containing standard characteristic peak information and weight coefficients of various characteristic substances is established. The method comprises the following steps: performing nuclear magnetic resonance detection on a to-be-detected sample to obtain a one-dimensional hydrogen spectrum, preprocessing, and intercepting a specific chemical shift interval in sections; baseline calibration and noise suppression are independently carried out on each segment, then convolution alignment with a standard characteristic peak is carried out, peak position offset and peak area proportion are extracted, and a multi-dimensional characteristic comparison vector is formed. And carrying out composite operation on the vector and the weight coefficient, calculating a comprehensive abnormal score, carrying out grading evaluation according to a preset threshold value, and outputting a judgment report. Through segmented fine processing and a multi-feature weighted fusion model, the detection sensitivity, objectivity and accuracy of trace adulteration are improved, and quantitative judgment of adulteration conditions is realized.

Description

Technical Field

[0001] This invention relates to the field of food adulteration detection technology, specifically to a nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream. Background Technology

[0002] In the field of dairy product safety and quality control, adulterating light cream with inexpensive vegetable fat cream to reduce costs is a common adulteration method. Currently, the main detection techniques for this type of adulteration include chromatography, mass spectrometry, and conventional nuclear magnetic resonance (NMR) spectral comparison. Chromatography and mass spectrometry methods typically require complex sample pretreatment and mainly target single or a few markers, making them difficult to address the challenges of detecting the complex composition of vegetable fat cream and its partial overlap with the light cream matrix. Conventional NMR full-spectrum analysis or simple characteristic peak comparison is easily affected by sample matrix interference, instrument drift, and subtle peak changes, resulting in insufficient sensitivity for identifying trace and gradual adulteration. The interpretation of results heavily relies on the analyst's experience and lacks objective and unified quantitative standards.

[0003] Existing technologies struggle to reliably and sensitively extract the faint characteristic signals of various adulterants from complex one-dimensional proton spectra. Because light cream and vegetable fat cream overlap in their main components, their characteristic signals are often masked or slightly shifted, making it difficult for conventional global spectral processing methods to effectively separate and focus this crucial identification information. Furthermore, existing methods lack an intelligent discrimination model capable of integrating multi-dimensional spectral features and assigning scientific weights to different characteristic substances, thus limiting the objectivity and accuracy of detection conclusions and hindering the quantitative assessment and automatic grading of adulteration levels. Summary of the Invention

[0004] The purpose of this invention is to provide a nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for detecting adulterated vegetable fat cream in light cream using nuclear magnetic resonance imaging, the method comprising: Establish a standard nuclear magnetic resonance spectral fingerprint library for light cream and vegetable fat cream, which includes standard characteristic peak information and weighting coefficients of various characteristic substances; The light cream sample to be tested was subjected to nuclear magnetic resonance detection to obtain the original one-dimensional proton NMR spectrum, and the one-dimensional proton NMR spectrum was preprocessed in multiple steps. From the preprocessed spectrum, multiple specific chemical shift intervals corresponding to various characteristic substances in the standard spectral fingerprint library are segmented and extracted to form a set of segmented spectra of the sample to be tested. Independent baseline calibration and noise suppression are performed on each segment of the spectrum set of the test sample; After noise suppression, each segment of the spectrum is convolved and aligned with the standard feature peak information of the corresponding feature substance in the standard NMR spectrum fingerprint database. The peak position offset and peak area ratio of each segment of the spectrum are extracted to form a multi-dimensional feature comparison vector. The multi-dimensional feature comparison vector is combined with the weight coefficients of each feature substance in the standard spectral fingerprint database to calculate a comprehensive anomaly score that reflects the possibility that the light cream sample to be tested is adulterated with vegetable fat cream. Based on the preset adulteration judgment threshold, the comprehensive anomaly score is graded and evaluated, and the final adulteration judgment conclusion report is output.

[0006] Preferably, the step of establishing a standard nuclear magnetic resonance spectral fingerprint library for light cream and vegetable fat cream, containing standard characteristic peak information and weighting coefficients of various characteristic substances, is specifically implemented as follows: A large number of pure light cream and pure vegetable fat cream samples from known sources were collected and analyzed to obtain their respective one-dimensional proton NMR spectra. The nuclear magnetic resonance spectra of pure light cream and pure vegetable fat cream were analyzed to screen out characteristic substances that showed stable and significant differences in the two types of samples. These characteristic substances include, but are not limited to, characteristic peaks of saturated fatty acids of different chain lengths, characteristic peaks of unsaturated fatty acids, and characteristic peaks of esters. For each selected characteristic substance, its standard characteristic peak information is precisely extracted from its nuclear magnetic resonance spectrum. The standard characteristic peak information includes the standard value of the center chemical shift of the characteristic peak, the standard peak shape profile data, and the standard half-peak width data. Based on the inherent differences in the content of each characteristic substance in light cream and vegetable fat cream and its importance in adulteration identification, a preset weighting coefficient is assigned to each characteristic substance. The standard characteristic peak information of each characteristic substance and its corresponding weight coefficient are systematically stored and managed to form a standard nuclear magnetic resonance spectrum fingerprint library.

[0007] Preferably, the step of performing nuclear magnetic resonance (NMR) detection on the cream sample to be tested to obtain the original one-dimensional proton NMR spectrum, and performing multi-step preprocessing on the one-dimensional proton NMR spectrum, specifically including: Under constant magnetic field and temperature conditions, nuclear magnetic resonance scanning was performed on the cream sample to be tested, and the original free induction decay signal was acquired. The acquired free induction decay signal is subjected to Fourier transform to generate the initial time-domain transformed nuclear magnetic resonance spectrum; Phase correction is performed on the initial time-domain transformed nuclear magnetic resonance spectrum to ensure that the peak shapes of all characteristic peaks are symmetrical; Perform global baseline adjustment to eliminate low-frequency baseline drift caused by the instrument or the sample itself; The amplitude of the spectrum after phase correction and baseline adjustment is normalized to eliminate signal intensity differences caused by small fluctuations in sample concentration or instrument parameters.

