Simulation degree digital evaluation method and system for microbial protein meat product

Abstract

The invention provides a simulation degree digital evaluation method and system for a microbial protein meat product, and relates to the technical field of digital evaluation. Collecting physical property mechanical data, characteristic volatile organic compound chromatographic data, multi-channel bionic sensing array response data and structured sensory evaluation data of the to-be-detected microbial protein meat product in parallel; preprocessing each modal data, and extracting a key quantitative index in each modal to form a quality characterization vector; according to the method, accurate, objective and standardized digital evaluation of the simulation degree is realized, the accuracy and normalization of an evaluation result are improved, and the defect that the evaluation method neglects mutual influence among indexes is overcome.

Description

Technical Field

[0001] This invention relates to the field of digital evaluation technology, and in particular to a method and system for digitally evaluating the simulation degree of microbial protein meat products. Background Technology

[0002] With the increasing application of microbial protein in the food industry, the development of meat-like products using it as a raw material has gradually become a research hotspot. Microbial protein meat products are nutritionally close to animal meat and have advantages such as high production efficiency and minimal environmental impact. However, existing microbial protein meat products still have certain gaps in sensory attributes such as texture and flavor compared to real meat products. How to scientifically and quantitatively evaluate their "simulation degree" to real meat products has become one of the key issues restricting product optimization and market promotion. Currently, the quality evaluation of microbial protein meat products mostly relies on single-dimensional physicochemical index testing or sensory evaluation.

[0003] Taking the research and development of dried meat products as an example, researchers typically measure the product's textural properties and flavor components separately, and then combine these with sensory evaluation scores for a comprehensive assessment. However, this multi-source data fusion process often faces the problem of unreasonable weight allocation. Different indicators contribute differently to the simulation accuracy, and there may be some degree of mutual influence between the indicators. For example, changes in certain flavor substances may indirectly affect the sensory score. Existing evaluation methods usually use subjective or single objective weighting methods when determining the weights of each indicator, making it difficult to fully consider the information content of the indicator itself and its correlation with other indicators. This may have a certain impact on the accuracy of the evaluation results. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a digital evaluation method and system for the simulation degree of microbial protein meat products, so as to improve the simulation effect of microbial protein meat products.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a digital evaluation method for the simulation degree of microbial protein meat products, the method comprising: Step 1: Parallel acquisition of physical and mechanical data of the microbial protein meat products to be tested, chromatographic data of characteristic volatile organic compounds, response data of multi-channel biomimetic sensor array, and structured sensory evaluation data; preprocessing of each modal data and extraction of key quantitative indicators under each modality to form a quality characterization vector. Step 2: Calculate the information entropy weight of each modality's key quantitative indicators and the conflict weight between each indicator and the remaining indicators. Combine the two to generate the initial static weight of each indicator. Pair the indicator vectors of different modalities, analyze the coupling relationship between any two indicator data sequences, and construct an initial interaction matrix that reflects the strength of mutual influence between cross-modal indicators. Step 3: The initial interaction matrix is ​​analogous to the stress interference field of a multi-crack system in mechanics of materials, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference strength; calculate the net interference effect of each crack source under the combined action of all remaining crack sources, and dynamically correct the initial static weights of the key quantitative indices of each mode based on the net interference effect to generate the corrected dynamic weights. Step 4: Using the corrected dynamic weights, the quality representation vectors are weighted and fused to generate a comprehensive feature expression vector; the comprehensive feature expression vector is input into the pre-trained similarity evaluation model to map and output a quantified simulation score.

[0006] Secondly, a digital evaluation system for the simulation degree of microbial protein meat products includes: The acquisition module is used to collect in parallel the physical and mechanical data of the microbial protein meat products under test, the chromatographic data of characteristic volatile organic compounds, the response data of the multi-channel biomimetic sensor array, and the structured sensory evaluation data. Each modal data is preprocessed, and key quantitative indicators under each modality are extracted to form a quality characterization vector. The information entropy weights of the key quantitative indicators for each modality are calculated, and the conflict weights between each indicator and the remaining indicators are calculated. These two are combined to generate the initial static weights of each indicator. The indicator vectors of different modalities are paired, and the coupling relationship between any two indicator data sequences is analyzed to construct an initial interaction matrix reflecting the intensity of mutual influence between cross-modal indicators. The processing module is used to analogize the initial interaction matrix to the stress interference field of a multi-crack system in materials mechanics, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference intensity. It calculates the net interference effect of each crack source under the combined action of all remaining crack sources. Based on the net interference effect, it dynamically corrects the initial static weights of the key quantitative indices of each mode to generate corrected dynamic weights. Using the corrected dynamic weights, it performs weighted fusion of the quality characterization vector to generate a comprehensive feature expression vector. The comprehensive feature expression vector is input into a pre-trained similarity evaluation model to map and output a quantified simulation score.

[0007] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0008] The above-described solution of the present invention has at least the following beneficial effects: A multi-dimensional and systematic digital evaluation system for the simulation degree of microbial protein meat products was constructed, breaking through the limitations of existing evaluation methods that rely on single-dimensional detection and subjective evaluation. This system achieves comprehensive coverage and quantitative characterization of core quality attributes such as texture, flavor, and sensory experience of simulated meat products, filling the gap in scientific evaluation methods for the simulation degree of microbial protein meat products. Initial static weights were first determined by combining information entropy weights and conflict weights, taking into account both the information content of the indicators themselves and the correlation characteristics between indicators. Then, the weights were dynamically adjusted through multi-crack interference effect analysis, fully considering the mutual influence between cross-modal indicators. This made the weight allocation more in line with the actual evaluation scenario of multi-source data coexistence, improving the objectivity and rationality of the weights.

[0009] By preprocessing multimodal data, extracting features, and weighted fusion, a comprehensive feature expression vector reflecting the overall quality of the sample under test is generated. Combined with a pre-trained similarity evaluation model, the simulation degree is quantitatively output, upgrading the evaluation results from qualitative description to quantitative value. This more intuitively and accurately reflects the similarity between microbial protein meat products and standard meat products, providing an objective and quantifiable basis for product quality judgment. The quantitative simulation degree score can pinpoint the quality gap between imitation meat products and natural meat products, improving the simulation effect of microbial protein meat products. The standardized multimodal data acquisition and processing process reduces the interference of subjective human factors on the evaluation results, improving the repeatability and consistency of the simulation degree evaluation of microbial protein meat products. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the digital evaluation method for the simulation degree of microbial protein meat products provided in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of a digital evaluation system for the simulation degree of microbial protein meat products provided in an embodiment of the present invention. Detailed Implementation

[0012] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0013] like Figure 1 As shown, embodiments of the present invention propose a digital evaluation method for the simulation degree of microbial protein meat products, the method comprising the following steps: Step 1: Parallel acquisition of physical and mechanical data of the microbial protein meat products to be tested, chromatographic data of characteristic volatile organic compounds, response data of multi-channel biomimetic sensor array, and structured sensory evaluation data; preprocessing of each modal data and extraction of key quantitative indicators under each modality to form a quality characterization vector. Step 2: Calculate the information entropy weight of each modality's key quantitative indicators and the conflict weight between each indicator and the remaining indicators. Combine the two to generate the initial static weight of each indicator. Pair the indicator vectors of different modalities, analyze the coupling relationship between any two indicator data sequences, and construct an initial interaction matrix that reflects the strength of mutual influence between cross-modal indicators. Step 3: The initial interaction matrix is ​​analogous to the stress interference field of a multi-crack system in mechanics of materials, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference strength; calculate the net interference effect of each crack source under the combined action of all remaining crack sources, and dynamically correct the initial static weights of the key quantitative indices of each mode based on the net interference effect to generate the corrected dynamic weights. Step 4: Using the corrected dynamic weights, the quality representation vectors are weighted and fused to generate a comprehensive feature expression vector; the comprehensive feature expression vector is input into the pre-trained similarity evaluation model to map and output a quantified simulation score.

