Method and system for automated monitoring and management of high-moisture vegetable protein meat processing

Abstract

The application provides an automatic monitoring and management method and system for high-moisture plant protein meat processing, comprising: performing optical path transmission loss coefficient correction processing according to a spectral data matrix, eliminating environmental interference by using a temperature compensation correction factor and a humidity influence correction parameter, extracting feature peak position coordinates by first derivative transformation to obtain standardized spectral feature data after environmental correction; constructing a sampling time interval sequence by using component concentration data of each detection point, correcting time drift influence by using a signal strength attenuation curve, and performing interpolation calculation on the component concentration gradient distribution according to spatial coordinate positioning information to obtain a spatial distribution map reflecting the overall component state of the raw material. The deficiencies of traditional chemical analysis and manual judgment in the raw material processing process are solved, the automation level and efficiency of plant protein meat production are improved, and the accuracy of raw material proportioning is improved.

Description

Technical Field

[0001] This invention relates to the field of production monitoring and management technology, and in particular to an automated monitoring and management method and system for high-moisture processing of plant-based meat. Background Technology

[0002] High-humidity plant-based meat processing is an important direction for the transformation and upgrading of the food industry. Through extrusion molding under high temperature and high humidity conditions, plant protein raw materials are transformed into food with the texture of meat. Its production efficiency and product quality directly affect the development level of the entire plant-based food industry.

[0003] A similar prior art is Chinese Patent Publication No. CN119574488A, which proposes a method and system for intelligent protein quality detection based on spectral analysis. This method includes: collecting infrared spectral data of a preset number of protein samples; determining the local fluctuation coefficient of transmittance at each wavelength; determining the fluctuation coefficient of the infrared spectral data of each protein sample; based on the fluctuation coefficient, screening the infrared spectral data of all protein samples to obtain fluctuation data of each suspected impurity; dividing each suspected impurity fluctuation data into segments; determining the anomaly index of the suspected impurity fluctuation data within each segment; correcting the spatial domain weight and value domain weight of each filter window in the bilateral filtering algorithm based on the anomaly index to denoise the infrared spectral data of the protein samples; determining the difference coefficient of the infrared spectral data of each protein sample, and detecting protein quality. This improves the accuracy of protein quality detection.

[0004] Furthermore, a similar prior art exists in Chinese Patent Publication No. CN119643500A, which proposes a method and system for efficient and rapid detection of protein content and water content in plant-based products based on near-infrared spectroscopy and deep learning. The method includes: obtaining spectral data of plant-based samples and experimentally determining protein content and water content; combining multi-step data preprocessing and adaptive dimensionality reduction to extract feature signals highly correlated with target detection; constructing a high-precision detection model using a deep learning model based on convolutional neural networks and self-attention mechanisms; obtaining a final stable detection model through multiple verifications and optimizations of the model; and integrating this model with a portable spectrometer to form a detection system. The detection system includes a 3D-printed shell, a spectral sensor, a conversion line, and a data processing unit. It allows for rapid and non-destructive detection of protein content and water content in samples with a single click, without the need for chemical reagents. This invention offers advantages such as high detection speed, high portability, ease of operation, and wide applicability, making it particularly suitable for applications requiring efficient, real-time on-site detection.

[0005] The above-mentioned technical solutions all solve the problem of component detection through spectroscopy. However, although these methods can improve the detection accuracy to some extent, they still fail to solve the problem of real-time, multi-dimensional component detection of plant protein raw materials in high-humidity processing, especially the complex interference factors that exist in the mixing of multiple raw materials and production process. Summary of the Invention

[0006] In a first aspect, the present invention provides an automated monitoring and management method for high-moisture processing of plant-based meat, comprising:

[0007] Multi-point spectral data acquisition of plant protein raw materials; establishment of detection grid on the surface of raw materials based on spatial coordinate positioning information; acquisition of spectral band segmentation marks and absorption intensity range of each detection point by scanning within a preset wavelength range; and obtaining a spectral data matrix containing spatial location information.

[0008] The optical path transmission loss coefficient is corrected based on the spectral data matrix. Environmental interference is eliminated by using temperature compensation correction factor and humidity influence correction parameter. The position coordinates of characteristic peaks are extracted by first derivative transformation to obtain standardized spectral characteristic data after environmental correction.

[0009] Principal component analysis is used to compress the vector dimension of the standardized spectral feature data, extract the principal components whose cumulative variance contribution rate reaches a preset value, and use the feature weight allocation coefficient to calculate the importance ranking of each principal component to obtain a set of low-dimensional feature vectors representing different plant protein raw material types.

[0010] The inter-class distance measurement parameters are calculated based on the set of low-dimensional feature vectors. The raw material type classification of each detection point is determined by Euclidean distance calculation, and classification result data containing raw material type information is obtained.

[0011] According to a preferred embodiment of the present invention, the acquisition of a spectral data matrix includes:

[0012] Based on the geometric dimensions and surface characteristics of the plant protein raw materials, a grid division algorithm is used to establish a uniformly distributed detection grid on the surface of the raw materials to obtain the spatial coordinate information of each grid node. The detection points corresponding to the spatial coordinates are spectrally scanned by a near-infrared spectrometer to collect spectral response signals within a preset wavelength range, thereby obtaining spectral data for each detection point. The preset range is divided into N characteristic bands by a band segmentation algorithm, and the spectral data matrix is ​​constructed based on the coordinate information of each detection point, the identifier of each characteristic band, and the spectral data.