[0008] Preferably, the step of segmenting and extracting multiple specific chemical shift intervals corresponding to various characteristic substances in the standard spectral fingerprint database from the preprocessed spectrum to form a set of segmented spectra of the sample to be tested is specifically implemented as follows: Read the standard nuclear magnetic resonance spectrum fingerprint database to obtain the standard characteristic peak information of each characteristic substance recorded therein; Based on the standard chemical shift values ​​recorded in the standard characteristic peak information, and combined with their standard half-peak width data, the complete chemical shift range of each characteristic substance on the nuclear magnetic resonance spectrum is calculated. On the pre-processed nuclear magnetic resonance spectrum of the sample to be tested, precise spectral cuts are made based on the calculated complete chemical shift range of each characteristic substance. For each extracted spectral image segment, associate it with the corresponding characteristic substance name; All the spectral segments with associated labels are collected to form a structured set of segmented spectra of the sample to be tested.

[0009] Preferably, the detailed process of performing independent baseline calibration and noise suppression on each segment of the spectrum set of the sample to be tested includes: Iterate through each spectral segment in the set of segmented spectra of the sample to be tested; For the spectrum image segment currently being processed, a polynomial fitting method is used to perform local quadratic fitting of its baseline to eliminate any possible local baseline distortion within the spectrum image segment; On the spectrum image segments that have completed baseline calibration, a wavelet transform threshold denoising algorithm is applied to identify and filter out high-frequency random noise signals in the spectrum image segments. Record the baseline calibration parameters and noise suppression parameters used in the independent processing of each spectral image segment; All spectral segments that have undergone independent baseline calibration and noise suppression are updated to the sample segment spectrum set to be tested.

[0010] Preferably, the specific steps of convolving and aligning each segmented spectrum after noise suppression with the standard feature peak information of the corresponding feature substance in the standard NMR spectral fingerprint database, extracting the peak position offset and peak area ratio of each segmented spectrum, and forming a multi-dimensional feature comparison vector are as follows: From the standard nuclear magnetic resonance spectrum fingerprint database, retrieve the standard characteristic peak information of each characteristic substance in sequence, especially its standard peak shape profile data; From the updated set of segmented spectra of the test sample, extract the noise-suppressed spectral segment data corresponding to the current characteristic substance; The spectral image segment data is convolved and cross-correlated with the standard peak shape contour data of the current characteristic substance. By finding the maximum value point of the convolution result, the two are precisely aligned in the chemical shift dimension. After the alignment operation is completed, the difference between the actual center chemical shift of the characteristic peak in the spectrum segment and the standard value of the center chemical shift recorded in the standard nuclear magnetic resonance spectrum fingerprint library is calculated as the peak position offset of the characteristic substance. The net peak area of ​​the characteristic peak in the spectrum segment is calculated by integration, and it is compared with the peak area of ​​the specified internal standard reference peak in the spectrum of the same sample to calculate the peak area ratio of the characteristic substance. The peak position offsets and peak area ratios extracted from all characteristic substances are arranged and combined in a predetermined order to form a multi-dimensional feature comparison vector containing all comparison features.

[0011] Preferably, the comprehensive anomaly score reflecting the possibility of adulteration with vegetable fat cream in the tested light cream sample is calculated by performing a composite operation on the multi-dimensional feature comparison vector and the weight coefficients of each feature substance in the standard spectral fingerprint database. The calculation logic is as follows: Read the preset weighting coefficients for each characteristic substance from the standard nuclear magnetic resonance spectral fingerprint database; For each feature comparison data in the multi-dimensional feature comparison vector, namely the peak position offset and peak area ratio of each feature substance, the deviation from the standard value of pure cream is calculated. The weighted individual anomaly contribution values ​​corresponding to all feature comparison data in the multi-dimensional feature comparison vector are summed to obtain a preliminary cumulative anomaly score. The preliminary cumulative anomaly score is standardized and mapped to a preset fixed numerical range to obtain the final comprehensive anomaly score. The value of the comprehensive anomaly score directly reflects the likelihood of adulteration.

[0012] Preferably, the step of classifying and evaluating the comprehensive anomaly score according to a preset adulteration determination threshold includes the following detailed evaluation process: Multiple adulteration detection thresholds with progressively increasing values ​​are pre-set. These adulteration detection thresholds divide the possibility of adulteration into different level ranges, including "not detected", "low suspicion", "medium suspicion" and "high suspicion". The calculated comprehensive anomaly score is compared sequentially with multiple preset adulteration judgment thresholds; The adulteration risk level of the cream sample to be tested is determined based on the specific level range in which the comprehensive anomaly score falls. Record the specific value of the comprehensive anomaly score and the adulteration risk level it determines.

[0013] Preferably, the step of outputting the final adulteration determination report specifically includes the following: Generate a structured report on adulteration determination; The adulteration determination report clearly lists the unique identification information of the tested cream sample; The adulteration determination report shall record in detail the specific value of the calculated comprehensive anomaly score; The adulteration risk level determined based on the comparison results is clearly marked in the adulteration determination report; The adulteration determination report includes a summary of the key feature comparison data in the multi-dimensional feature comparison vector, especially the peak position offset and peak area ratio data of the feature substances with significant deviation. The adulteration determination report provides analytical notes and subsequent processing suggestions based on the current adulteration determination conclusion.

[0014] Preferably, after the step of generating a structured adulteration determination report, the method further includes a report verification and archiving step, specifically: Logically associate the adulteration determination report with all intermediate data generated during this testing process, including the preprocessed NMR spectrum, the set of segmented spectra of the sample to be tested, the multi-dimensional feature comparison vector, and the key parameters in the calculation process. All related data and reports are stored in the test results database, and a unique test record number is assigned to each test. The complete testing records stored in the testing results database are backed up, and an unalterable testing process and results summary log is generated simultaneously for subsequent auditing and quality traceability.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By segmenting the raw one-dimensional proton spectrum according to the standard chemical shift ranges of various characteristic substances, and independently performing baseline calibration and noise suppression on each range, the complex full-spectrum analysis task is decomposed into multiple targeted signal extraction channels. This isolates the target analysis range from background interference from complex matrices, enhancing the capture and purification capabilities of weak characteristic signals at specific chemical shifts. Convolutional alignment of each processed spectrum with the corresponding standard characteristic peak information accurately compensates for minor peak shifts caused by instrument conditions or the sample microenvironment, ensuring the accuracy of subsequent characteristic parameter extraction and laying a stable data foundation for high-sensitivity identification.