[0014] In this embodiment of the invention, a multi-dimensional and systematic digital evaluation system for the simulation degree of microbial protein meat products is constructed. This system overcomes the limitations of existing evaluation methods that rely on single-dimensional detection and subjective evaluation. It achieves comprehensive coverage and quantitative characterization of core quality attributes such as texture, flavor, and sensory experience of simulated meat products, filling the gap in scientific evaluation methods for the simulation degree of microbial protein meat products. First, the initial static weights are determined by combining information entropy weights and conflict weights, taking into account both the information content of the indicators themselves and the correlation characteristics between indicators. Then, the weights are dynamically adjusted through multi-crack interference effect analysis, fully considering the mutual influence between cross-modal indicators. This makes the weight allocation more in line with the actual evaluation scenario of multi-source data coexistence, improving the objectivity and rationality of the weights.

[0015] By preprocessing multimodal data, extracting features, and weighted fusion, a comprehensive feature expression vector reflecting the overall quality of the sample under test is generated. Combined with a pre-trained similarity evaluation model, the simulation degree is quantitatively output, upgrading the evaluation results from qualitative description to quantitative value. This more intuitively and accurately reflects the similarity between microbial protein meat products and standard meat products, providing an objective and quantifiable basis for product quality judgment. The quantitative simulation degree score can pinpoint the quality gap between imitation meat products and natural meat products, improving the simulation effect of microbial protein meat products. The standardized multimodal data acquisition and processing process reduces the interference of subjective human factors on the evaluation results, improving the repeatability and consistency of the simulation degree evaluation of microbial protein meat products.

[0016] In a preferred embodiment of the present invention, step 1 involves the parallel acquisition of physical and mechanical data of the microbial protein meat product to be tested, chromatographic data of characteristic volatile organic compounds, response data of a multi-channel biomimetic sensor array, and structured sensory evaluation data; preprocessing of each modal data, and extraction of key quantitative indicators under each modality to form a quality characterization vector, including: Step 1a: Clean and standardize the physical property data, remove outliers, and extract key quantitative indicators characterizing the texture properties of microbial protein meat products from the processed data to form a physical property index vector; perform baseline correction, peak identification, and normalization on the chromatographic data of characteristic volatile organic compounds, determine the abundance values ​​of each characteristic flavor substance based on the spectrum, and form a flavor substance abundance vector; perform noise reduction and normalization on the response data of the multi-channel biomimetic sensor array, extract steady-state response values ​​or characteristic response patterns from the response curves of each channel to form a sensor response pattern vector; perform consistency verification and standardization on the structured sensory evaluation data, summarize the scores of each evaluation dimension to form a sensory score vector, specifically including: from the texture property dimensions of the microbial protein meat products to be tested, select brightness, redness, yellowness, shear force, hardness, crispness, viscosity, elasticity, etc. The core testing indicators were viscosity, cohesiveness, adhesiveness, chewiness, and resilience. A Canon CR-S400w handheld colorimeter was used to measure color indicators by closely adhering to the sample surface without gaps. After the values ​​stabilized, brightness, redness, and yellowness data were recorded. Each sample was measured six times and the results were recorded. A physical property analyzer with a 10mm diameter circular sampler was used to measure the sample shear force. A pulp column was drilled along the sample fiber direction to complete the test. Each sample was measured five times and the results were recorded. A TA・XTPlus physical property analyzer with an HDP / BS probe was used to measure the sample texture properties. Uniform parameters were set: a pre-test speed of 10mm / s, a test speed of 1mm / s, and a post-test speed of 10mm / s. A full texture analysis was performed on the samples, recording hardness, brittleness, viscosity, elasticity, cohesiveness, adhesiveness, chewiness, and resilience data. Each sample was measured five times and the results were recorded. All collected raw physical property data were cleaned, and outliers were removed using the Grubbs method. The calculation process was as follows: first, the sample mean of all repeated measurements for a certain index was calculated; then, the sample standard deviation of that index was calculated; subsequently, the Grubbs statistic for each data point was calculated, which is equal to the absolute value of the data point and the sample mean divided by the sample standard deviation. The calculated statistic was compared with the critical value at the corresponding significance level (assumed to be 0.05). Data points greater than the critical value were identified as outliers and removed. The cleaned data were then subjected to Z-score standardization to eliminate dimensional differences between indices. The calculation process was as follows: the standardized value of a data point was equal to the original value of the data point minus the mean of all samples for that index, and then divided by the standard deviation of all samples for that index. The standardized values ​​of brightness, redness, yellowness, shear force, hardness, brittleness, viscosity, elasticity, cohesiveness, adhesiveness, chewiness, and resilience indices were extracted after cleaning and standardization. These indices were arranged in a fixed order to form a one-dimensional numerical vector of physical property indices.

[0017] Gas chromatography-mass spectrometry (GC-MS) was used to analyze the samples. First, headspace solid-phase microextraction (HSP) was performed as a pretreatment. 5g of the sample was placed in a 20mL headspace vial, and 1μL of 0.816μg / μL 2-methyl-3-heptanone internal standard was added before sealing. A 50 / 30μm DVB / CAR / PDMS fiber extraction needle was pretreated and activated at 250℃ for 60min in the GC inlet. The activated needle was then inserted into the headspace vial, equilibrated at 50℃ for 10min, and extracted for 30min. After extraction, the needle was desorbed at 250℃ for 6min. The sample was analyzed using split injection mode. Volatile organic compounds in the sample were separated by gas chromatography with a TG-WaxMS polar column. High-purity helium was used as the carrier gas, and the chromatographic parameters were set at a column flow rate of 1.0 mL / min. The chamber temperature was maintained at 40℃ for 3 min, then increased to 200℃ at 5℃ / min and maintained for 2 min, and finally increased to 230℃ at 10℃ / min and maintained for 3 min. The substances were qualitatively identified by mass spectrometry using full scan mode, with a scan range of m / z 40 to 600 and an electron ionization voltage of 70 eV. The gas chromatography-mass spectrometry chromatogram and related information on volatile metabolites of the sample were obtained. Baseline correction was performed on the spectra to eliminate the influence of baseline drift on the data during detection and ensure the accuracy of chromatographic peaks. Peak identification was then performed, combining mass spectrometry databases and retention indices to identify various volatile organic compounds (VOCs) in the sample based on the retention time of the chromatographic peaks and the characteristics of the mass spectra. Subsequently, peak areas of the identified VOCs were normalized. The calculation process was as follows: the abundance value of a VOC equals its peak area divided by the sum of the peak areas of all identified VOCs, multiplied by 100%. Finally, characteristic flavor substances that play a key role in the flavor of the sample were screened using relative odor activity thresholds. A relative odor activity threshold ≥1 was set as the screening criterion, and substances with a relative odor activity threshold <1 or no significant flavor contribution were removed. The abundance values ​​of each characteristic flavor substance, such as artemisia argyi, β-pinene, γ-terpinene, and acetic acid, were extracted after screening and arranged in a fixed order to form a one-dimensional numerical flavor substance abundance vector.