[0013] According to a preferred embodiment of the present invention, the acquisition of standardized spectral characteristic data includes:

[0014] Based on the intensity values ​​of each wavelength point in the spectral data matrix, the transmission attenuation coefficient is calculated using the optical path loss correction model, and the optical path loss correction parameter table is obtained.

[0015] Based on the aforementioned correction parameter table, the spectral intensity is linearly corrected using a temperature compensation factor to obtain temperature-corrected spectral response data.

[0016] Based on the temperature-corrected spectral response data, the amplitude of each band absorption peak is adjusted by humidity correction parameters to obtain the purified spectral signal after environmental interference is eliminated.

[0017] The purified spectral signal is differentiated using a first-order derivative transformation algorithm to extract the slope change information of the spectral curve, determine the wavelength position of the characteristic peak, and obtain the standardized spectral feature data.

[0018] According to a preferred embodiment of the present invention, obtaining a set of low-dimensional feature vectors includes:

[0019] Obtain a standardized spectral feature data matrix and calculate the covariance relationship between each feature dimension. Based on the covariance matrix, use principal component analysis (PCA) to calculate eigenvalues ​​and eigenvectors, and sort them in descending order of eigenvalue size to determine the contribution distribution of principal components. Using the cumulative variance contribution rate calculation formula, sum the variance contribution values ​​of each principal component to determine the number of retained principal components. Use feature weight allocation coefficients to evaluate the importance of the retained principal components, calculate the explanatory power of each principal component in the feature space, and obtain a set of low-dimensional feature vectors representing different plant protein raw material types based on the principal component importance ranking results.

[0020] A preferred embodiment of the present invention includes obtaining classification result data containing raw material type information, comprising:

[0021] A low-dimensional feature vector set of data is acquired, and the Euclidean distance measurement parameter between each raw material type is calculated. A distance matrix is ​​constructed to record the spatial distribution relationship between different types. The K-means clustering algorithm is used to optimize and adjust the coordinate points of the cluster centers, calculate the optimal center position for each raw material type, and generate a stable type coding label mapping relationship. Through the type coding label mapping relationship, a database corresponding to raw material types and coding identifiers is established to record the feature vector range and classification boundary information of each type. For the feature vector data to be detected, the Euclidean distance value between it and the coordinate points of each cluster center is calculated, and the Euclidean distance value is used as the measurement parameter. Based on the measurement parameter and the raw material type database, a classification result data set containing complete raw material type information is generated.

[0022] According to a preferred embodiment of the present invention, after obtaining classification result data containing raw material type information, the method further includes:

[0023] Based on the classification results, the component concentration gradient distribution of each detection point is analyzed. A quantitative relationship model between spectral features and protein content is established using the partial least squares regression algorithm. The component content prediction values ​​at different depths are obtained by dividing the detection depth into layers, thus obtaining the component concentration data of each detection point.

[0024] According to a preferred embodiment of the present invention, obtaining the spatial distribution map of the overall composition state of the raw materials further includes:

[0025] By constructing a sampling time interval sequence using the component concentration data at each detection point, the influence of time drift is corrected using the signal intensity attenuation curve, and the component concentration gradient distribution is interpolated based on the spatial coordinate positioning information to obtain a spatial distribution map reflecting the overall component state of the raw material.

[0026] The deviation between the spatial distribution map and the preset component ratio standard is calculated. The weighted average algorithm is used to adjust the raw material mixing ratio according to the characteristic weight allocation coefficient of each detection point, and the final raw material ratio scheme is determined and output to the production control system.

[0027] In a preferred embodiment of the present invention, the preset wavelength range is 1000-2500nm.

[0028] Secondly, the present invention also provides an automated monitoring and management system for high-moisture processing of plant-based meat, for implementing the above-mentioned method, the system comprising:

[0029] The spectral data acquisition module is used to acquire multi-point spectral data of plant protein raw materials using a near-infrared spectrometer. It establishes a detection grid on the surface of the raw materials based on spatial coordinate positioning information, and obtains the spectral band segmentation marks and absorption intensity value range of each detection point by scanning through a preset wavelength range, thus obtaining a spectral data matrix containing spatial location information.

[0030] The environmental correction module is used to perform optical path transmission loss coefficient correction processing based on the spectral data matrix, eliminate environmental interference by using temperature compensation correction factor and humidity influence correction parameter, and extract the position coordinates of characteristic peaks through first derivative transformation to obtain standardized spectral feature data after environmental correction.

[0031] The feature extraction module is used to control the vector dimension compression ratio of the standardized spectral feature data through principal component analysis, extract principal components whose cumulative variance contribution rate reaches a preset value, calculate the importance ranking of each principal component using feature weight allocation coefficients, and obtain a set of low-dimensional feature vectors representing different plant protein raw material types.

[0032] The classification and recognition module is used to calculate the inter-class distance measurement parameters based on the set of low-dimensional feature vectors, determine the raw material type classification of each detection point through Euclidean distance calculation, and obtain classification result data containing raw material type information.