[0016] Multi-dimensional parameters such as peak position shift and peak area ratio are extracted from segmented spectra processed independently, and a feature comparison vector is constructed. This vector is then combined with pre-calibrated weight coefficients for each feature substance to generate a quantitative comprehensive anomaly score. A mathematical model based on multi-feature fusion and weighted decision-making is established. Subjective spectral interpretation is eliminated, and the identification process is transformed into objective data calculation. Different feature substances are assigned different weights according to their differences in discriminative power, enabling the final score to more scientifically and sensitively reflect the overall anomaly state of the sample. The quantitative score can be directly used to set clear judgment thresholds, achieving automated graded output of adulteration conclusions, significantly improving the objectivity, consistency, and interpretability of the detection results, and distinguishing different levels of adulteration. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to the present invention. Figure 2 A flowchart for establishing a standard nuclear magnetic resonance spectral fingerprint database; Figure 3 This is a flowchart of sample NMR detection and spectrum preprocessing; Figure 4 A biaxial plot comparing the peak parameters of characteristic substances for detecting adulteration in light cream; Figure 5 This is a trend chart of key parameters in the preprocessing stage of nuclear magnetic resonance spectra. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides a method for detecting adulterated vegetable fat cream in light cream using nuclear magnetic resonance (NMR). The method includes: establishing a standard NMR spectral fingerprint library for light cream and vegetable fat cream, which contains standard characteristic peak information and weighting coefficients for various characteristic substances; performing NMR detection on the light cream sample to be tested to obtain the original one-dimensional proton NMR spectrum, and performing multi-step preprocessing on the one-dimensional proton NMR spectrum; from the preprocessed spectrum, segmenting multiple specific chemical shift intervals corresponding to various characteristic substances in the standard spectral fingerprint library to form a set of segmented spectra of the sample to be tested; and processing the sample in the set of segmented spectra... Each segment undergoes independent baseline calibration and noise suppression. The noise-suppressed spectrum of each segment is convolved and aligned with the standard characteristic peak information of the corresponding characteristic substances in the standard NMR spectral fingerprint database. The peak position offset and peak area ratio of each segment are extracted to form a multi-dimensional feature comparison vector. This multi-dimensional feature comparison vector is then combined with the weight coefficients of each characteristic substance in the standard spectral fingerprint database to calculate a comprehensive anomaly score reflecting the possibility of adulteration with vegetable fat cream in the tested cream sample. Based on a preset adulteration judgment threshold, the comprehensive anomaly score is graded and evaluated, and a final adulteration judgment report is output.

[0020] In one embodiment of the present invention, see [reference] Figure 2 A large number of pure cream and pure vegetable fat cream samples from known sources were collected and analyzed to obtain their respective one-dimensional proton NMR spectra. The NMR spectra of pure cream and pure vegetable fat cream were analyzed to screen out characteristic substances that exhibited stable and significant differences in the two types of samples. These characteristic substances include, but are not limited to, characteristic peaks of saturated fatty acids of different chain lengths, characteristic peaks of unsaturated fatty acids, and characteristic peaks of esters. For each screened characteristic substance, its standard characteristic peak information was precisely extracted from its NMR spectrum. The standard characteristic peak information includes the standard value of the center chemical shift of the characteristic peak, standard peak shape data, and standard half-peak width data. Based on the inherent content differences of each characteristic substance in cream and vegetable fat cream and its importance in adulteration identification, a preset weighting coefficient was assigned to each characteristic substance. The standard characteristic peak information and its corresponding weighting coefficient of each characteristic substance were systematically stored and managed to construct a standard NMR spectrum fingerprint database.

[0021] In practice, a standard NMR spectral fingerprint library for whipped cream and vegetable-based cream was established. This involved collecting and analyzing one-dimensional proton NMR spectral data from a large number of pure whipped cream and pure vegetable-based cream samples from known sources. These pure samples were sourced from verified manufacturers or prepared using standard methods, and their NMR detection was performed under identical instrument parameters and sample pretreatment conditions to ensure the comparability and consistency of the obtained spectral data. The NMR spectra of pure whipped cream and pure vegetable-based cream were analyzed to screen for characteristic substances exhibiting stable and significant differences between the two types of samples. This screening process involves comparing the spectra of a large number of pure samples to identify peak regions with systematic differences in chemical shift and peak intensity. These characteristic substances include, but are not limited to, characteristic peaks of saturated fatty acids of different chain lengths, characteristic peaks of unsaturated fatty acids, and characteristic peaks of esters.

[0022] In some embodiments, for each selected characteristic substance, its standard characteristic peak information is precisely extracted from its nuclear magnetic resonance spectrum. This extraction operation involves high-precision fitting and analysis of the spectral intervals of the selected characteristic peaks. The standard characteristic peak information includes the standard value of the center chemical shift of the characteristic peak, standard peak shape profile data, and standard half-maximum width (HWW) data. In specific implementations, the standard value of the center chemical shift is obtained by statistically averaging the center positions of the corresponding characteristic peaks of all pure samples. The standard peak shape profile data is obtained by superimposing and averaging the characteristic peak segments of multiple pure samples and then normalizing them. The standard HWW data is derived from the statistical analysis of the HWW measurements of the characteristic peak in the pure sample spectra.

[0023] In some embodiments, based on the inherent differences in the content of each characteristic substance in light cream and vegetable fat cream and its importance in adulteration identification, a preset weighting coefficient is assigned to each characteristic substance. Optionally, one method for assigning weighting coefficients is as follows: in: Indicates assigning the first Weighting coefficients for each characteristic substance; Indicates the first The absolute difference in the average normalized peak area of ​​the characteristic substances in pure vegetable fat cream and pure light cream; Indicates the first Standard deviation of peak positions of a characteristic substance in the spectrum of a pure light cream sample; and To adjust constants, the influence of content differences and peak position stability is balanced. In practice, the standard characteristic peak information and its corresponding weighting coefficients for each characteristic substance are systematically stored and managed. Optionally, the storage format uses a structured database or data file to ensure that each piece of information is uniquely associated with the corresponding characteristic substance name, thus constructing a standard NMR spectrum fingerprint library.