[0018] The multi-channel biomimetic sensor array response data includes electronic nose and electronic tongue response data. Both types of data are acquired simultaneously. In the electronic nose detection, 1.0±0.1g of the crushed and mixed sample is accurately measured and placed in a headspace vial. The sample is equilibrated at 25℃ for 1 hour. Fixed parameters are set: incubation temperature 50℃, incubation time 180s, and injection flow rate 300mL / min. The electronic nose, equipped with 10 sensors including W1C, W5S, and W3C, collects the response signals of volatile substances from the sample. The sensor zeroing time is 10s, the instrument equilibration time is 180s, and the total signal acquisition time is 90s. The entire process is recorded. The process response curve data; In the electronic tongue detection, 10±0.1g of the sample to be tested after being crushed and mixed was mixed with 100mL of ultrapure water, and then crushed by a homogenizer until there were no obvious lumps. After centrifugation at 4℃ and 10000rpm for 10min, 60mL of the filtrate was used as the detection solution after filtration with filter paper. The tasteless solution of 30mM KCl and 0.03mM tartaric acid was used as the reference sample. The uniform parameters of sample detection time of 120s and washing time of 10s were set. The taste response signal of the sample was collected by an electronic tongue equipped with 5 sensors for umami, astringency, saltiness, sourness and bitterness, and the entire response curve data was recorded. The raw response curve data of the electronic nose and electronic tongue are denoised using a moving average method to reduce detection noise. The calculation process is as follows: the denoised response value at a certain moment is equal to the sum of the raw response values ​​at that moment and a fixed number of adjacent moments, divided by the total number of moments involved in the calculation. Subsequently, the denoised response data is normalized to eliminate the difference in response values ​​between sensor channels. The calculation process is as follows: the normalized value of a channel at a certain moment is equal to the raw response value at that moment minus the minimum value of all response values ​​for that channel, divided by the total number of moments involved in the calculation. The difference between the maximum and minimum response values ​​is calculated. Finally, steady-state response values ​​are extracted from the response curves of each channel, namely the response value of the electronic nose signal acquisition at 70 seconds and the average response value of the electronic tongue response curve when it tends to stabilize, as the core detection data. At the same time, invalid data below 0.05 of the tasteless point in the electronic tongue are removed. The steady-state normalized response values ​​of the 10 channels of the electronic nose and the channels of the electronic tongue such as umami and astringency are extracted. The steady-state response values ​​of all channels are arranged in the order of electronic nose and electronic tongue to form a one-dimensional numerical sensor response mode vector.

[0019] Structured sensory evaluation of the tested microbial protein meat products was conducted in a professional sensory analysis laboratory. The evaluation was carried out in groups of five assessment units, with texture, color, aroma, taste, tenderness, and juiciness as the core evaluation dimensions. A graded evaluation standard of 1 to 7 points and corresponding score ranges were established for each dimension, with 6 to 7 points being the best grade and 1 to 2 points the worst grade. Scores for each dimension were recorded for each sample, and each sample underwent multiple repeated evaluations with simultaneous recording of all score data. Consistency testing was performed on all collected raw scoring data using Kendall's coefficient of harmony. The calculation process involved first calculating the rank sum of the scores for each dimension of the same sample in each evaluation, then calculating the sum of squared deviations from the mean of the rank sums, and finally calculating the Kendall's coefficient of harmony. The coefficient is equal to the fixed coefficient 12 calculated by Kendall's coefficient of harmony, multiplied by the sum of squared deviations from the mean of the rank sum, and then divided by the square of the number of assessments multiplied by the cube of the number of assessment dimensions and the difference between the number of assessment dimensions. The closer the coefficient of harmony is to 1, the better the consistency of the assessment data. After removing the scoring data with poor consistency, the valid scoring data is retained. The valid scoring data is subjected to Z-score standardization. The calculation process is as follows: the standardized value of a score is equal to the original value of the score minus the mean of all valid scores of the assessment dimension, and then divided by the standard deviation of all valid scores of the assessment dimension. The standardized score values ​​of the tissue state, color, aroma, taste, tenderness, and juiciness dimensions after consistency testing and standardization are extracted. The assessment dimensions are arranged in a fixed order to form a one-dimensional numerical sensory score vector.

[0020] Step 1b involves combining the physical property index vector, flavor substance abundance vector, sensor response mode vector, and sensory score vector to form a quality characterization vector. Specifically, this includes concatenating the constructed physical property index vector, flavor substance abundance vector, sensor response mode vector, and sensory score vector in a fixed order. This fixed order follows a quality analysis logic that progresses from objective physicochemical texture detection and flavor component detection to the physicochemical response detection of the biomimetic sensor array, and finally to the subjective experience characterization of sensory evaluation. This approach aligns with the cognitive and characterization patterns of microbial protein meat product quality characteristics, moving from objective physicochemical attributes to subjective sensory experiences. It also ensures consistency in the construction of the quality characterization vector and the orderliness of the data dimensions, preventing computational deviations in subsequent weight allocation and feature fusion steps due to chaotic vector concatenation order. Furthermore, it ensures that the quality characterization vectors of different batches and different test samples have a unified dimensional arrangement, achieving comparability between sample analyses and standardization of algorithmic operations. This results in a comprehensive one-dimensional numerical vector, which serves as a quality characterization vector capable of fully characterizing the quality characteristics of the tested microbial protein meat product.

[0021] This embodiment develops targeted preprocessing procedures for different types of multimodal data, effectively eliminating various interference factors such as detection noise, human error, and equipment deviation through operations such as outlier removal, noise reduction, and consistency checks, thereby improving the reliability and effectiveness of data across all dimensions. Standardization of all types of data is achieved through quantitative calculations, uniformly eliminating differences in dimensions and ranges between indicators, making quality indicators of different types and dimensions comparable and synergistic. The key quantitative indicators extracted from each dimension of data are all characteristic indicators that accurately characterize the core quality of microbial protein meat products, and each indicator is constructed into a single-dimensional feature vector in a fixed order, ensuring the accuracy of the characterization of each quality dimension. Furthermore, it achieves the digital and structured expression of quality characteristics; by splicing single-dimensional feature vectors to form a comprehensive quality characterization vector, it realizes the orderly integration of multi-source data, bringing together previously scattered quality information such as texture, flavor, biomimetic sensor response, and sensory evaluation into a whole, comprehensively covering the core quality dimensions of microbial protein meat products, and making up for the deficiency that single-dimensional data cannot fully characterize product quality; the processing flow of multimodal data, index extraction rules, and vector splicing order are all fixed and standardized, ensuring that the quality characterization vectors of different batches and different test samples have consistency and repeatability, providing a unified and standardized numerical carrier for subsequent weight calculation, feature fusion, and simulation quantification analysis.

[0022] In a preferred embodiment of the present invention, step 2, which involves calculating the information entropy weights of key quantification indicators for each modality and calculating the conflict weights between each indicator and the remaining indicators, and combining the two to generate the initial static weights of each indicator, includes: Step 200a: For all key quantitative indicators in the quality characterization vector, calculate the proportion of each sample's indicator value and the information entropy value of the corresponding indicator. Determine the information entropy weight of the corresponding indicator based on the information entropy value. Specifically, this includes: assuming the total number of key quantitative indicators in the quality characterization vector is n and the total number of samples participating in the evaluation is m, forming an indicator value matrix of n indicators and m samples, where each element in the matrix is ​​the standardized value of the corresponding sample under the corresponding indicator; for the j-th indicator, calculate the proportion of the standardized value of the i-th sample under that indicator. The calculation process is as follows: the proportion of the indicator value of the i-th sample under the j-th indicator is equal to the standardized value of that sample under that indicator, divided by the sum of the standardized values ​​of all m samples under the j-th indicator. If the sum of the values ​​of all samples under a certain indicator is 0, then the proportion of the indicator value of each sample under that indicator is set to 1 / m to avoid meaningless calculations; calculate the information entropy value based on the proportion of the sample indicator values ​​under each indicator. The information entropy value reflects the information discrimination ability of the indicator. The smaller the value, the higher the discrimination of the indicator on the sample and the more effective information it contains. The calculation process is as follows: the information entropy value of the j-th indicator... , where constant =1 / m, For the first The first indicator The percentage of an indicator value in a sample; if the percentage of an indicator value in a certain sample... If the ratio is 0, then the product of that ratio and the natural logarithm is... Set it to 0; normalize the information entropy value of each indicator to obtain the information entropy weight of each indicator. The calculation process is as follows: the information entropy weight of the j-th indicator is equal to 1 minus the information entropy value of the j-th indicator, and then divided by the sum of the differences between 1 and the corresponding information entropy value of all n indicators. Finally, the sum of the information entropy weights of all indicators is 1.

