Honey refining process characterization method and system based on near infrared spectrum and color characteristics

Abstract

The invention belongs to the technical field of traditional Chinese medicine honey refining process detection and analysis. The invention provides a honey refining process characterization method and system based on near infrared spectrum and color characteristics. The method comprises the following steps: acquiring near infrared spectrum data and corresponding color images of a sample in a flow cell through an outer circulation pipeline; extracting low-dimensional color features from the image, coupling and fusing the low-dimensional color features with the preprocessed spectral data, and constructing a honey refining process state characterization vector; establishing a partial least squares regression model as a main model, and introducing an extreme gradient lifting model as nonlinear supplement to obtain a comprehensive characterization result; and performing variable contribution analysis on the double models by utilizing a hierarchical interpretation framework, and verifying the consistency of a model prediction mechanism and a physical and chemical reaction path in a honey refining process. According to the method, spectrum and visual information are fused, linear and nonlinear modeling advantages are combined, reliability of a model mechanism is ensured through interpretability analysis, and a new way is provided for accurate monitoring of the honey refining process.

Description

Technical Field

[0001] This invention relates to the field of detection and analysis technology of honey refining process of traditional Chinese medicine, specifically to a method and system for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Honey refining is a crucial step in the processing of traditional Chinese medicine, and its quality directly determines the efficacy of the finished product. Traditional monitoring relies heavily on manual observation of macroscopic indicators such as color and viscosity, which is highly subjective and time-consuming. In recent years, near-infrared spectroscopy, due to its rapid and non-destructive characteristics, has been introduced into process analysis, reflecting changes in internal chemical composition such as sugar and moisture. Simultaneously, machine vision technology can quantify the color characteristics of materials and capture apparent color evolution caused by Maillard reactions. Combining spectral and image data can theoretically construct a more comprehensive digital profile of the honey refining process, providing a data foundation for intelligent control.

[0004] However, existing technologies, when integrating spectral and color features, often focus solely on improving prediction accuracy, lacking interpretability analysis of the model's decision-making mechanism. This results in a "black box" for the constructed comprehensive characterization model, failing to elucidate how spectral wavelengths or color variables specifically drive the prediction of quality attributes, and making it even more difficult to verify whether the model's prediction logic aligns with the actual physicochemical reaction pathways in the honey refining process. This opacity prevents technicians from confirming whether the model truly captures key process characteristics, thus hindering reliable process control and quality traceability based on model results. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for characterizing the honey refining process based on near-infrared spectroscopy and color features. This addresses the problems in existing technologies, such as incomplete characterization of the honey refining process, insufficient correspondence between detection results and process states, excessively high visual modality dimensions, and lack of mechanistic explanation in the models.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics.

[0007] A method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics includes the following steps: Near-infrared spectral data and corresponding color images of the samples in the flow cell were obtained. The samples in the flow cell were samples input through the external circulation pipeline during the honey refining process. Low-dimensional color features are extracted from color images, and preprocessed near-infrared spectral data are coupled and fused with low-dimensional color features to construct a state characterization vector for the honey refining process. Based on the state representation vector of the honey refining process, a partial least squares regression model is established as the main representation model, and an extreme gradient boosting model is introduced as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. Based on the comprehensive characterization results of the honey refining process, a hierarchical interpretation framework was used to conduct variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model to determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

[0008] In one implementation of the first aspect of the present invention, extracting low-dimensional color features from a color image includes: Perform center cropping on the color image and convert the image to the CIE Lab color space; Calculate in CIE Lab color space , and The average value is used to obtain low-dimensional color features.

[0009] In one implementation of the first aspect of the present invention, constructing a honey refining process state representation vector includes: The internal composition state variables formed by the preprocessed near-infrared spectral data and the appearance thermal reaction state variables formed by the low-dimensional color features are standardized and integrated into a unified feature matrix, which serves as the state characterization vector for the honey refining process.

[0010] In one implementation of the first aspect of the present invention, a partial least squares regression model is established as the main representation model, including: The state characterization vector of the honey refining process is input into the partial least squares regression model. By extracting the latent variable score matrix and loading matrix, the multimodal input is mapped to a unified latent variable space, and a linear comprehensive characterization relationship is established between the internal composition state variables, the external thermal reaction state variables and the key quality attributes. In one implementation of the first aspect of the present invention, an extreme gradient boosting model is introduced as a nonlinear supplementary analysis model, including: An extreme gradient boosting model is trained using the state representation vector of the honey refining process. The model complexity is controlled by minimizing the loss function and introducing a regularization term to supplement the analysis of the nonlinear expressive power of key quality attributes.