[0024] In one embodiment of the present invention, see [reference] Figure 3 Under constant magnetic field and temperature conditions, nuclear magnetic resonance (NMR) scanning was performed on the cream sample to be tested, and the original free induction decay signal was acquired. Fourier transform was performed on the acquired free induction decay signal to generate an initial time-domain transformed NMR spectrum. Phase correction was performed on the initial time-domain transformed NMR spectrum to ensure the peak shape symmetry of all characteristic peaks. Global baseline adjustment was performed to eliminate low-frequency baseline drift caused by the instrument or the sample itself. Amplitude normalization was performed on the phase-corrected and baseline-adjusted spectra to eliminate signal intensity differences caused by small fluctuations in sample concentration or instrument parameters.

[0025] In practice, the NMR detection of the cream sample requires performing the NMR scan under constant magnetic field and temperature conditions. For example, the magnetic field strength is stabilized at 14.1 Tesla, and the sample temperature is controlled at 298.0 ± 0.1 Kelvin. Under these conditions, the original free-induction decay signal is acquired. The scanning parameters include a spectral width of 20 ppm, a sampling number of 65536, and a relaxation delay time of 5.0 seconds to ensure sufficient acquisition of the resonance information of hydrogen nuclei in the sample. The acquired free-induction decay signal is then subjected to a Fourier transform to generate an initial time-domain transformed NMR spectrum. In practice, before the Fourier transform, the free-induction decay signal is zero-padded to 131072 points and an exponential window function with a 0.3 Hz linewidth factor is applied to optimize the spectral resolution and signal-to-noise ratio.

[0026] In some embodiments, phase correction is performed on the initial time-domain transformed NMR spectrum to ensure the peak shape symmetry of all characteristic peaks. Optionally, phase correction is achieved by manually adjusting the zero-order and first-order phase correction parameters, or by using an automatic phase correction algorithm iteratively until the real part signal of all peaks in the spectrum reaches its maximum and the peak shape is symmetrical. In a specific implementation, the automatic phase correction algorithm automatically adjusts the zero-order and first-order phase correction parameters through an iterative optimization process. The algorithm aims to maximize the real part signal intensity and peak shape symmetry of all characteristic peaks in the NMR spectrum. During algorithm initialization, the initial value of the phase parameters is set, usually zero or an estimated value based on historical data. Subsequently, in each iteration, the overall intensity or peak shape symmetry index of the real part signal of the spectrum under the current phase parameters is calculated, and the parameters are updated based on gradient descent or similar optimization methods. The iteration continues until the real part signal intensity no longer increases significantly and the peak shape of all characteristic peaks is close to symmetrical. At this point, the algorithm terminates and outputs the final phase correction parameters, achieving fully automatic spectrum phase correction without manual intervention. Global baseline adjustment is performed on the phase-corrected spectrum to eliminate low-frequency baseline drift caused by the instrument or the sample itself. In practice, global baseline adjustment uses a polynomial fitting method, such as using a fifth-order polynomial to fit the spectrum baseline over the entire chemical shift range, and then subtracting the fitted baseline curve from the original spectrum.

[0027] In some embodiments, amplitude normalization is performed on the spectrum after phase correction and baseline adjustment to eliminate signal intensity differences caused by minor fluctuations in sample concentration or instrument parameters. It can be understood that amplitude normalization maps the signal intensity of the entire spectrum to a uniform range. In specific implementations, one method of amplitude normalization scales the spectrum based on the peak area of ​​a specified internal standard reference. Optionally, another method achieves amplitude normalization by calculating a certain norm of the total signal intensity of the spectrum, the relationship of which can be expressed as: in: Indicates chemical shift The normalized signal strength at the location; Indicates chemical shift The original signal strength at that location; This indicates that all sampling points in the entire spectrum are considered. The sum of squares of the corresponding original signal strength; This is a preset constant scaling factor used to adjust the normalized signal strength to an order of magnitude that is easy to observe and calculate.

[0028] In one embodiment of the present invention, the standard nuclear magnetic resonance (NMR) spectrum fingerprint database is read to obtain the standard characteristic peak information of each characteristic substance recorded therein; based on the standard chemical shift standard value recorded in the standard characteristic peak information, combined with its standard half-peak width data, the complete chemical shift range of each characteristic substance on the NMR spectrum is calculated; on the preprocessed NMR spectrum of the sample to be tested, according to the calculated complete chemical shift range of each characteristic substance, a precise spectrum segment is performed; for each segmented spectrum, it is associated with the name of the corresponding characteristic substance; all associated spectrum segments are collected to form a structured set of segmented spectra of the sample to be tested. The algorithm iterates through each spectral segment in the sample segment spectrum set. For the currently being processed spectral segment, a polynomial fitting method is used to perform local quadratic fitting of its baseline to eliminate local baseline distortion within the spectral segment. On the spectral segments that have completed baseline calibration, a wavelet transform threshold denoising algorithm is applied to identify and filter out high-frequency random noise signals in the spectral segments. The baseline calibration parameters and noise suppression parameters used for each spectral segment during independent processing are recorded. All spectral segments that have undergone independent baseline calibration and noise suppression processing are updated in the sample segment spectrum set.

[0029] In practice, the operation of segmenting specific chemical shift intervals from the preprocessed spectrum begins by reading the standard NMR spectrum fingerprint database to obtain the standard characteristic peak information for each characteristic substance. This information is stored in a data structure containing the characteristic substance name, the standard value of the central chemical shift, and the standard half-width at half-maximum (HWHM). Based on the standard chemical shift standard value recorded in the standard characteristic peak information, and its standard HWHM, the complete chemical shift interval range for each characteristic substance on the NMR spectrum is calculated. Optionally, one calculation method is to define the lower limit of the interval as "the standard value of the central chemical shift minus three times the standard HWHM" and the upper limit as "the standard value of the central chemical shift plus three times the standard HWHM," ensuring complete coverage of the characteristic peak distribution range. On the preprocessed NMR spectrum of the sample to be tested, precise spectrum segmentation is performed based on the calculated complete chemical shift interval range for each characteristic substance. In practice, the segmentation operation extracts all signal intensity data points within the indexed interval by locating the corresponding chemical shift index in the spectrum data array.