[0023] Step 201a: For each indicator, calculate the correlation coefficient between each indicator and the remaining indicators, and use the sum of the absolute values ​​of the correlation coefficients as the conflict weight of the corresponding indicator. Combine the information entropy weight and the conflict weight through multiplication or weighted summation to obtain the initial static weight of each indicator. Specifically, for the j-th indicator, use it as a reference indicator and perform pairwise analysis with each of the remaining n-1 indicators. Calculate the Pearson correlation coefficient between the reference indicator and each comparison indicator. The value of the correlation coefficient reflects the degree of linear association between the two indicators; the larger the absolute value, the higher the association between the indicators. Sum the absolute values ​​of the correlation coefficients between the j-th indicator and all the remaining indicators. This summation result is the original conflict weight of the j-th indicator. The original conflict weight reflects the overall conflict association between the indicator and other indicators; the larger the value, the stronger the association between the indicator and other indicators. The more significant the correlation between the targets, the stronger the conflict characteristics in the evaluation system. The original conflict weights of all n indicators are normalized. The calculation process is as follows: the normalized conflict weight of the j-th indicator = the original conflict weight of the j-th indicator ÷ the sum of the original conflict weights of all n indicators. The sum of the normalized conflict weights of all indicators is 1. The normalized conflict weights are then fused with the information entropy weights obtained in step 200a. Two synthesis methods are provided: multiplicative synthesis and weighted summation. The choice can be made based on the evaluation focus of the microbial protein meat product simulation degree and the actual application scenario. Multiplicative synthesis is suitable for general evaluation scenarios without a clear evaluation focus, where a balance between the effective information content of the indicators themselves and the correlation characteristics between the indicators is desired. Weighted summation is suitable for targeted evaluation scenarios with a clear evaluation orientation, where the emphasis needs to be placed on highlighting the information differentiation ability of the indicators themselves or the coupling correlation characteristics between the indicators. The information entropy weight is denoted as... The range of values ​​is This value is obtained by normalizing the information entropy weight. A value greater than 0 ensures that the effective information content of a single indicator has a basic contribution to the evaluation system, while a value less than 1 prevents the information characteristics of a single indicator from dominating the overall evaluation system. Furthermore, the sum of the information entropy weights of all indicators is 1. The conflict weight is denoted as... The value range is 0 < <1, this value is obtained after normalization of conflict weights. A value greater than 0 ensures that the coupling correlation between indicators is taken into account in weight allocation, while a value less than 1 prevents the conflict characteristics of a single indicator from excessively affecting the weight allocation result. Furthermore, the sum of the conflict weights of all indicators is 1. Both synthesis methods guarantee that the sum of the initial static weights of all indicators after synthesis is 1. The multiplicative synthesis method is as follows: the initial static weight calculation formula for the j-th indicator is... (in This represents the total number of key quantitative indicators in the quality characterization vector. (For the index sequence number), complete the weight normalization; the weighted summation method is to first set the information entropy weight allocation coefficient as λ and the conflict weight allocation coefficient as... Both allocation coefficients are greater than 0 and their sum is 1. Commonly used values ​​for allocation coefficients are... =0.6 and =0.4、 =0.4 and =0.6、 =0.5 and =0.5, where λ=0.6 and 1 =0.4 is suitable for targeted evaluation scenarios that emphasize the ability of indicators to distinguish information, because the core of evaluating the simulation degree of microbial protein meat products is to distinguish the test sample from the standard sample through the characteristics of the indicator itself. Highlighting this dimension can make the weight more in line with the essential needs of product quality characterization. =0.4 and =0.6 is suitable for targeted evaluation scenarios that emphasize the coupling and correlation characteristics between indicators, because the interaction between indicators in a multimodal data fusion system has a significant impact on the simulation results. Strengthening this dimension can improve the adaptability of the weights to the correlation characteristics of multi-source data. =0.5 and =0.5 is suitable for targeted evaluation scenarios that balance the discriminative power of the indicators themselves with the coupling and correlation characteristics between indicators. This is because it achieves a balance between the two weighting features, adapting to targeted evaluation needs without a single-dimensional focus. The initial static weight calculation formula for each indicator is as follows: .

[0024] In this embodiment, the calculation of information entropy weights is entirely based on the standardized values ​​of the indicators themselves and the characteristics of the sample distribution. Objective weight assignment is achieved through quantitative calculation, avoiding interference from subjective experience in weight allocation. Simultaneously, the information entropy value accurately reflects the effective information content of the indicators, allowing indicators with higher discriminative power and greater value for sample evaluation to receive higher weights, thus improving the rationality of weight allocation. The calculation of conflict weights fully considers the linear correlation characteristics between key quantitative indicators, overcoming the shortcomings of the single entropy weight method, which only focuses on the indicator itself and ignores the mutual influence between indicators. This ensures that weight allocation not only reflects the value of the indicator itself but also its correlation characteristics within the entire evaluation system. Two flexible weight fusion methods, multiplicative synthesis and weighted summation, are employed to tailor the weighting to the evaluation focus and actual needs of different microbial protein meat products such as dried meat, meat patties, and sausages. The method seeks to select initial static weights that align with the quality evaluation characteristics of different products, thus improving its adaptability and practicality. All weight calculations are normalized to ensure that the sum of information entropy weights, conflict weights, and the final initial static weights is 1, forming a standardized weight system. This provides a standardized and unified numerical basis for subsequent dynamic weight correction and multimodal data weighted fusion, preventing deviations in subsequent calculations due to non-standard weight values. The improved entropy-weight-CRITIC hybrid method combines the information differentiation advantage of the entropy weight method with the conflict correlation analysis advantage of the CRITIC method. This allows the initial static weights to consider both the information value of the indicators themselves and the interactions between indicators. Compared to single objective or subjective assignment methods, the weight allocation is more aligned with the actual needs of evaluating the simulation degree of microbial protein meat products.