[0033] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0034] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0035] This invention provides an automated monitoring and management method and system for high-humidity plant protein meat processing based on near-infrared spectroscopy. The system utilizes multi-point spectral data acquisition, environmental correction, principal component analysis, low-dimensional feature extraction, and classification to achieve rapid, non-destructive detection, accurate classification, and automatic proportioning adjustment of plant protein raw materials. By establishing a spectral data matrix of plant protein raw materials, and correcting for optical path loss and environmental interference, standardized spectral feature data is extracted. Principal component analysis is used for dimensionality reduction to obtain a set of low-dimensional feature vectors, effectively classifying different raw material types. The system further establishes a quantitative relationship model between spectral features and protein content using partial least squares regression, enabling accurate prediction of component concentration. This method allows for the construction of a spatial distribution map reflecting the state of raw material components, real-time adjustment of raw material proportioning, ensuring product quality and stability, and overcoming the shortcomings of traditional chemical analysis and manual judgment in raw material processing. It improves the automation level and efficiency of plant protein meat production, while also enhancing the accuracy of raw material proportioning, ensuring the consistency of the product's nutritional value and taste characteristics. Attached Figure Description

[0036] Figure 1 This is a flowchart of an automated monitoring and management method for high-humidity plant protein meat processing in an embodiment of the present invention;

[0037] Figure 2 This is a flowchart of the method for obtaining standardized spectral feature data in an embodiment of the present invention;

[0038] Figure 3 This is a flowchart of the method for obtaining the overall compositional state space distribution map of raw materials in an embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the automated monitoring and management system for high-moisture processing of plant-based meat according to an embodiment of the present invention. Detailed Implementation

[0040] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0041] like Figure 1-4 The automated monitoring and management method and system for high-moisture processing of plant-based meat in this embodiment may specifically include:

[0042] Step S1: Use a near-infrared spectrometer to collect spectral data from multiple points on the plant protein raw material. Establish a detection grid on the surface of the raw material based on the spatial coordinate positioning information. Obtain the spectral band segmentation mark and absorption intensity value range of each detection point by scanning within a preset wavelength range, and obtain a spectral data matrix containing spatial location information.

[0043] Specifically, in the high-humidity processing of plant-based meat, near-infrared spectrometers are used to collect multi-point spectral data of the plant protein raw materials. Based on the geometric dimensions and surface characteristics of the raw materials, a grid-based algorithm is used to establish a uniformly distributed detection grid on the surface, with each grid node corresponding to a detection point. The spectrometer scans each detection point, collecting spectral response signals within a preset wavelength range. The collected signals include spectral intensity values ​​at multiple wavelengths. A band segmentation algorithm divides the preset spectral bands into several feature bands, assigning a unique identifier to each feature band. The system combines these collected spectral data with the spatial coordinate information of each detection point to construct a spectral data matrix containing spatial location information, band identifiers, and absorption intensity values. This technical solution not only provides a foundation for subsequent data analysis but also lays the data foundation for feature extraction, classification, and quantitative analysis.

[0044] Step S2: Perform optical path transmission loss coefficient correction processing based on the spectral data matrix, use temperature compensation correction factor and humidity influence correction parameter to eliminate environmental interference, extract the position coordinates of characteristic peaks through first derivative transformation, and obtain standardized spectral feature data after environmental correction.

[0045] Specifically, due to environmental factors such as temperature and humidity, the quality of spectral acquisition may be affected. The system first calculates the transmission attenuation coefficient using an optical path loss correction model to compensate for signal attenuation caused by optical path transmission loss. This process calculates the attenuation coefficient for each detection point based on the Lambert-Beer law and generates an optical path loss correction parameter table. Next, the spectral intensity is corrected based on temperature and humidity: a temperature compensation factor is used to correct for temperature effects in the spectral data, while humidity correction parameters are used to adjust the amplitude of absorption peaks in each band, thereby eliminating the influence of humidity on the spectral data. After this series of environmental interference corrections, the obtained spectral data becomes more accurate and stable. The system uses a first-order derivative transformation algorithm to extract the precise location coordinates of characteristic peaks, obtaining standardized spectral feature data, providing accurate input for subsequent feature extraction and data analysis.

[0046] Step S3: The standardized spectral feature data is compressed by controlling the vector dimension ratio using principal component analysis. Principal components with cumulative variance contribution rates reaching a preset value are extracted. The importance ranking of each principal component is calculated using feature weight allocation coefficients to obtain a set of low-dimensional feature vectors representing different plant protein raw material types.

[0047] Specifically, Principal Component Analysis (PCA) is used to compress the vector dimension of standardized spectral feature data, reducing data dimensionality while retaining the maximum amount of key information. By calculating the covariance matrix of the spectral data, the PCA algorithm extracts the principal components in the data, and controls the compression ratio according to the cumulative variance contribution rate to ensure that the retained principal components can explain most of the data variation. The system selects the principal components with the cumulative variance contribution rate that reaches the preset value by sorting the feature values, and obtains a set of low-dimensional feature vectors. The system uses feature weight allocation coefficients to evaluate the importance of each principal component, and generates a set of low-dimensional feature vectors representing different plant protein raw material types based on these importance rankings. The above technical solution reduces data redundancy, extracts the most representative features, reduces computational complexity, and retains sufficient raw material information, providing accurate feature representation for subsequent classification and identification.

[0048] Step S4: Calculate the inter-class distance measurement parameters based on the set of low-dimensional feature vectors, determine the raw material type classification of each detection point by Euclidean distance calculation, and obtain classification result data containing raw material type information.