[0030] In some embodiments, each extracted spectral segment is associated with its corresponding characteristic substance name. This association is achieved by adding a characteristic substance name attribute to each spectral segment object in the data structure. All associated spectral segments are then collected to form a structured set of segmented spectra of the test sample. In specific implementations, this set can be organized using a list or dictionary data structure, where each element contains the characteristic substance name, chemical shift interval data, and the corresponding signal intensity array. Each spectral segment in the set is traversed. For the currently processed segment, a polynomial fitting method is used to perform a local quadratic fitting of its baseline to eliminate local baseline distortion within the segment. In specific implementations, the local quadratic fitting uses the least squares method to perform a second-order polynomial fitting on the data points in the signal-free regions on both sides of the current segment, and then subtracts the fitted curve from the entire segment signal.

[0031] On the spectrum image segments after baseline calibration, a wavelet transform threshold denoising algorithm is applied to identify and filter out high-frequency random noise signals in the spectrum image segments. Optionally, the core steps of the wavelet transform threshold denoising algorithm involve multi-level wavelet decomposition of the signal, applying a threshold function to shrink the high-frequency wavelet coefficients, and reconstructing the signal. One form of the threshold function is expressed as follows: in: This represents the coefficient after thresholding. Represents the original wavelet decomposition coefficients; This indicates a threshold value set based on an estimate of the noise level. It is a symbolic function; This represents the absolute value operator. It is used to obtain the raw wavelet coefficients. The range. It is a maximum value function. It records the baseline calibration parameters and noise suppression parameters used in the independent processing of each spectral image segment, such as the coefficients of the local quadratic fitting, the mother wavelet type selected for wavelet denoising, the number of decomposition levels, and the specific values ​​of the threshold λ. All spectral image segments that have undergone independent baseline calibration and noise suppression processing are then updated in the sample segment spectrum set, replacing the original original spectral image segment data.

[0032] In one embodiment of the present invention, standard characteristic peak information, particularly standard peak shape contour data, for each characteristic substance is sequentially retrieved from the standard NMR spectral fingerprint database. From the updated set of segmented spectra of the test sample, noise-suppressed spectral segment data corresponding to the current characteristic substance is extracted. The spectral segment data and the standard peak shape contour data of the current characteristic substance are convolved and cross-correlated. By finding the maximum value of the convolution result, precise alignment in the chemical shift dimension is achieved. After alignment, the difference between the actual central chemical shift of the characteristic peak in the spectral segment and the standard value of the central chemical shift recorded in the standard NMR spectral fingerprint database is calculated as the peak position offset of the characteristic substance. The net peak area of ​​the characteristic peak in the spectral segment is calculated by integration and compared with the peak area of ​​the specified internal standard reference peak in the same sample spectrum to calculate the peak area ratio of the characteristic substance. The peak position offsets and peak area ratios extracted for all characteristic substances are arranged and combined in a predetermined order to form a multi-dimensional feature alignment vector containing all alignment features. The preset weighting coefficients for each characteristic substance are read from the standard NMR spectral fingerprint database; for each feature comparison data in the multi-dimensional feature comparison vector, i.e., the peak position offset and peak area ratio of each characteristic substance, the deviation from the standard value of pure cream is calculated; the weighted single-item anomaly contribution values ​​corresponding to all feature comparison data in the multi-dimensional feature comparison vector are summed to obtain a preliminary cumulative anomaly score; the preliminary cumulative anomaly score is standardized and mapped to a preset fixed numerical range to obtain the final comprehensive anomaly score, the value of which directly reflects the likelihood of adulteration.

[0033] In practice, the noise-suppressed segmented spectra are convolved and aligned with the standard characteristic peak information of the corresponding characteristic substances in the standard NMR spectral fingerprint database. This begins by sequentially retrieving the standard characteristic peak information of each characteristic substance from the database, particularly its standard peak shape profile data. This standard peak shape profile data is understood to be a discrete sequence of data points, representing the ideal linear shape of the characteristic peaks of a pure substance. From the updated set of segmented spectra of the sample to be tested, noise-suppressed spectral segment data corresponding to the current characteristic substance is extracted. This data is also a signal intensity sequence at the same chemical shift resolution. The spectral segment data and the standard peak shape profile data of the current characteristic substance are then convolved and cross-correlated. By finding the maximum value in the convolution result sequence, precise alignment in the chemical shift dimension is achieved. In practice, the convolution and cross-correlation operation is performed in the frequency domain to improve computational efficiency. After alignment, the starting chemical shift index of the spectral segment to be tested is adjusted based on the position of the maximum value.

[0034] In some embodiments, after the alignment operation is completed, the difference between the actual central chemical shift of the characteristic peak in the spectrum segment and the standard value of the central chemical shift recorded in the standard NMR spectral fingerprint library is calculated as the peak position offset of the characteristic substance. The unit of the peak position offset is ppm or Hz. The net peak area of ​​the characteristic peak in the spectrum segment is calculated by integration and compared with the peak area of ​​the specified internal standard reference peak in the spectrum of the same sample to calculate the peak area ratio of the characteristic substance. Optionally, the internal standard reference peak can be a characteristic peak that is stable in the sample matrix and is not affected by adulteration, such as the signal of the lactose characteristic peak at a specific chemical shift. The peak position offsets and peak area ratios extracted for all characteristic substances are arranged and combined in a predetermined order to form a multi-dimensional feature alignment vector containing all alignment features. In specific implementations, this vector can be represented as a numerical array whose order is strictly consistent with the recording order of the characteristic substances in the standard NMR spectral fingerprint library.

[0035] Weighting coefficients are pre-defined for each characteristic substance from a standard NMR spectral fingerprint database. For each feature comparison data point in the multi-dimensional feature comparison vector—namely, the peak position shift and peak area ratio of each characteristic substance—the deviation from the standard value of pure cream is calculated. The standard value of pure cream includes the standard values ​​of the central chemical shift and peak area ratio of each characteristic substance in pure cream. The deviation can be calculated as an absolute difference, a relative difference, or a standardized distance metric. The weighted individual anomaly contribution values ​​corresponding to all feature comparison data in the multi-dimensional feature comparison vector are summed to obtain a preliminary cumulative anomaly score. The weighted individual anomaly contribution value is the product of the deviation of each characteristic substance and its weighting coefficient. The preliminary cumulative anomaly score is standardized and mapped to a pre-defined fixed numerical range, such as 0 to 100, to obtain the final comprehensive anomaly score. The value of the comprehensive anomaly score directly reflects the likelihood of adulteration. One method of standardization is as follows: in: This indicates the final comprehensive anomaly score; Represents a standardized function; This indicates the initial cumulative anomaly score; This represents the initial cumulative set of anomaly scores for a set of reference samples; This refers to comparing the preliminary cumulative anomaly score of the sample to be tested with the preliminary cumulative anomaly scores of a set of representative pure cream reference samples in order to eliminate systematic bias and improve the comparability of the scores. and These represent the minimum and maximum values ​​in the reference set, respectively. This represents the upper limit of the preset range, for example, 100. The reference sample is selected based on its representativeness as pure whipped cream. Its initial cumulative anomaly score is derived from the weighted summation of individual anomaly contribution values ​​of the multi-dimensional feature comparison vectors of each reference sample. This calculation logic is consistent with that of the test sample, thus ensuring the stability and comparability of the scoring benchmark. By comparing the initial cumulative anomaly score of the test sample with the statistical characteristics (such as minimum and maximum values) of the reference sample set, the anomaly score can be standardized, better reflecting the degree of deviation of the test sample from the normal samples. See Table 1.