[0011] In one implementation of the first aspect of the present invention, a hierarchical interpretation framework is used to perform variable contribution analysis on the principal representation model, including: Calculate the variable projection importance value of each variable in the main representation model, and identify the key wavelength variables and key color variables that have a significant impact on the prediction of key quality attributes within the overall sample range; The calculation process for the projected importance value of variables is as follows: the sum of squares of the explanations of each latent variable component to the response variable is calculated, and the weights of each variable in each component are combined to obtain the weighted sum of the contributions of each variable to all latent variable components, thereby quantifying the importance of each variable in constructing a unified latent variable space.

[0012] In one implementation of the first aspect of the present invention, a hierarchical interpretation framework is used to perform variable contribution analysis on a nonlinear supplementary analysis model, including: The SHAP method is used to quantify the contribution of each feature in the nonlinear supplementary analysis model to a single prediction result, and a SHAP summary plot is generated. Based on the generated SHAP summary diagram, analyze the direction and intensity of the contribution of key wavelength variables and key color variables to the prediction results of key quality attributes. The calculation logic for the contribution value is as follows: based on the expected output of the model, calculate the marginal contribution of the model output when each feature exists or not, and obtain the independent contribution of each feature to the final predicted value by arranging and combining the average marginal contribution of all feature subsets.

[0013] In one implementation of the first aspect of the present invention, the key quality attributes include glucose content, fructose content, moisture content, pH value, and 5-hydroxymethylfurfural content. Partial least squares regression model was used to characterize the linear trends of glucose content, fructose content, moisture content, pH value and 5-hydroxymethylfurfural content; The extreme gradient boosting model is used to detect whether there are deviations in glucose content, fructose content, moisture content, pH value, and 5-hydroxymethylfurfural content caused by nonlinear coupling between internal composition state variables and external thermal reaction state variables.

[0014] As a further limitation of the first aspect of the present invention, determining the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process includes: Based on the variable projection importance values ​​and SHAP summary plot, the set of key variables driving the model prediction is determined; Analyze whether the contribution direction of the key color variable in the key variable set is consistent with the trend of increased browning caused by sugar dehydration and Maillard reaction during honey refining; If the key color variable representing a decrease in brightness contributes positively to the prediction of 5-hydroxymethylfurfural content, and the key color variable representing an increase in redness contributes positively to the prediction of 5-hydroxymethylfurfural content, then the model prediction mechanism is considered to be consistent with the physicochemical reaction pathway.

[0015] In one implementation of the first aspect of the present invention, the hierarchical interpretation framework further includes single-sample decision path analysis, which includes: By selecting representative samples and using single samples, we aim to demonstrate how the prediction results of representative samples are jointly driven by key wavelength variables and key color variables. By comparing the variable contribution distributions of the single-spectral model and the multimodal fusion model on representative samples, we can determine whether low-dimensional color features are used to supplement missing information in near-infrared spectral data or exist as redundant variables.

[0016] In one implementation of the first aspect of the present invention, the output of the comprehensive characterization result includes: Quantitative predictions of key quality attributes based on the output of the main characterization model and the nonlinear supplementary analysis model; Based on the hierarchical interpretation framework, the report on the contribution analysis of key variables includes the global importance ranking and local contribution direction of key wavelength and key color variables to key quality attributes. The process status credibility evaluation generated based on the consistency judgment results is used to guide the status judgment, process trend analysis and quality control assistance of the honey refining process.

[0017] Secondly, the present invention provides a honey refining process characterization system based on near-infrared spectroscopy and color characteristics.

[0018] A honey refining process characterization system based on near-infrared spectroscopy and color characteristics, comprising: The spectral image acquisition unit is configured to acquire near-infrared spectral data of the sample in the flow cell and the color image corresponding to the sample, wherein the sample in the flow cell is the sample input through the external circulation pipeline during the honey refining process; The feature fusion construction unit is configured to: extract low-dimensional color features from the color image, and couple and fuse the preprocessed near-infrared spectral data with the low-dimensional color features to construct a honey refining process state representation vector; The dual-mode modeling and representation unit is configured to: establish a partial least squares regression model as the main representation model based on the state representation vector of the honey refining process, and introduce an extreme gradient boosting model as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. The mechanism consistency criterion unit is configured to: based on the comprehensive characterization results of the honey refining process, use a hierarchical interpretation framework to perform variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model, and determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

[0019] Thirdly, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the honey refining process characterization method based on near-infrared spectral and color characteristics of the first aspect of the present invention.