[0049] Specifically, the system measures the similarity between different types of raw materials by calculating the Euclidean distance, and constructs a distance matrix to record the spatial distribution relationships between different types. To further optimize classification, the system uses the K-means clustering algorithm to optimize and adjust the coordinates of the cluster centers, thereby calculating the optimal center position for each type of raw material. Each cluster center corresponds to one type of plant protein raw material, and the system generates a stable type coding label for each raw material type. These labels help the system accurately identify and label different types of plant protein raw materials. Through Euclidean distance and clustering algorithms, the system can assign corresponding raw material type classification results to each detection point, generating a classification dataset containing raw material type information. The above technical solution achieves accurate classification of different plant protein raw material types, providing real-time data for raw material identification and quality control in the production process.

[0050] Furthermore, the acquisition of the spectral data matrix includes:

[0051] Based on the geometric dimensions and surface characteristics of plant protein raw materials, a grid division algorithm is used to establish a uniformly distributed detection grid on the surface of the raw materials to obtain the spatial coordinate information of each grid node;

[0052] The detection points corresponding to the spatial coordinates are spectrally scanned using a near-infrared spectrometer, and spectral response signals within a preset wavelength range are collected to obtain spectral data for each detection point.

[0053] The preset range is divided into N feature bands by a band segmentation algorithm, and the spectral data matrix is ​​constructed based on the coordinate information of each detection point, the identifier of each feature band, and the spectral data.

[0054] Specifically, firstly, a laser rangefinder is used to perform a three-dimensional scan of the plant protein raw material surface, establishing a three-dimensional coordinate system with the origin as the reference point to obtain the spatial coordinate information of the sample surface. Based on the obtained spatial coordinate data, a grid division algorithm is used to divide the raw material surface into multiple detection grids. Then, the spectral scanning module of the near-infrared spectrometer is activated, and the surface of the plant protein raw material is scanned within a preset wavelength range to obtain spectral data at multiple wavelength points. For each detection point, the spectrometer automatically collects the reflectance spectral intensity value of that point. A preset band segmentation algorithm is used to divide the preset range into N characteristic bands, where N is a positive integer greater than or equal to 2, and each band is assigned a unique identifier, for example, from B001 to B015. The system records the absorption intensity values ​​of each detection point in different bands in real time, i.e., the aforementioned spectral data. Finally, the spatial coordinate information, band identifiers, and absorption intensity values ​​are integrated to form a spectral data matrix, where the first dimension represents the number of detection points, the second dimension represents the number of wavelength points, and the third dimension stores the wavelength value, absorption intensity, and spatial coordinate information, respectively. The above technical solution provides a complete data foundation for subsequent spectral data analysis and the establishment of a protein content prediction model.

[0055] Furthermore, the acquisition of standardized spectral feature data, such as Figure 2 As shown, it includes:

[0056] Based on the intensity values ​​of each wavelength point corresponding to each detection point in the spectral data matrix, the transmission attenuation coefficient is calculated using an optical path loss correction model, and the average loss coefficient is calculated based on the optical path loss correction parameter table. When the average loss coefficient is greater than a set value, the spectral intensity is linearly corrected using a temperature compensation factor to obtain temperature-corrected spectral response data. Based on the temperature-corrected spectral response data, the amplitude of each band absorption peak is adjusted using humidity correction parameters to obtain a purified spectral signal after environmental interference is eliminated. The purified spectral signal is differentiated using a first-order derivative transformation algorithm to extract the slope change information of the spectral curve, determine the wavelength position of the characteristic peak, and obtain the standardized spectral feature data.

[0057] Specifically, since the spectral data extraction process may be affected by the optical path and temperature and humidity, resulting in inaccurate spectral data, an optical path loss correction model is used to calculate the transmission attenuation coefficient based on the intensity values ​​of each wavelength point corresponding to each detection point in the spectral data matrix. This involves establishing a loss compensation algorithm based on the Lambert-Beer law, normalizing the original spectral intensity of each detection point using the reflectivity reference value of the built-in standard reflector, and calculating the output optical path transmission loss coefficient for each detection point. ,That Standard reflection intensity, The measured intensity value is the original spectral intensity. Based on the transmission attenuation coefficient of each detection point, an optical path loss correction parameter table is generated. The average loss coefficient of multiple detection points within a preset range is calculated based on this parameter table. When the average loss coefficient exceeds a set value, influencing factors, namely temperature and humidity, are determined using environmental data. If the ambient temperature exceeds a preset range, a temperature compensation factor is used to linearly correct the spectral intensity, obtaining temperature-corrected spectral response data. Based on the temperature-corrected spectral response data, the amplitude of absorption peaks in each band is adjusted using humidity correction parameters to obtain a purified spectral signal after environmental interference is eliminated. For example, the temperature sensor records the ambient temperature as the measured temperature, and the relative humidity sensor displays the ambient humidity as the measured humidity. The system then calls a preset temperature compensation correction algorithm. ,in, For temperature coefficient, As the reference temperature, The temperature correction factor was calculated based on the measured temperature. Simultaneously apply humidity effect correction parameters , Humidity coefficient Based on the baseline humidity, the humidity correction factor is calculated as follows: The system sequentially multiplies each intensity value in the spectral data matrix by a loss coefficient, a temperature correction factor, and a humidity correction factor to complete environmental interference elimination processing, obtaining a purified spectral signal. A first-order derivative transform algorithm is then used to differentiate the purified spectral signal, extracting the slope change information of the spectral curve to determine the precise wavelength position of the characteristic peaks. This is determined by the characteristic that the slopes on both sides of the peak position satisfy the rule that the left side is positive and the right side is negative. Based on the characteristic peak positions, a wavelength-intensity coordinate mapping relationship is established to obtain a characteristic peak distribution matrix containing coordinate position information. The characteristic peak distribution matrix is ​​normalized using a standardization processing algorithm to eliminate numerical differences under different detection conditions, resulting in a standardized spectral feature vector. This vector is composed of the standardized spectral feature data, which includes a set of key spectral information capable of quantifying the chemical composition of plant protein raw materials, such as characteristic peak position coordinates, standardized absorption intensity, slope variation information, and feature vector dimension. Based on the standardized spectral feature vector, a multidimensional feature space model is constructed to obtain a spectral fingerprint database of plant protein raw materials. This technical solution can obtain accurate standardized spectral feature data, laying the foundation for further analysis of the composition of plant protein meat raw materials during production.