[0025] In a preferred embodiment of the present invention, index vectors of different modalities are paired, the coupling relationship between any two index data sequences is analyzed, and an initial interaction matrix reflecting the strength of mutual influence between cross-modal indicators is constructed, including: Step 200b: Treat each indicator in the quality characterization vector as a node. For any two indicators, extract the standardized data sequences of the corresponding indicators in multiple samples, and calculate the correlation metric between the two sequences. Specifically, for any selected i-th and j-th indicators, extract the standardized indicator values ​​of the two indicators in all participating microbial protein meat product samples and standard meat product samples, forming two standardized data sequences of equal length. The length of the data sequence is consistent with the total number of samples participating in the evaluation (e.g., if there are two samples, semi-fermented beef jerky and semi-fermented microbial protein jerky, the data sequence length is 2; if the batch...). The total number of samples can be adjusted according to actual testing needs to ensure that the data sequences of the two indicators are constructed based on the same batch of samples, thus guaranteeing the effectiveness of the correlation calculation. For example, shear force from the physical property index and β-pinene from the flavor index are selected, and the standardized values ​​of shear force (2173.91±932.05 and 2457.71±44.14 after standardization) and β-pinene (1696.67±250.95 and 3629.82±239.22 after standardization) of semi-fermented beef jerky and semi-fermented microbial protein jerky are extracted respectively, forming two standardized data sequences of length 2.Based on two extracted standardized data sequences, the Pearson correlation coefficient method is used to calculate their correlation measure. This value is the core parameter that quantifies the linear coupling relationship between the two indicators. The specific calculation process is as follows: First, the sample means of the two data sequences are calculated separately. Then, the difference between each value in the first data sequence and the mean of that sequence, and the difference between each value in the second data sequence and the mean of that sequence, are calculated. The two differences at corresponding positions are multiplied and summed to obtain the covariance numerator. Subsequently, the sample standard deviations of the two data sequences are calculated separately, and the two standard deviations are multiplied to obtain the covariance denominator. The final correlation measure is equal to the covariance numerator divided by the covariance denominator. For example, when calculating the correlation measure between shear force and β-pinene, the means of the two standardized data sequences are first calculated, and then the difference between the standardized shear force value and the mean, and the standardized β-pinene value for each sample are calculated sequentially. The numerator is the sum of the differences between the two values ​​and the mean. The denominator is the sum of the corresponding differences. The standard deviations of the two sequences are calculated and multiplied to obtain the denominator. The denominator is the product of the two values. The correlation measure ranges from -1 to 1. The larger the absolute value, the stronger the coupling relationship and the more significant the interaction between the two indicators. A positive value indicates a positive correlation between the two indicators. A change in the value of one indicator will drive the other indicator to change in the same direction. For example, elasticity and cohesion both show positive changes in their standardized values, and the correlation measure is positive. A negative value indicates a negative correlation between the two indicators. A change in the value of one indicator will drive the other indicator to change in the opposite direction. For example, viscosity and chewiness. When the viscosity value increases, the chewiness value decreases, and the correlation measure is negative. A value of 0 indicates that there is no obvious linear coupling relationship between the two indicators, and the interaction can be ignored. Following the above calculation process, the correlation metric between any two index nodes in the quality characterization vector is calculated, achieving a comprehensive quantification of the interaction strength between all cross-modal and same-modal indicators. Cross-modal indicators include combinations of indicators of different dimensions, such as physical and mechanical indicators and flavor substance abundance indicators (e.g., shear force and β-pinene), flavor substance abundance indicators and sensor response mode indicators (e.g., acetic acid and electronic nose W1C sensor response value), and sensor response mode indicators and structured sensory evaluation indicators (e.g., electronic tongue umami sensor response value and sensory taste score). Same-modal indicators include combinations of same-dimensional indicators among physical and mechanical indicators (e.g., hardness and chewiness), among flavor substance abundance indicators (e.g., β-pinene and γ-terpinene), among sensor response mode indicators (e.g., electronic nose W1C and W3C sensor response values), and among structured sensory evaluation indicators (e.g., aroma and taste score).

[0026] Step 201b: The correlation measure is used as the interaction strength between two corresponding indicators, and the diagonal elements of the matrix are set to 1. The interaction strengths between all indicators are arranged in n×n matrix form according to the indicator order, where n is the total number of indicators, thus constructing the initial interaction matrix. Specifically, the correlation measure calculated between any two indicators is directly used as the quantified value of the interaction strength between them. The absolute value of this value reflects the strength of the mutual influence between the indicators; the larger the absolute value, the stronger the mutual influence, and the smaller the absolute value, the weaker the mutual influence. The positive or negative sign of the value reflects the direction of the mutual influence between the indicators. A positive value indicates a positive correlation between the two indicators; a change in the value of one indicator will drive the other indicator to change in the same direction. For example, hardness and adhesion have a positive correlation measure; as hardness increases, adhesion also increases. The value indicates the negative correlation between two indicators; a change in one indicator will cause the other to change in the opposite direction. For example, viscosity and elasticity have a negative correlation; as viscosity increases, elasticity decreases. This quantified value is the core foundational data for subsequent multi-crack interference effect analysis, characterizing the interaction between indicators across various dimensions of physical properties, flavor, sensing, and sensory perception. For instance, the correlation between the electronic nose W2W sensor response value and the sensory aroma score in this example accurately reflects the strength of the interaction between sensing and sensory indicators. The elements on the diagonal of the matrix are uniformly set to 1 because individual indicators do not interact with themselves. Setting them to 1 ensures the integrity of the matrix and the standardization of subsequent multi-crack interference effect algorithm calculations. It also represents the autocorrelation characteristic of the indicators themselves as perfectly correlated. For example, the diagonal element corresponding to shear force is set to 1, and brightness L... The corresponding diagonal element is also set to 1. Based on the total number of indicators n in the quality characterization vector, an n×n square matrix is ​​constructed. The rows and columns of the matrix correspond to the fixed splicing order of the physical property indicator vector, flavor substance abundance vector, sensor response mode vector, and sensory score vector in the quality characterization vector. That is, the first part of the matrix is ​​the interaction strength between physical property indicators, the second part is the interaction strength between physical property indicators and flavor indicators, and so on. The element in the x-th row and y-th column of the matrix is ​​the quantified value of the interaction strength between the x-th indicator and the y-th indicator; for example, the element in the 3rd row (corresponding to shear force) and the 15th column (corresponding to β-pinene) is the correlation measure value between shear force and β-pinene. According to the above assignment rules and arrangement, the interaction strength values ​​between all indicators are sequentially filled into the corresponding positions to form a complete n×n matrix. This matrix is ​​the initial interaction matrix that can comprehensively, systematically and quantitatively reflect the mutual influence strength and direction between all key quantitative indicators. It can be directly used as the basis for analogical analysis of the stress interference field of a multi-crack system, and is suitable for the evaluation needs of different types of microbial protein meat products (dried meat, meat patties) in the embodiments.

[0027] This embodiment treats each indicator as an independent node for interaction analysis, achieving refined decomposition and analysis of multimodal indicators. It breaks down analytical barriers between different modal indicators, capturing cross-modal coupling relationships among indicators across physical properties, flavor, sensing, and sensory dimensions, thus overcoming the shortcomings of evaluation methods that neglect inter-indicator interactions. Correlation metrics are calculated based on standardized data sequences from the same batch of samples, ensuring sample consistency and data validity in the correlation analysis. Furthermore, the Pearson correlation coefficient method is used to quantitatively calculate the intensity of interactions, transforming the previously abstract inter-indicator interactions into calculable and analyzable numerical values, improving the objectivity and accuracy of the analysis results. Comprehensive calculation of correlation metrics between any two indicators achieves seamless coverage of all interactions within the indicator system, including both internal interactions between indicators of the same modality and... This method incorporates external interactions between cross-modal indicators, fully reconstructing the overall interaction network under multi-source indicator coexistence. An initial interaction matrix is ​​constructed in matrix form, systematically integrating the interaction strengths between all indicators. The rows and columns of the matrix correspond to the fixed order of the indicators, making the interaction relationships more intuitive and structurally clear. The initial interaction matrix contains both the intensity and direction of the interactions between indicators, allowing dynamic weight adjustments to accurately reflect the true impact of different indicators under comprehensive interaction. The entire process, from indicator node setting and data sequence extraction to matrix construction, follows fixed and standardized rules, ensuring the repeatability and consistency of the initial interaction matrix construction. Evaluations of different batches and types of microbial protein meat products can obtain standardized matrix results through this process, improving the standardization and universal applicability of the method.