[0020] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively addresses the technical problems of existing honey refining process characterization models lacking interpretability and struggling to verify whether the prediction logic conforms to the actual reaction path by constructing a coupling and fusion mechanism of near-infrared spectroscopy and color features, and employing a dual-mode modeling strategy of partial least squares regression and extreme gradient boosting. Based on the comprehensive characterization results of the honey refining process, a hierarchical interpretation framework is used to analyze the variable contributions of the main characterization model and the nonlinear supplementary analysis model, quantitatively revealing the specific driving role of key spectral wavelengths and color features in the prediction of quality attributes. This mechanism not only breaks the "black box" state of traditional models but also achieves consistency verification between the model prediction mechanism and the actual reaction path by comparing the direction of variable contributions with the physicochemical changes such as sugar dehydration and Maillard reaction during the honey refining process. Therefore, this invention ensures that the comprehensive characterization results not only have high precision but also clear physicochemical significance, enabling technicians to conduct precise process control and quality traceability based on reliable model logic, significantly improving the scientific rigor and reliability of honey refining process control.

[0021] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 A flowchart illustrating a honey refining process characterization method based on near-infrared spectroscopy and color features, provided as an exemplary embodiment of the present invention; Figure 2 A spectral color fusion feature VIP curve diagram provided as an exemplary embodiment of the present invention; Figure 3 A summary diagram of spectral color fusion features (SHAP) provided as an exemplary embodiment of the present invention; Figure 4 A representative sample SHAP force map provided for an exemplary embodiment of the present invention; Figure 5A schematic diagram of a honey refining process characterization system based on near-infrared spectroscopy and color characteristics, provided as an exemplary embodiment of the present invention; Figure 6 A solution for a computer device provided as an exemplary embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] This implementation proposes a characterization method for the honey refining process based on near-infrared spectroscopy and color characteristics. The following is a brief introduction to the technical terms and related concepts involved in this processing scheme: "Honey refining" refers to the process of concentrating and refining honey under vacuum and heating conditions to achieve specific viscosity, water content, and color to meet the production requirements of traditional Chinese medicine honey pills and other preparations.

[0027] "Near-infrared spectroscopy" refers to spectral data with wavelengths or wavenumbers in the near-infrared range that can characterize the absorption information of groups such as OH, CH, and CO in a sample.

[0028] "Lab color features" refers to the CIE Lab color space parameters extracted from the sample image, where... Indicates brightness. Indicates the red-green axis. This indicates the yellow-blue axis.

[0029] "Multimodal fusion" refers to the unified modeling of data from different physical sources. In this invention, it specifically refers to the feature-level fusion of near-infrared spectral data and image color feature data.

[0030] "5-HMF" stands for 5-hydroxymethylfurfural, which is an important quality indicator reflecting the degree of thermal processing reaction in honey.

[0031] "Characteristics" refers to the unified description of the changes in internal composition, macroscopic appearance evolution, and related quality status of samples during the honey refining process by integrating near-infrared spectral information and color feature information.