[0058] Furthermore, obtaining the set of low-dimensional feature vectors includes:

[0059] Obtain a standardized spectral feature data matrix and calculate the covariance relationship between each feature dimension; based on the covariance matrix, use principal component analysis algorithm to calculate eigenvalues ​​and eigenvectors, sort them in descending order of eigenvalue size, and determine the contribution distribution of principal components; use the cumulative variance contribution rate calculation formula to accumulate the variance contribution values ​​of each principal component to determine the number of principal components to retain.

[0060] The importance of the retained principal components is evaluated using feature weight allocation coefficients. The explanatory power of each principal component in the feature space is calculated. Based on the principal component importance ranking results, a set of low-dimensional feature vectors representing different plant protein raw material types is obtained.

[0061] Specifically, the vector composed of the standardized spectral feature data is used as the standardized spectral feature data matrix. The covariance relationship between each feature dimension is calculated, and a covariance matrix is ​​constructed as the input basis for principal component analysis. Based on the covariance matrix, principal component analysis algorithms are used to calculate eigenvalues ​​and eigenvectors. The eigenvalues ​​are then sorted in descending order to determine the contribution distribution of the principal components. The variance contribution rate is calculated by summing the variance contribution values ​​of each principal component. If the cumulative contribution rate reaches a preset threshold, the number of retained principal components is determined. The importance of the retained principal components is evaluated using feature weight allocation coefficients, and the explanatory power of each principal component in the original feature space is calculated to obtain the principal component importance ranking result. Based on the importance ranking result, a principal component transformation matrix is ​​constructed, and a linear transformation operation is performed on the original spectral feature data to obtain the dimensionality-reduced principal component score data. Cluster analysis is used to group the principal component score data, identifying feature patterns of different plant protein raw material types and establishing a mapping relationship between type labels and principal component features. Based on the mapping relationship, a low-dimensional feature vector set containing plant protein raw material type information is generated, forming a feature database for subsequent classification and identification.

[0062] For example, the system automatically starts the principal component analysis module, taking the standardized spectral feature data matrix as input, and first calculates the covariance matrix. Where X is the centered spectral data matrix, and n is the number of samples. The system calls the eigenvalue decomposition algorithm to decompose the covariance matrix, obtaining M eigenvalues ​​and corresponding eigenvectors. The variance contribution rate of each principal component is calculated, and the system automatically selects the top k principal components based on the cumulative variance contribution rate reaching a preset threshold. Then, a feature weight allocation algorithm is applied to calculate the importance of each principal component, using the information entropy weighting method. ,in Let m be the normalized loading value of the i-th principal component in the j-th sample, and m be the total number of samples. The weight coefficients of the first K principal components are then derived. The system transforms the original multidimensional spectral data using a transformation matrix. Projecting onto a K-dimensional principal component space, where W is a transformation matrix composed of the first k eigenvectors, ultimately generates a set of low-dimensional eigenvectors with dimension n x k. This technical solution improves the data compression ratio while retaining key spectral feature information of different plant protein raw material types, such as soybean protein, pea protein, and wheat protein, thereby improving the accuracy of plant protein meat raw material identification and laying the foundation for improving the production quality of plant protein meat.

[0063] Furthermore, the acquisition of classification result data containing raw material type information includes: acquiring low-dimensional feature vector set data, calculating Euclidean distance measurement parameters between each raw material type, constructing a distance matrix to record the spatial distribution relationship between different types; using the K-means clustering algorithm to optimize and adjust the coordinate points of the cluster centers, calculating the optimal center position for each raw material type, and generating a stable type coding label mapping relationship; establishing a database corresponding to raw material types and coding identifiers through the type coding label mapping relationship, recording the feature vector range and classification boundary information of each type; for the feature vector data to be detected, calculating its Euclidean distance value with the coordinate points of each cluster center, and using the Euclidean distance value as the measurement parameter; based on the measurement parameter and the raw material type database, generating a classification result data set containing complete raw material type information.