[0036] Table 1: Multidimensional Feature Comparison Vector Data Table Characteristic substance name Peak position offset (ppm) Peak area ratio Standard peak area ratio of pure cream Short-chain saturated fatty acids -0.002 0.152 0.165 Long-chain saturated fatty acids +0.005 0.431 0.398 Oleic acid characteristic peak +0.012 0.287 0.215 Linoleic acid characteristic peak -0.001 0.085 0.048 Specific ester characteristic peaks +0.008 0.123 0.072 See Figure 4 This is a biaxial graph comparing the peak parameters of characteristic substances in the detection of adulterated cream. Its core purpose is to show the differences between the tested sample and pure cream in terms of peak area ratio and peak position offset. In food adulteration detection, this type of graph is used for visual difference analysis of spectral features, supporting the detection process of "feature comparison → anomaly quantification → adulteration determination." This graph is a core tool in the feature comparison stage of NMR detection of adulterated cream, intuitively identifying the peak parameter differences between the tested sample and pure cream; providing basic data for subsequent "deviation calculation → anomaly scoring → adulteration determination"; and quickly locating the most sensitive characteristic substances for adulteration identification. Specifically, the feature comparison step extracts the peak position offset and peak area ratio of each characteristic substance from the segmented spectrum of the tested sample through convolutional alignment operations, forming a multi-dimensional feature comparison vector. This vector comprehensively reflects the spectral differences between the tested sample and pure cream in multiple characteristic substances. Secondly, the anomaly quantification step involves compounding the multi-dimensional feature comparison vector with the weighting coefficients of each feature substance in the standard spectral fingerprint database. These weighting coefficients are pre-set based on the inherent content differences of each feature substance in pure cream and pure vegetable fat cream, as well as its importance in adulteration identification. Specifically, the weighting coefficients are determined by statistically analyzing the normalized peak area difference and peak position stability parameters of each feature substance in a large number of pure samples, and then balancing these parameters with a quantification factor. Next, the deviation of each feature comparison data from the standard value of pure cream is calculated, and all deviations are weighted, summed, and standardized to generate a quantified comprehensive anomaly score. Finally, the adulteration determination step classifies the comprehensive anomaly score into different risk levels based on multiple preset adulteration determination thresholds and outputs a structured determination report, achieving automated adulteration conclusion output.

[0037] The multi-dimensional feature comparison vector extracted in the feature comparison step contains key parameters such as the peak position offset and peak area ratio of each feature substance. These parameters are directly used as input data for deviation calculation. The weight coefficients used in the anomaly quantification step are obtained from the standard spectral fingerprint database. Their setting is based on the content difference and peak position stability of the feature substances, ensuring the scientific nature of the deviation calculation and the objectivity of the weight allocation. The comprehensive anomaly score generated by feature comparison and anomaly quantification becomes the direct basis for the adulteration judgment step, enabling subsequent deviation calculations to be based on the actually extracted spectral parameters. The anomaly score improves the sensitivity of the judgment by weighted fusion of multi-feature data, while the adulteration judgment achieves risk classification through threshold comparison. The entire process is progressive, and the basic parameters provided in the feature comparison stage ensure the reliability of the data source and the coherence of the calculation process in subsequent steps.

[0038] In one embodiment of the present invention, multiple adulteration judgment thresholds with progressively increasing values ​​are preset. These adulteration judgment thresholds divide the possibility of adulteration into different level intervals, including "not detected," "low suspicion," "moderate suspicion," and "high suspicion." The calculated comprehensive anomaly score is compared sequentially with the preset multiple adulteration judgment thresholds. Based on the specific level interval in which the comprehensive anomaly score falls, the adulteration risk level of the cream sample to be tested is determined. The specific value of the comprehensive anomaly score and the determined adulteration risk level are recorded. A structured adulteration judgment conclusion report is generated. The adulteration judgment conclusion report clearly lists the unique identification information of the cream sample to be tested. The adulteration judgment conclusion report records in detail the specific value of the calculated comprehensive anomaly score. The adulteration judgment conclusion report clearly marks the adulteration risk level determined based on the comparison results. The adulteration judgment conclusion report includes a summary of the key feature comparison data in the multi-dimensional feature comparison vector, especially the peak position offset and peak area ratio data of feature substances with significant deviation. The adulteration judgment conclusion report provides analytical notes and subsequent processing suggestions based on the current adulteration judgment conclusion. The adulteration determination report is logically correlated with all intermediate data generated during this testing process, including the preprocessed NMR spectrum, the set of segmented spectra of the sample to be tested, the multi-dimensional feature comparison vector, and the key parameters in the calculation process; all correlated data and reports are stored in the test result database, and a unique test record number is assigned to this test; the complete test record stored in the test result database is backed up, and an immutable test process and result summary log is generated simultaneously for subsequent auditing and quality traceability.

[0039] In practice, the process of grading the comprehensive anomaly score based on preset adulteration judgment thresholds begins with setting multiple progressively increasing adulteration judgment thresholds. These thresholds divide the possibility of adulteration into different level ranges, including "not detected," "low suspicion," "moderate suspicion," and "high suspicion." The specific values ​​of these thresholds are determined based on statistical analysis of the comprehensive anomaly score distributions of a large number of known pure cream samples and known adulterated samples. The calculated comprehensive anomaly score is then compared sequentially with the preset adulteration judgment thresholds. Based on the specific level range the comprehensive anomaly score falls into, the adulteration risk level of the cream sample to be tested is determined. The specific value of the comprehensive anomaly score and its determined adulteration risk level are recorded. In practice, this recording operation writes the score and level information into a structured detection result object.