[0032] Existing offline analytical methods, such as high-performance liquid chromatography (HPLC), can provide highly accurate quantitative results, but they suffer from long detection cycles, feedback lag, and difficulty in being applied to high-frequency industrial processes, making it difficult to meet the practical needs of continuous monitoring and process control in honey refining. Near-infrared spectroscopy has been widely used for rapid component analysis and process monitoring, and can effectively reflect changes in the internal composition and molecular structure of samples. However, in high-viscosity systems like honey refining, with heterogeneous optical properties, spectral measurements are easily affected by scattering variations and baseline drift. Furthermore, the honey refining process involves not only molecular-level component changes but also significant macroscopic changes in appearance, especially the gradually deepening browning caused by cumulative thermal reactions. This information about cumulative thermal reactions cannot be fully represented by a single spectral feature; therefore, the characterization of the honey refining process using a single near-infrared spectroscopy method remains limited. On the other hand, image-based color descriptors can provide complementary information related to the thermal reaction process. The intensity of browning during honey refining directly reflects the progress of the thermal reaction, and 5-HMF is an important quality marker of the degree of thermal reaction. Therefore, fusing the molecular layer information represented by spectroscopy with the macroscopic appearance changes represented by color is beneficial to enhancing the characterization ability of the honey refining process. However, existing multimodal studies in the food field often rely on high-dimensional visual features or end-to-end deep learning models. Such methods typically require large-scale datasets and strict data collection standardization conditions. In real industrial environments, honey refining process data usually exhibits a clear batch structure, a limited number of samples, and is accompanied by process disturbances and collection fluctuations. Under these conditions, directly using high-dimensional visual features or complex deep learning models can easily lead to problems such as high model complexity, increased risk of overfitting, and insufficient interpretability, which is not conducive to stable application in industrial scenarios. Furthermore, even if online sampling or process monitoring devices are set up in existing technologies, if the analysis model still mainly relies on single-modal information or uses a high-dimensional black box model that is difficult to interpret, it is still difficult to simultaneously ensure the comprehensive characterization ability of the honey refining process, model robustness, and engineering application feasibility. For complex thermal processing systems like honey refining where internal composition changes and appearance evolution occur simultaneously, there is a lack of a multimodal comprehensive characterization scheme based on low-dimensional color features rather than high-dimensional visual features, capable of synergistic fusion with near-infrared spectroscopy, and possessing clear physical interpretation capabilities. While retaining the ability of near-infrared spectroscopy to characterize changes in the sample's internal composition, this study introduces color features that reflect the appearance evolution of the thermal reaction and uses low-dimensional, physically meaningful features to construct a fusion model. This allows for a more robust characterization of the honey refining process state and key quality attributes, even under conditions of limited sample size, significant batch structure, and process disturbances, providing a foundation for further process monitoring and quality assessment.

[0033] This invention proposes a technical solution to address the following problems in the analysis of the honey refining process: existing offline detection methods have long feedback cycles, making it difficult to meet the requirements of continuous analysis and timely control of the honey refining process; single near-infrared spectroscopy methods mainly reflect changes in the internal composition of the sample, making it difficult to fully characterize the comprehensive color evolution caused by the accumulation of thermal reactions during the honey refining process; although single color information can reflect changes in appearance, it is difficult to independently reflect changes in the internal composition of the sample; existing multimodal methods based on high-dimensional visual features or end-to-end deep learning usually rely on large-scale datasets and strictly standardized acquisition conditions, which are prone to problems such as high model complexity, high risk of overfitting, and insufficient physical interpretability in batch, small-sample scenarios with process disturbances, such as industrial honey refining.

[0034] In view of the problems existing in the existing solutions, the present invention provides a method for characterizing the honey refining process based on the interpretable fusion of near-infrared spectroscopy and color features, such as... Figure 1 As shown, the input on the left includes spectral signals and an array of honey sample images; multidimensional spectral features are extracted through "spectral preprocessing". The Lab mean features of the CIELAB space were obtained through "low-dimensional color feature extraction". The two are concatenated into a joint feature vector in the "Feature Fusion" module. ; The input is a "baseline model," which uses PLS modeling and XGBoost modeling in parallel for prediction. The model output is connected to the "model interpretability module," which achieves global and local interpretability through variable importance analysis, SHAP plot of important variables, and SHAP plot of representative samples. Finally, the output is a "comprehensive characterization result of the honey refining process," forming a closed-loop analysis process from multi-source perception → feature fusion → intelligent modeling → mechanism explanation.

[0035] This invention uses near-infrared spectroscopy as a characterization method for changes in internal composition and color features as a characterization method for appearance evolution during thermal processing. Instead of using high-dimensional raw images or complex end-to-end visual depth models, it employs low-dimensional color features with clear physical meaning, preferably Lab color features, to construct a low-dimensional, interpretable multimodal fusion framework. Based on this, a linear comprehensive characterization model (PLS model) and a nonlinear comprehensive characterization model (XGBoost model) are established to characterize key quality attributes such as glucose (GLU), fructose (FRU), moisture content (H2O), pH, and 5-HMF during the honey refining process. The XGBoost model supplements the quality attributes that rely on the expressive power of the nonlinear model. A hierarchical interpretation framework is then constructed, including VIP curves, SHAP total plots, and the SHAP single-sample force graph interpretation method. The key variables driving the model predictions and their contribution directions to the prediction results are systematically analyzed. The physical rationality behind the improved performance of the multimodal model is verified, and the consistency between the model prediction mechanism and the physicochemical reaction pathways in the honey refining process is determined, further enhancing the reliability and interpretability of the proposed modeling strategy.