[0028] In a preferred embodiment of the present invention, step 3 includes: Step 300: Treat each element in the initial interaction matrix as a stress interference coefficient between two corresponding indices, and treat each key quantitative index as a crack source. Specifically, this involves: analogizing the initial interaction matrix, which characterizes the interaction relationship between indices, to the stress interference field of a multi-crack system in materials mechanics, and completing the one-to-one correspondence assignment between matrix elements and stress interference coefficients, and between indices and crack sources. Specifically, each element in the initial interaction matrix is ​​directly regarded as a stress interference coefficient between two corresponding key quantitative indices. The absolute value of this coefficient reflects the stress interference intensity generated by the interaction between the two indices; the larger the absolute value, the higher the stress interference intensity. For example, a larger absolute value of the correlation metric between hardness and adhesiveness corresponds to a higher stress interference intensity between the two. The positive or negative value of the numerical value reflects the direction of stress interference generated by the interaction between the two indicators. A positive value indicates positive interference, and a negative value indicates negative interference. For example, a positive value of the correlation metric between hardness and adhesiveness corresponds to positive interference, while a negative value of the correlation metric between viscosity and elasticity corresponds to negative interference. At the same time, each key quantitative indicator in the quality characterization vector is regarded as an independent crack source in a multi-crack system. The total number of crack sources is completely consistent with the total number of indicators in the quality characterization vector. For example, there are 48 key indicators including physical properties, flavor, sensing, and sensory properties, which correspond to 48 independent crack sources.

[0029] Step 301: For each crack source, calculate the net interference effect value of the corresponding crack source under the combined action of all remaining crack sources based on the stress interference coefficients between the corresponding crack source and all remaining crack sources in the initial interaction matrix. This includes: using each off-diagonal element in the initial interaction matrix as the stress interference coefficient between two corresponding crack sources; for the i-th crack source, extracting the stress interference coefficients between the corresponding crack source and all remaining crack sources from the initial interaction matrix; summing all stress interference coefficients and using the summation result as the net interference effect value of the i-th crack source under the combined action of all remaining crack sources. Specifically, this includes: Each off-diagonal element in the initial interaction matrix is ​​used as the effective stress interference coefficient between two different crack sources. The diagonal elements of the matrix are assigned autocorrelation values ​​of the indices themselves and do not participate in the calculation of the net interference effect value, ensuring that the calculation results only reflect the mutual influence between different indices. For example, taking shear force (the 3rd crack source) as an example, its corresponding diagonal element (shear force itself) is not included in the calculation. Only all elements in the 3rd row of the matrix other than the diagonal are extracted (i.e., the stress interference coefficients of shear force and the other 47 indices). For the i-th crack source, according to the row and column arrangement rules of the initial interaction matrix, all effective stress interference coefficients between this crack source and all the remaining crack sources in the matrix are extracted. The number of coefficients extracted is consistent with the number of remaining crack sources, ensuring no omissions or repetitions. If the i-th crack source (e.g., β-pinene) has n-1 remaining crack sources (e.g., 47), then 47 effective stress interference coefficients are extracted, corresponding to the interaction strength of β-pinene and all other indices. All extracted effective stress interference coefficients are directly summed without additional weighting calculations. The final summation is directly used as the net interference effect value of the i-th crack source under the combined action of all remaining crack sources. For example, the 47 effective stress interference coefficients corresponding to β-pinene are directly added together, and the summation result is the net interference effect value of β-pinene. The net interference effect value is a dimensionless quantitative value. The positive or negative value directly reflects the type of comprehensive effect of other indicators on the indicator. A positive value indicates that other indicators have a comprehensive positive enhancing effect on the indicator, and a negative value indicates that other indicators have a comprehensive negative weakening effect on the indicator. The magnitude of the absolute value reflects the intensity of the comprehensive effect of other indicators on the indicator. The larger the absolute value, the more significant the comprehensive effect, and the smaller the absolute value, the weaker the comprehensive effect. For example, if the net interference effect value of the sensory aroma score is positive, it means that other indicators (such as β-pinene and the response value of the electronic nose W2W sensor) have a comprehensive enhancing effect on it, and the larger the absolute value, the more significant the enhancing effect. Following the above calculation process, the net interference effect value of all crack sources (indicators) is calculated, and the degree of influence of each indicator on the combined effect of other indicators is fully quantified, adapting to the dynamic weight correction requirements of different types of microbial protein meat products.

[0030] Step 302: Based on the net interference effect value of each indicator, adjust the initial static weight of the corresponding indicator. When the net interference effect value is an enhancing effect, increase the initial static weight; when the net interference effect value is a weakening effect, decrease the initial static weight. Generate the adjusted dynamic weight for each indicator, including: comparing the net interference effect value of each indicator with zero; if the net interference effect value is greater than zero, the net interference effect value is an enhancing effect; if the net interference effect value is less than zero, the net interference effect value is a weakening effect; for indicators with a net interference effect value greater than zero, multiply the initial static weight of the corresponding indicator by an enhancement coefficient greater than 1 to obtain the adjusted dynamic weight of the corresponding indicator; for indicators with a net interference effect value less than zero... The indicator is adjusted by multiplying its initial static weight by a weakening coefficient less than 1 to obtain its modified dynamic weight. Specifically, this involves comparing the net interference effect value of each indicator with zero to determine the combined effect of other indicators. If the net interference effect value is greater than zero, it is considered an enhancing effect, indicating that the indicator is positively strengthened by the combined effect of all other indicators, and its actual importance in the evaluation system is higher than reflected by the initial static weight. If the net interference effect value is less than zero, it is considered a weakening effect, indicating that the indicator is negatively weakened by the combined effect of all other indicators, and its actual importance in the evaluation system is lower than reflected by the initial static weight. If the value is zero, it indicates that the indicator is neutral due to the combined effect of all other indicators, and the weight remains unchanged from the initial static weight. A uniform enhancement and weakening coefficient are set. The enhancement coefficient is set to a fixed value of 1.20, which is greater than 1 and has a moderate range. This value achieves a reasonable positive amplification of the initial static weight of the enhancement effect indicator without causing the weight to be excessively high due to an excessively large coefficient. The weakening coefficient is set to a fixed value of 0.80, which is less than 1 and greater than 0. This value achieves a reasonable negative reduction of the initial static weight of the weakening effect indicator without causing the weight to be excessively low due to an excessively small coefficient. Both types of coefficients follow the principles of standardization and appropriate adjustment to ensure the rationality and standardization of the adjusted weights. For indicators judged as having an enhancement effect... For indicators with a weakening effect, the initial static weight is multiplied by the enhancement coefficient, and the result is the dynamic weight of the indicator after correction. For indicators with a net interference effect of zero, the dynamic weight remains consistent with the initial static weight. The corrected weights of all indicators are then normalized. The calculation process is as follows: the final dynamic weight of a single indicator = the corrected weight of that indicator ÷ the sum of the corrected weights of all indicators. This process ensures that the sum of the dynamic weights of all indicators after correction is 1, forming a standardized and complete dynamic weight system.