[0040] In some embodiments, the step of outputting the final adulteration determination report specifically involves generating a structured adulteration determination report. This report format uses a document template that includes fixed fields and free text areas. The adulteration determination report clearly lists the unique identification information of the tested cream sample, such as sample number, submission date, and batch number. It also records the detailed numerical value of the calculated comprehensive anomaly score. The report clearly indicates the adulteration risk level determined based on the comparison results. The report includes a summary of key feature comparison data from the multi-dimensional feature comparison vector, particularly the peak position shift and peak area ratio data of significantly deviating feature substances. Optionally, the summary may list the top three deviating feature substances and their comparison data in tabular form. The report provides analytical notes and subsequent processing suggestions based on the current adulteration determination conclusion. For example, for a "moderately suspected" level, the notes suggest retesting or using other detection methods for confirmation.

[0041] Following the step of generating a structured adulteration determination report, the process includes report verification and archiving. Specifically, this involves logically linking the adulteration determination report with all intermediate data generated during the testing process, including preprocessed NMR spectra, the set of segmented spectra of the sample to be tested, multi-dimensional feature comparison vectors, and key parameters from the calculation process. All linked data and the report are then stored in the test results database, and a unique test record number is assigned to each test. The rule for generating this test record number can be expressed as follows: in: This indicates a unique test record number for this test; This represents a pre-defined prefix string for the organization or project code. This represents a string concatenation operation; This string represents the date the test was performed, formatted as a two-digit year, two-digit month, and two-digit day. The sequence number representing the daily detection serial number is fixed in length and padded with leading zeros. A complete backup of the detection records stored in the detection results database is performed, and an immutable detection process and result summary log is generated simultaneously for subsequent auditing and quality traceability. In some embodiments, the immutable summary log is achieved by calculating the hash value of all associated data and writing it to an append-only blockchain or secure log file.

[0042] See Figure 5 This is a trend chart of key parameters in the NMR spectrum preprocessing stage, clearly showing the numerical changes of the five samples under test in four core processing indicators. The Fourier transform signal-to-noise ratio (SNR) generally shows a steady upward trend, with a slight decrease for sample 4, reflecting the clarity of the original signal converted into the spectrum. The higher the value, the less noise interference the effective information of the spectrum is affected by. The baseline calibration error generally shows a small fluctuation and a downward trend, measuring the baseline correction effect. The lower the error, the flatter the spectrum baseline and the more accurate the feature peak identification. The wavelet denoising threshold value is stable at around 5 with minimal fluctuation, indicating high stability of the denoising parameters and good consistency of the preprocessing process, avoiding additional errors introduced by parameter fluctuations. The convolution alignment accuracy remains consistently above 98%, almost stable, ensuring accurate matching between the spectrum under test and the standard fingerprint library spectrum.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream, characterized in that, Includes the following steps: Establish a standard nuclear magnetic resonance spectral fingerprint library for light cream and vegetable fat cream, which includes standard characteristic peak information and weighting coefficients of various characteristic substances; The light cream sample to be tested was subjected to nuclear magnetic resonance detection to obtain the original one-dimensional proton NMR spectrum, and the one-dimensional proton NMR spectrum was preprocessed in multiple steps. From the preprocessed spectrum, multiple specific chemical shift intervals corresponding to various characteristic substances in the standard spectral fingerprint library are segmented and extracted to form a set of segmented spectra of the sample to be tested. Independent baseline calibration and noise suppression are performed on each segment of the spectrum set of the test sample; After noise suppression, each segment of the spectrum is convolved and aligned with the standard feature peak information of the corresponding feature substance in the standard NMR spectrum fingerprint database. The peak position offset and peak area ratio of each segment of the spectrum are extracted to form a multi-dimensional feature comparison vector. The multi-dimensional feature comparison vector is combined with the weight coefficients of each feature substance in the standard spectral fingerprint database to calculate a comprehensive anomaly score that reflects the possibility that the light cream sample to be tested is adulterated with vegetable fat cream. Based on the preset adulteration judgment threshold, the comprehensive anomaly score is graded and evaluated, and the final adulteration judgment conclusion report is output.

2. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 1, characterized in that, The step of establishing a standard nuclear magnetic resonance spectral fingerprint library for light cream and vegetable fat cream, containing standard characteristic peak information and weighting coefficients of various characteristic substances, is specifically implemented as follows: A large number of pure light cream and pure vegetable fat cream samples from known sources were collected and analyzed to obtain their respective one-dimensional proton NMR spectra. The nuclear magnetic resonance spectra of pure light cream and pure vegetable fat cream were analyzed to screen out characteristic substances that showed stable and significant differences in the two types of samples. These characteristic substances include, but are not limited to, characteristic peaks of saturated fatty acids of different chain lengths, characteristic peaks of unsaturated fatty acids, and characteristic peaks of esters. For each selected characteristic substance, its standard characteristic peak information is precisely extracted from its nuclear magnetic resonance spectrum. The standard characteristic peak information includes the standard value of the center chemical shift of the characteristic peak, the standard peak shape profile data, and the standard half-peak width data. Based on the inherent differences in the content of each characteristic substance in light cream and vegetable fat cream and its importance in adulteration identification, a preset weighting coefficient is assigned to each characteristic substance. The standard characteristic peak information of each characteristic substance and its corresponding weight coefficient are systematically stored and managed to form a standard nuclear magnetic resonance spectrum fingerprint library.

3. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 2, characterized in that, The method involves performing nuclear magnetic resonance (NMR) detection on the cream sample to be tested, obtaining the original one-dimensional proton NMR spectrum, and then performing multi-step preprocessing on the one-dimensional proton NMR spectrum. The specific steps include: Under constant magnetic field and temperature conditions, nuclear magnetic resonance scanning was performed on the cream sample to be tested, and the original free induction decay signal was acquired. The acquired free induction decay signal is subjected to Fourier transform to generate the initial time-domain transformed nuclear magnetic resonance spectrum; Phase correction is performed on the initial time-domain transformed nuclear magnetic resonance spectrum to ensure that the peak shapes of all characteristic peaks are symmetrical; Perform global baseline adjustment to eliminate low-frequency baseline drift caused by the instrument or the sample itself; The amplitude of the spectrum after phase correction and baseline adjustment is normalized to eliminate signal intensity differences caused by small fluctuations in sample concentration or instrument parameters.

4. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 3, characterized in that, The process involves segmenting the preprocessed spectrum and extracting multiple specific chemical shift intervals corresponding to various characteristic substances in the standard spectral fingerprint database to form a set of segmented spectra of the sample to be tested. The specific implementation method is as follows: Read the standard nuclear magnetic resonance spectrum fingerprint database to obtain the standard characteristic peak information of each characteristic substance recorded therein; Based on the standard chemical shift values ​​recorded in the standard characteristic peak information, and combined with their standard half-peak width data, the complete chemical shift range of each characteristic substance on the nuclear magnetic resonance spectrum is calculated. On the pre-processed nuclear magnetic resonance spectrum of the sample to be tested, precise spectral cuts are made based on the calculated complete chemical shift range of each characteristic substance. For each extracted spectral image segment, associate it with the corresponding characteristic substance name; All the spectral segments with associated labels are collected to form a structured set of segmented spectra of the sample to be tested.

5. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 4, characterized in that, The detailed process of performing independent baseline calibration and noise suppression on each segment of the spectrum set of the sample to be tested includes: Iterate through each spectral segment in the set of segmented spectra of the sample to be tested; For the spectrum image segment currently being processed, a polynomial fitting method is used to perform local quadratic fitting of its baseline to eliminate any possible local baseline distortion within the spectrum image segment; On the spectrum image segments that have completed baseline calibration, a wavelet transform threshold denoising algorithm is applied to identify and filter out high-frequency random noise signals in the spectrum image segments. Record the baseline calibration parameters and noise suppression parameters used in the independent processing of each spectral image segment; All spectral segments that have undergone independent baseline calibration and noise suppression are updated to the sample segment spectrum set to be tested.

6. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 5, characterized in that, The specific steps for convolving and aligning the noise-suppressed segmented spectra with the standard feature peak information of the corresponding characteristic substances in the standard NMR spectral fingerprint database, extracting the peak position offset and peak area ratio of each segmented spectrum, and forming a multi-dimensional feature comparison vector are as follows: From the standard nuclear magnetic resonance spectrum fingerprint database, retrieve the standard characteristic peak information of each characteristic substance in sequence, especially its standard peak shape profile data; From the updated set of segmented spectra of the test sample, extract the noise-suppressed spectral segment data corresponding to the current characteristic substance; The spectral image segment data is convolved and cross-correlated with the standard peak shape contour data of the current characteristic substance. By finding the maximum value point of the convolution result, the two are precisely aligned in the chemical shift dimension. After the alignment operation is completed, the difference between the actual center chemical shift of the characteristic peak in the spectrum segment and the standard value of the center chemical shift recorded in the standard nuclear magnetic resonance spectrum fingerprint library is calculated as the peak position offset of the characteristic substance. The net peak area of ​​the characteristic peak in the spectrum segment is calculated by integration, and it is compared with the peak area of ​​the specified internal standard reference peak in the spectrum of the same sample to calculate the peak area ratio of the characteristic substance. The peak position offsets and peak area ratios extracted from all characteristic substances are arranged and combined in a predetermined order to form a multi-dimensional feature comparison vector containing all comparison features.

7. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 6, characterized in that, The multi-dimensional feature comparison vector is combined with the weight coefficients of each feature substance in the standard spectral fingerprint database to calculate a comprehensive anomaly score reflecting the possibility that the light cream sample to be tested is adulterated with vegetable fat cream. The calculation logic is as follows: Read the preset weighting coefficients for each characteristic substance from the standard nuclear magnetic resonance spectral fingerprint database; For each feature comparison data in the multi-dimensional feature comparison vector, namely the peak position offset and peak area ratio of each feature substance, the deviation from the standard value of pure cream is calculated. The weighted individual anomaly contribution values ​​corresponding to all feature comparison data in the multi-dimensional feature comparison vector are summed to obtain a preliminary cumulative anomaly score. The preliminary cumulative anomaly score is standardized and mapped to a preset fixed numerical range to obtain the final comprehensive anomaly score. The value of the comprehensive anomaly score directly reflects the likelihood of adulteration.

8. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 7, characterized in that, The comprehensive anomaly score is graded and evaluated according to a preset adulteration judgment threshold. The detailed evaluation process includes: Multiple adulteration detection thresholds with progressively increasing values ​​are pre-set. These adulteration detection thresholds divide the possibility of adulteration into different level ranges, including "not detected", "low suspicion", "moderate suspicion" and "high suspicion". The calculated comprehensive anomaly score is compared sequentially with multiple preset adulteration judgment thresholds; The adulteration risk level of the cream sample to be tested is determined based on the specific level range in which the comprehensive anomaly score falls. Record the specific value of the comprehensive anomaly score and the adulteration risk level it determines.

9. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 8, characterized in that, The steps for outputting the final adulteration determination report specifically include the following: Generate a structured report on adulteration determination; The adulteration determination report clearly lists the unique identification information of the tested cream sample; The adulteration determination report shall record in detail the specific value of the calculated comprehensive anomaly score; The adulteration risk level determined based on the comparison results is clearly marked in the adulteration determination report; The adulteration determination report includes a summary of the key feature comparison data in the multi-dimensional feature comparison vector, especially the peak position offset and peak area ratio data of the feature substances with significant deviation. The adulteration determination report provides analytical notes and subsequent processing suggestions based on the current adulteration determination conclusion.

10. The nuclear magnetic resonance detection method for adulterated vegetable fat cream in light cream according to claim 9, characterized in that, Following the step of generating a structured adulteration determination report, the report also includes a report verification and archiving step, specifically: Logically associate the adulteration determination report with all intermediate data generated during this testing process, including the preprocessed NMR spectrum, the set of segmented spectra of the sample to be tested, the multi-dimensional feature comparison vector, and the key parameters in the calculation process. All related data and reports are stored in the test results database, and a unique test record number is assigned to each test. The complete testing records stored in the testing results database are backed up, and an unalterable testing process and results summary log is generated simultaneously for subsequent auditing and quality traceability.

Images