[0064] Specifically, by using a set of low-dimensional feature vectors to process the spectral feature data of the raw materials, in the above steps, the spectral data of the plant protein raw materials have been collected by near-infrared spectroscopy, and standardized spectral feature data have been obtained through environmental correction and feature extraction steps. Next, principal component analysis (PCA) is used to compress the vector dimensions of these spectral data, thereby extracting a set of low-dimensional feature vectors. PCA not only reduces the dimensionality of the data but also retains crucial information for raw material classification. Each low-dimensional feature vector represents a characteristic of a plant protein raw material, and this vector set provides the foundational data for subsequent classification and analysis. Using these low-dimensional feature vectors, the Euclidean distance between different plant protein raw material types is calculated. Euclidean distance is an important indicator for measuring the differences between different categories of raw materials. By constructing a distance matrix, the spatial distribution relationship between different plant protein raw materials can be identified. If the distance between different types exceeds a set threshold, the system further optimizes the cluster centers using the K-means clustering algorithm to find the optimal center position for each plant protein type. Each cluster center corresponds to a plant protein raw material type, and these types are identified using stable type coding labels in the raw material type database. Based on the coordinate information of the cluster centers, the system assigns a unique coding label to each plant protein raw material type. Through these labels, the system can associate different spectral data with raw material types, forming a classification result dataset containing raw material type information. This dataset not only provides clear raw material type identification for subsequent production processes but also offers real-time classification information for the production control system. In practical applications, this technological feature effectively addresses the challenges posed by the diversity and complexity of plant protein raw material components. Through precise spectral data classification, the system can monitor raw material quality in real time and provide reliable raw material type information for subsequent processing, avoiding errors inherent in traditional manual judgment and chemical analysis. Furthermore, the combination of cluster analysis and Euclidean distance helps identify subtle differences between different types of raw materials, ensuring accurate classification and proportioning of raw materials. For example, if there is a change in the type of raw material or the addition of a new batch of raw materials during the production process, the system can automatically classify and readjust the proportions based on low-dimensional feature vectors and Euclidean distance measurements, ensuring the stability and consistency of product quality. The production line can not only process complex plant protein raw materials more efficiently, but also reduce the impact of human factors on production, thereby improving the automation level of production and the overall quality of products. The above technical solution, through steps such as low-dimensional feature vector set, Euclidean distance calculation, and cluster analysis, forms a complete automated monitoring and management solution. Through accurate classification and real-time feedback of plant protein raw materials, it significantly improves the efficiency of the production process and the accuracy of raw material proportioning, thus effectively solving the problem of the inability to accurately handle the mixing of multiple raw materials in existing technologies.

[0065] Furthermore, after obtaining the classification result data containing raw material type information, such as... Figure 3 As shown, it also includes:

[0066] Based on the classification results, the component concentration gradient distribution of each detection point is analyzed. A quantitative relationship model between spectral characteristics and protein content is established using partial least squares regression. Component content prediction values ​​at different depths are obtained through detection depth layering, yielding component concentration data for each detection point. A sampling time interval sequence is constructed using the component concentration data from each detection point. The influence of time drift is corrected using a signal intensity attenuation curve. The component concentration gradient distribution is interpolated based on spatial coordinate positioning information to obtain a spatial distribution map reflecting the overall component state of the raw materials. The deviation between the spatial distribution map and the preset component ratio standard is calculated. A weighted average algorithm is used to adjust the raw material mixing ratio according to the characteristic weight allocation coefficient of each detection point, determining the final raw material ratio scheme and outputting it to the production control system.

[0067] Specifically, the spectral characteristics of each detection point were obtained by combining the spectral data of plant protein raw materials from the classification results with spatial coordinate information. Based on this spectral data, the system established a precise quantitative relationship model between spectral absorption intensity values ​​and protein content using a partial least squares regression algorithm. The introduction of the partial least squares regression algorithm allows the features extracted from complex spectral data to be accurately mapped to protein content. Especially in the case of multiple component interactions, this algorithm can effectively reduce the impact of multicollinearity on the analysis results, ensuring the robustness and accuracy of the quantitative model. Using this model, the system can accurately predict the protein content of each detection point, providing a basis for further adjustment of raw material ratios. To address the issue of component concentration differences at different depth levels, the system also performs depth level division for each detection point. This division is based on the three-dimensional spatial coordinates of the sample. By assigning each point to a specific depth level, such as surface, middle, and deep, the system can establish protein content prediction models for different depth levels. These hierarchical models not only help identify changes in protein content in surface and deep regions, but also provide a high-resolution concentration prediction map that clearly reflects the compositional distribution of the entire sample, solving the problem that traditional methods cannot accurately reflect spatial variations in composition.

[0068] Furthermore, the accurate predictions of the partial least squares regression model can effectively handle differences between different samples, especially differences in component concentrations at different depths. Traditional analytical methods often fail to capture these subtle spatial variations, while this technical solution, through stratified analysis, accurately presents the gradient changes in component concentration. This method allows for personalized concentration predictions at each detection point in the production process, ensuring the accuracy of raw material mixing ratios and avoiding product quality fluctuations caused by uneven component distribution. Moreover, by interpolating the concentration data at different depths, the system can fuse the predicted component concentration values ​​from different depths to obtain a complete, fine-grained component concentration distribution map. This map not only reflects the component status at each detection point but also reveals the overall component variation trend of the sample, providing a precise basis for raw material formulation during the production process. For example, within a production batch, the system can monitor and adjust the protein content at each depth level in real time, ensuring that the component ratio at each level meets predetermined standards.