[0031] This embodiment, for the first time, introduces the multi-crack interference effect from materials mechanics into the weight correction of the simulation evaluation of microbial protein meat products. Through scientific model analogy, it transforms the abstract interaction between indicators into a quantifiable stress interference field, achieving a quantitative characterization of the comprehensive effect between indicators and filling the technical gap that weight allocation methods cannot consider the mutual influence between indicators. The calculation of the net interference effect value is based on the initial interaction matrix, quantifying the combined effect of a single indicator on all other indicators through direct summation. The calculation process is simple, intuitive, and verifiable, reflecting the actual impact of indicators in a multi-source data coexistence environment, providing an objective and reliable quantitative basis for weight correction. Based on the positive or negative nature of the net interference effect value, the enhancement and weakening effects are clearly determined, and then... The initial static weights are specifically modified using fixed coefficients, making the weight adjustments more closely reflect the actual importance of the indicators and avoiding the problem that a single static weight cannot adapt to the dynamic interactions between indicators. After the weights are modified, a unified normalization process is performed to ensure the standardization of the dynamic weight system. The sum of the dynamic weights of all indicators is always 1, avoiding deviations in subsequent algorithm calculations due to non-standard weight values ​​and ensuring the continuity of the evaluation process. The dynamic weights fully consider the comprehensive interaction between all indicators across modalities and within the same modality, so that the weight of each indicator reflects both its own information content and conflict characteristics, as well as the comprehensive impact of other indicators on it. Compared with static weights, this is more in line with the actual evaluation of multimodal data fusion of microbial protein meat products, allowing the weights to truly reflect the actual contribution of indicators in the evaluation system.

[0032] In a preferred embodiment of the present invention, step 4 includes: Step 400: The corrected dynamic weights are weighted and multiplied by the corresponding key quantitative indicators in the physical property index vector, flavor compound abundance vector, sensor response mode vector, and sensory score vector contained in the quality characterization vector. The weighted vectors are then concatenated or summed to generate a comprehensive feature expression vector. Specifically, this includes: multiplying the standardized value of each key quantitative indicator by the corrected dynamic weight corresponding to that indicator. The calculation process is: weighted index value = standardized value of the indicator × dynamic weight of the indicator. The above weighting calculation is performed on all indicators in the physical property index vector, flavor compound abundance vector, sensor response mode vector, and sensory score vector. All weighted index values ​​are concatenated according to the original fixed order of physical properties, flavor, sensors, and sensory characteristics to form a one-dimensional comprehensive feature expression vector. This vector not only retains the complete quality characteristic information of the sample, but also reflects the true contribution of each index to the evaluation system through the weighting effect of dynamic weights. The dynamic weights are obtained by comprehensively correcting the information content of the index itself and the interaction between the indicators. The higher the weight value of the index, the higher the proportion of its weighted value in the comprehensive feature expression vector, and the stronger its representation of the overall quality characteristics. The lower the weight value of the index, the lower the proportion of its weighted value, and the weaker its representation. This reflects the actual impact of different indicators on the simulation degree evaluation of microbial protein meat products.

[0033] Step 401: Input the comprehensive feature expression vector into the pre-trained similarity evaluation model. The similarity evaluation model calculates the distance or similarity between the corresponding comprehensive feature expression vector and the comprehensive feature expression vector of the standard meat product sample in the feature space. Specifically, the construction and training process of the similarity evaluation model is as follows: Using the comprehensive feature expression vectors of multiple sets of standard meat product samples and microbial protein meat product samples with different simulation degrees as the training dataset, the comprehensive feature expression vector of the standard meat product samples is used as the positive sample reference, and the comprehensive feature expression vector of the microbial protein meat product samples is used as the sample to be matched. Either Euclidean distance or cosine similarity is selected as the loss function. A single loss function type is clearly defined for model training to avoid evaluation bias caused by mixing multiple loss functions. The model parameters are optimized through iterative training until the similarity or feature distance error calculated by the model reaches a preset threshold. The threshold is set to 0.05. This value ensures the accuracy of the model calculation while avoiding overtraining and overfitting due to an excessively low threshold. This completes the model construction and training, ensuring that the model can accurately calculate the similarity between two vectors. The similarity evaluation model uses either Euclidean distance or cosine similarity (consistent with the loss function type selected during model training). If Euclidean distance is used, the calculation process is to subtract the values ​​at corresponding positions of the two vectors, square the result, sum all the squares, and then take the square root of the sum to obtain the feature space distance. If cosine similarity is used, the calculation process is to multiply the values ​​at corresponding positions of the two vectors, sum the result to obtain the inner product, and then divide by the product of the magnitudes of the two vectors to obtain the similarity value. The model output is a dimensionless value between 0 and 1. The larger the value, the closer the quality characteristics of the sample to the standard meat product are.

[0034] Step 402: Map the distance or similarity to a quantitative simulation score within the range of 0 to 100%, which serves as the final simulation evaluation result for the microbial protein meat product under test. Specifically, if the output of step 401 is a similarity value (value from 0 to 1), then the simulation score = similarity value × 100%; if the output of step 401 is a feature space distance (Euclidean distance, value ≥ 0), first normalize the distance value. The calculation process is: normalized distance = feature space distance ÷ maximum feature space distance (maximum feature space distance is the maximum Euclidean distance between all samples in the training set and the standard sample), and then map it using simulation score = (1 - normalized distance) × 100% to ensure that the score after mapping is still within the range of 0 to 100%.

[0035] This embodiment employs dynamic weighting to weight multimodal indicators, making the simulation evaluation more closely reflect the true importance of the indicators in the system and improving the rationality and accuracy of the evaluation results. The comprehensive feature expression vector formed after weighting retains both objective physicochemical characteristics and subjective sensory characteristics, achieving deep integration of multi-dimensional quality information. A pre-trained similarity evaluation model is used for quantitative comparison, ensuring a stable and repeatable evaluation process and avoiding errors and fluctuations caused by subjective human judgment. The similarity is directly mapped to a simulation score from 0 to 100%, making the results intuitive and easy to understand, facilitating actual production, testing, and product optimization.

[0036] like Figure 2 As shown, embodiments of the present invention also provide a digital evaluation system for the simulation degree of microbial protein meat products, including: The acquisition module is used to collect in parallel the physical and mechanical data of the microbial protein meat products under test, the chromatographic data of characteristic volatile organic compounds, the response data of the multi-channel biomimetic sensor array, and the structured sensory evaluation data. Each modal data is preprocessed, and key quantitative indicators under each modality are extracted to form a quality characterization vector. The information entropy weights of the key quantitative indicators for each modality are calculated, and the conflict weights between each indicator and the remaining indicators are calculated. These two are combined to generate the initial static weights of each indicator. The indicator vectors of different modalities are paired, and the coupling relationship between any two indicator data sequences is analyzed to construct an initial interaction matrix reflecting the intensity of mutual influence between cross-modal indicators. The processing module is used to analogize the initial interaction matrix to the stress interference field of a multi-crack system in materials mechanics, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference intensity. It calculates the net interference effect of each crack source under the combined action of all remaining crack sources. Based on the net interference effect, it dynamically corrects the initial static weights of the key quantitative indices of each mode to generate corrected dynamic weights. Using the corrected dynamic weights, it performs weighted fusion of the quality characterization vector to generate a comprehensive feature expression vector. The comprehensive feature expression vector is input into a pre-trained similarity evaluation model to map and output a quantified simulation score.