[0069] By constructing a sampling time interval sequence, the system can also monitor and correct data errors caused by time drift. During this process, the system calculates the signal attenuation curve of concentration changes and uses an appropriate attenuation correction algorithm to adjust the data, ensuring the stability and reliability of component concentration data during long-term sampling. This solves the concentration drift problem caused by time changes in conventional monitoring systems during continuous production, thereby improving data accuracy and effectiveness. Through the cooperation of the above steps, several technical challenges in plant-based meat processing are jointly solved. Through precise component concentration analysis and a real-time adjustment mechanism, the system can achieve precise control of raw materials during the production of high-wet-process plant-based meat. This not only improves production efficiency but also ensures the consistency and stability of the nutritional components and taste characteristics of the final product. For example, in a certain production stage, the system, through real-time analysis of protein content prediction data, discovered a deviation in protein concentration between the surface and deeper layers. The above technical solution solves the technical problem of uneven component distribution of plant protein raw materials at different depths and time periods, optimizes raw material management during production, and improves product quality control capabilities, making the production process more refined and efficient.

[0070] Furthermore, the preset wavelength range is 1000-2500nm.

[0071] This invention also provides an automated monitoring and management system for high-moisture processing of plant-based meat, used to implement the above-mentioned methods, such as... Figure 4 As shown, the system includes:

[0072] The spectral data acquisition module is used to acquire multi-point spectral data of plant protein raw materials using a near-infrared spectrometer. It establishes a detection grid on the surface of the raw materials based on spatial coordinate positioning information, and obtains the spectral band segmentation marks and absorption intensity value range of each detection point by scanning through a preset wavelength range, thus obtaining a spectral data matrix containing spatial location information.

[0073] The environmental correction module is used to perform optical path transmission loss coefficient correction processing based on the spectral data matrix, eliminate environmental interference by using temperature compensation correction factor and humidity influence correction parameter, and extract the position coordinates of characteristic peaks through first derivative transformation to obtain standardized spectral feature data after environmental correction.

[0074] The feature extraction module is used to control the vector dimension compression ratio of the standardized spectral feature data through principal component analysis, extract principal components whose cumulative variance contribution rate reaches a preset value, calculate the importance ranking of each principal component using feature weight allocation coefficients, and obtain a set of low-dimensional feature vectors representing different plant protein raw material types.

[0075] The classification and recognition module is used to calculate the inter-class distance measurement parameters based on the set of low-dimensional feature vectors, determine the raw material type classification of each detection point through Euclidean distance calculation, and obtain classification result data containing raw material type information.

[0076] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0077] In summary, the core of the technical solution of this invention lies in achieving accurate identification, quantitative analysis, and automatic proportioning adjustment of plant protein raw material components by combining near-infrared spectroscopy with multi-point data acquisition, environmental interference correction, principal component analysis, and partial least squares regression. First, in step S1, multi-point spectral data of the plant protein raw material surface is acquired using a near-infrared spectrometer, and a detection grid is established based on spatial coordinates to obtain a spectral data matrix containing location information. Next, in step S2, environmental interference is eliminated through environmental correction methods such as optical path loss correction, temperature and humidity compensation, and first-order derivative transformation to obtain standardized spectral feature data. Step S3 uses principal component analysis to compress the dimensionality of the standardized data, extracting a set of low-dimensional feature vectors to facilitate subsequent raw material classification and identification. Step S4 uses Euclidean distance calculation and K-means clustering algorithm to achieve accurate classification of plant protein raw material types. In step S5, a quantitative relationship between spectral features and component concentration is established using partial least squares regression algorithm to predict the component concentration at different detection points. Furthermore, by constructing a component concentration gradient distribution map, the raw material ratio scheme is adjusted in real time to ensure product consistency and quality. The coordinated operation of each step significantly improves data accuracy and the system's real-time response capability. The environmental calibration and data acquisition process from S1 to S2 provides a reliable data foundation for subsequent accurate classification and quantitative analysis; the principal component analysis and classification identification in S3 and S4 ensures the accurate differentiation of different types of plant proteins, providing a clear basis for raw material ratio; S5 and the concentration prediction and adjustment mechanism ensure the precise ratio and quality control of raw materials during the production process, ultimately forming a highly efficient, automated, and accurate plant protein meat processing monitoring and management system.

[0078] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An automated monitoring and management method for high-moisture processing of plant-based meat, characterized in that, The application relates to a plant protein raw material classification method based on near-infrared spectroscopy. The method comprises the following steps: collecting multi-point spectral data of a plant protein raw material, establishing a detection grid on the surface of the raw material according to spatial coordinate positioning information, obtaining spectral band segmentation marks and absorption intensity value ranges of each detection point by scanning a preset wavelength range, and obtaining a spectral data matrix containing spatial position information; According to the spectral data matrix, a light path transmission loss coefficient correction process is performed, temperature compensation correction factors and humidity influence correction parameters are used to eliminate environmental interference, and characteristic peak position coordinates are extracted through first-order derivative transformation to obtain standardized spectral feature data after environmental correction; The standardized spectral feature data is subjected to vector dimension compression ratio control through a principal component analysis method, main components with a cumulative variance contribution rate reaching a preset value are extracted, the importance of each main component is calculated by using a feature weight distribution coefficient, and a low-dimensional feature vector set representing different types of plant protein raw materials is obtained; According to the low-dimensional feature vector set, an inter-class distance measurement parameter is calculated, and the type classification of each detection point is determined through Euclidean distance calculation, and a classification result data containing raw material type information is obtained; After obtaining the classification result data containing the raw material type information, the method further comprises the following steps: A sampling time interval sequence is constructed by using the component concentration data of each detection point, a signal strength attenuation curve is used to correct the influence of time drift, and the component concentration gradient distribution is calculated by interpolation according to the spatial coordinate positioning information, so as to obtain a spatial distribution map reflecting the overall component state of the raw material; According to the spatial distribution map and a preset component distribution standard, a deviation is calculated, a weighted average algorithm is used to adjust the raw material mixing ratio according to the feature weight distribution coefficient of each detection point, a final raw material mixing ratio scheme is determined, and the final raw material mixing ratio scheme is output to a production control system.