[0037] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0038] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0039] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A digital evaluation method for the simulation degree of microbial protein meat products, characterized in that, The method includes: Step 1: Parallel acquisition of physical and mechanical data of the microbial protein meat products to be tested, chromatographic data of characteristic volatile organic compounds, response data of multi-channel biomimetic sensor array, and structured sensory evaluation data; preprocessing of each modal data and extraction of key quantitative indicators under each modality to form a quality characterization vector. Step 2: Calculate the information entropy weight of each modality's key quantitative indicators and the conflict weight between each indicator and the remaining indicators. Combine the two to generate the initial static weight of each indicator. Pair the indicator vectors of different modalities, analyze the coupling relationship between any two indicator data sequences, and construct an initial interaction matrix that reflects the strength of mutual influence between cross-modal indicators. Step 3: The initial interaction matrix is ​​analogous to the stress interference field of a multi-crack system in mechanics of materials, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference strength; calculate the net interference effect of each crack source under the combined action of all remaining crack sources, and dynamically correct the initial static weights of the key quantitative indices of each mode based on the net interference effect to generate the corrected dynamic weights. Step 4: Using the corrected dynamic weights, the quality representation vectors are weighted and fused to generate a comprehensive feature expression vector; the comprehensive feature expression vector is input into the pre-trained similarity evaluation model to map and output a quantified simulation score.

2. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 1, characterized in that, Each modal data is preprocessed separately, and key quantitative indicators for each modality are extracted to form a quality characterization vector, including: The physical property and mechanical data were cleaned and standardized to remove outliers. Key quantitative indicators characterizing the texture of microbial protein meat products were extracted from the processed data to form a physical property index vector. Baseline correction, peak identification, and normalization were performed on the characteristic volatile organic compound chromatographic data. The abundance values ​​of each characteristic flavor substance were determined based on the spectra to form a flavor substance abundance vector. Noise reduction and normalization were performed on the multi-channel biomimetic sensor array response data. Steady-state response values ​​or characteristic response patterns were extracted from the response curves of each channel to form a sensor response pattern vector. Consistency checks and standardization were performed on the structured sensory evaluation data. The scores of each evaluation dimension were summarized to form a sensory score vector. The physical property index vector, flavor compound abundance vector, sensor response mode vector, and sensory score vector are combined to form a quality characterization vector.

3. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 2, characterized in that, Calculate the information entropy weights of the key quantification indicators for each modality, and calculate the conflict weights between each indicator and the remaining indicators. Combine these two calculations to generate the initial static weights for each indicator, including: For all key quantitative indicators in the quality characterization vector, calculate the proportion of each sample's indicator value and the information entropy value of the corresponding indicator, and determine the information entropy weight of the corresponding indicator based on the information entropy value. For each indicator, the correlation coefficient between each indicator and the remaining indicators is calculated, and the sum of the absolute values ​​of the correlation coefficients is used as the conflict weight of the corresponding indicator. The information entropy weight and the conflict weight are combined by multiplication or weighted summation to obtain the initial static weight of each indicator.

4. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 3, characterized in that, By pairing index vectors from different modalities, the coupling relationship between any two index data sequences is analyzed, and an initial interaction matrix reflecting the strength of mutual influence between cross-modal indices is constructed, including: Each indicator in the quality characterization vector is regarded as a node. For any two indicators, the standardized data sequence of the corresponding indicator in multiple samples is extracted, and the correlation measure between the two sequences is calculated. The correlation measure is used as the interaction strength between two corresponding indicators, and the diagonal elements of the matrix are set to 1. The interaction strengths between all indicators are arranged in the order of the indicators into an n×n matrix, where n is the total number of indicators, thus constructing the initial interaction matrix.

5. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 4, characterized in that, Step 3 includes: Each element in the initial interaction matrix is ​​regarded as the stress interference coefficient between the corresponding two indices, and each key quantitative index is regarded as a crack source. For each crack source, the net interference effect value of the corresponding crack source under the combined action of all remaining crack sources is calculated based on the stress interference coefficient between the corresponding crack source and all remaining crack sources in the initial interaction matrix. Based on the net interference effect value of each indicator, the initial static weight of the corresponding indicator is adjusted. When the net interference effect value is an enhancing effect, the initial static weight is increased; when the net interference effect value is a weakening effect, the initial static weight is decreased, thus generating the adjusted dynamic weight of each indicator.

6. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 5, characterized in that, For each crack initiation, based on the stress interference coefficients between the corresponding crack initiation and all remaining crack initiations in the initial interaction matrix, calculate the net interference effect value of the corresponding crack initiation under the combined action of all remaining crack initiations, including: Each off-diagonal element in the initial interaction matrix is ​​used as the stress interference coefficient between the two corresponding crack sources. For the i-th crack source, extract the stress interference coefficients between the corresponding crack source and all remaining crack sources from the initial interaction matrix; sum all stress interference coefficients and use the summation result as the net interference effect value of the i-th crack source under the combined action of all remaining crack sources.

7. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 6, characterized in that, Based on the net interference effect value of each indicator, the initial static weights of the corresponding indicators are adjusted. When the net interference effect value is an enhancing effect, the initial static weights are increased; when the net interference effect value is a weakening effect, the initial static weights are decreased. This generates the adjusted dynamic weights for each indicator, including: The net interference effect value of each indicator is compared with zero. If the net interference effect value is greater than zero, the net interference effect value is a strengthening effect; if the net interference effect value is less than zero, the net interference effect value is a weakening effect. For indicators with a net interference effect value greater than zero, the initial static weight of the corresponding indicator is multiplied by an enhancement coefficient greater than 1 to obtain the dynamic weight of the corresponding indicator after correction; for indicators with a net interference effect value less than zero, the initial static weight of the corresponding indicator is multiplied by a weakening coefficient less than 1 to obtain the dynamic weight of the corresponding indicator after correction.

8. The digital evaluation method for the simulation degree of microbial protein meat products according to claim 7, characterized in that, Step 4 includes: The modified dynamic weights are multiplied by the key quantitative indicators corresponding to the physical property index vector, flavor substance abundance vector, sensor response mode vector and sensory score vector contained in the quality characterization vector, and the weighted vectors are concatenated or summed to generate a comprehensive feature expression vector. The comprehensive feature expression vector is input into a pre-trained similarity evaluation model, which calculates the distance or similarity between the corresponding comprehensive feature expression vector and the comprehensive feature expression vector of the standard meat product sample in the feature space. The distance or similarity is mapped to a quantitative simulation score within the range of 0 to 100%, which is used as the final simulation evaluation result of the microbial protein meat product under test.

9. A digital evaluation system for the simulation degree of microbial protein meat products, wherein the system implements the method as described in any one of claims 1 to 8, characterized in that, include: The acquisition module is used to collect in parallel the physical and mechanical data of the microbial protein meat products under test, the chromatographic data of characteristic volatile organic compounds, the response data of the multi-channel biomimetic sensor array, and the structured sensory evaluation data. Each modal data is preprocessed and key quantitative indicators under each modality are extracted to form a quality characterization vector. The information entropy weight of each modality key quantitative indicator is calculated, and the conflict weight between each indicator and the remaining indicators is calculated. The two are combined to generate the initial static weight of each indicator. By pairing index vectors of different modalities, analyzing the coupling relationship between any two index data sequences, and constructing an initial interaction matrix that reflects the strength of mutual influence between cross-modal indicators; The processing module is used to analogize the initial interaction matrix to the stress interference field of a multi-crack system in mechanics of materials, where each index is regarded as a crack source and the interaction strength between the indices is analogous to the stress interference strength. Calculate the net interference effect of each crack source under the combined action of all remaining crack sources. Based on the net interference effect, dynamically correct the initial static weights of the key quantitative indicators of each mode to generate the corrected dynamic weights. Using the modified dynamic weights, the quality representation vectors are weighted and fused to generate a comprehensive feature expression vector. The comprehensive feature representation vector is input into the pre-trained similarity evaluation model, and the output is a quantified simulation score.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 8.

Images