2. The method of claim 1, wherein, The acquisition of the spectral data matrix comprises the following steps: According to the geometric size and surface characteristics of the plant protein raw material, a grid division algorithm is used to establish a uniformly distributed detection grid on the surface of the raw material, spatial coordinate information of each grid node is obtained, spectral scanning is performed on the detection points corresponding to the spatial coordinates by using a near-infrared spectrometer, spectral response signals in a preset wavelength range are collected, and a spectral data matrix is constructed based on the coordinate information of each detection point, the identification of each characteristic band and the spectral data corresponding to each characteristic band.

3. The method of claim 1, wherein, The acquisition of the standardized spectral feature data comprises the following steps: According to the intensity values of the wavelength points in the spectral data matrix, a light path loss correction model is used to calculate a transmission attenuation coefficient, and a light path loss correction parameter table is obtained; A temperature compensation factor is used to linearly correct the spectral intensity based on the correction parameter table, and temperature-corrected spectral response data is obtained; According to the temperature-corrected spectral response data, the amplitude of each band absorption peak is adjusted through a humidity correction parameter, and a purified spectral signal after environmental interference elimination is obtained; The purified spectral signal is subjected to differential processing by using a first-order derivative transformation algorithm, the slope change information of the spectral curve is extracted, the wavelength position of the characteristic peak value is determined, and the standardized spectral feature data is obtained.

4. The method of claim 1, wherein, The acquisition of the low-dimensional feature vector set comprises the following steps: The standardized spectral feature data matrix is obtained, and the covariance relationship between each feature dimension is calculated; according to the covariance matrix, the eigenvalue and eigenvector are calculated by using the principal component analysis algorithm, the eigenvalues are arranged in descending order according to the size, and the contribution degree distribution of the principal component is determined; the variance contribution value of each principal component is added one by one through the cumulative variance contribution rate calculation formula, and the number of retained principal components is determined; the importance of the retained principal components is evaluated by using the feature weight distribution coefficient, the explanation ability of each principal component in the feature space is calculated, and the low-dimensional feature vector set representing different plant protein raw material types is obtained based on the principal component importance ranking result.

5. The method of claim 1, wherein, The acquisition of the classification result data containing the raw material type information includes: The low-dimensional feature vector set data is obtained, the Euclidean distance measurement parameter between each raw material type is calculated, the distance matrix recording the spatial distribution relationship between different types is constructed, the K-means clustering algorithm is used to optimize and adjust the cluster center coordinate points, the optimal center position of each raw material type is calculated, and a stable type coding label mapping relationship is generated; through the type coding label mapping relationship, a raw material type and coding identification corresponding database is established, and the feature vector range and classification boundary information of each type are recorded; for the feature vector data to be detected, the Euclidean distance value between the feature vector data and each cluster center coordinate point is calculated, and the Euclidean distance value is used as the measurement parameter; based on the measurement parameter and the raw material type database, a classification result data set containing complete raw material type information is generated.

6. The method of claim 1, wherein, After obtaining the classification result data containing the raw material type information, the following steps are further included: According to the classification result data, the component concentration gradient distribution of each detection point is analyzed, a quantitative relationship model between the spectral feature and the protein content is established by using the partial least squares regression algorithm, the component content prediction value at different depths is obtained through detection depth level division, and the component concentration data of each detection point is obtained.

7. The method of claim 1, wherein, The preset wavelength range is 1000-2500nm.

8. An automated monitoring and management system for high-moisture plant protein meat processing for implementing the method according to any one of claims 1 to 7, characterized in that The system includes: A spectrum data acquisition module is configured to acquire multi-point spectrum data of plant protein raw materials by using a near-infrared spectrometer, establish a detection grid on the surface of the raw materials according to spatial coordinate positioning information, obtain spectrum band segmentation marks and absorption intensity numerical ranges of each detection point by scanning through a preset wavelength range, and obtain a spectrum data matrix containing spatial position information; An environment correction module is configured to correct a light path transmission loss coefficient based on the spectrum data matrix, eliminate environmental interference by using a temperature compensation correction factor and a humidity influence correction parameter, and extract feature peak position coordinates by first derivative transformation to obtain standardized spectral feature data after environment correction; A feature extraction module is configured to control the vector dimension compression ratio of the standardized spectral feature data by using a principal component analysis method, extract principal components with a cumulative variance contribution rate reaching a preset value, calculate the importance ranking of each principal component by using a feature weight distribution coefficient, and obtain a low-dimensional feature vector set representing different plant protein raw material types. The classification recognition module is configured to calculate an inter-class distance measurement parameter according to the low-dimensional feature vector set, determine a raw material type classification to which each detection point belongs through Euclidean distance calculation, and obtain classification result data containing raw material type information.

9. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions, when executed by a processor, implement the method of any one of claims 1-7.

Images