[0036] This invention achieves sampling during the honey refining process through an external circulation pipeline, allowing the sample to enter a flow cell and complete near-infrared spectral acquisition within the cell. Color images corresponding to the sample are acquired, and low-dimensional Lab color features are extracted. The internal composition state variables formed by the near-infrared spectrum are coupled and fused with the appearance thermal reaction state variables formed by the low-dimensional color features to constitute a state representation vector for the honey refining process. Based on this state representation vector, a partial least squares regression (PLS) model is established, and XGBoost is introduced as a nonlinear supplementary analysis model. A hierarchical interpretation framework for both the PLS and XGBoost models is employed to analyze key variables and their mechanisms of action, thereby obtaining a comprehensive representation of the honey refining process.

[0037] More specifically, it includes the following processes: Step S1: Sample acquisition during the honey refining process.

[0038] During the honey refining process, the honey raw materials are placed in a refining tank for heat processing. An external circulation pipeline is installed outside the refining tank. The honey material is drawn out of the refining tank through the external circulation pipeline and enters the external detection loop. After passing through the flow cell, it is sampled to achieve process sampling. External circulation sampling allows the sample to enter the detection system without significantly interfering with the main process, while retaining a sampling port for offline reference analysis and model maintenance.

[0039] Step S2: Near-infrared spectral acquisition in the flow cell.

[0040] In step S1, the externally circulated sample enters the flow cell, where near-infrared spectral acquisition is completed. The near-infrared detection unit consists of a near-infrared spectrometer, a flow cell, and an optical fiber coupling structure, allowing spectral detection to be performed under relatively stable flow conditions. The spectral acquisition position is fixed within the flow cell, rather than being randomly measured directly in the honey-refining tank, making the near-infrared spectral detection conditions more stable and thus improving the consistency of the spectrum's characterization of internal compositional changes.

[0041] Step S3: Honey color image acquisition.

[0042] Color images are acquired for the samples corresponding to step S2. These images are acquired under controlled lighting and fixed shooting parameters to minimize the impact of ambient light, background, and shooting geometry on color characteristics. The color images characterize the overall color evolution of the samples during the honey refining process, including changes in brightness, overall color difference, and browning degree. Based on the principle of time synchronization, spectral samples and image samples from the honey refining process are paired to form a spectral visual inspection dataset.

[0043] Step S4: Near-infrared spectroscopy preprocessing.

[0044] To reduce the impact of instrument noise, scattering effects, and baseline drift on subsequent modeling results, a systematic spectral preprocessing was performed on the raw spectral data. Parallel preprocessing steps included multivariate scattering correction (MSC), standard normal transformation (SNV), Savitzky-Golay (SG) smoothing, SG first derivative, and SG second derivative. Due to the significant interaction between the baseline model structure and the processing strategy, various baseline models were matched with different preprocessing methods to select the optimal method. The best preprocessing method for each modeling algorithm was determined based on evaluation parameters. After preprocessing, a near-infrared spectral feature vector was obtained: .

[0045] Step S5: Low-dimensional color feature extraction.

[0046] For the sample images, a center-cropping strategy was employed to reduce edge effects and background interference, cropping the image size to 32×32 pixels. Then, low-dimensional color features were extracted from the images, including the mean and standard deviation of the RGB channels, and the mean and standard deviation of the Lab channels. Channel histogram features, preferably using Lab features as the primary color modality input, convert the image from the RGB color space to the CIE Lab color space, and calculate... , and The average value of the three channels is used to characterize the brightness and chromaticity properties of the sample, and Lab features are extracted. The extracted features are used for multimodal feature fusion and quantitative modeling analysis.

[0047] Step S6: Construct the state representation vector of the honey refining process.

[0048] The preprocessed spectral features and the Lab color features extracted from the image were standardized and integrated into a unified feature matrix for quantitative regression analysis. The fused features are: This approach unifies the spectral information of internal composition state variables, which characterize changes in the sample's internal composition, with the low-dimensional color information of external thermal reaction state variables, which characterize the evolution of the sample's appearance, into a single descriptive space. This results in a more complete input representation of the honey refining process. When it is necessary to unify the dimensions of different modalities, the spectral and color features are first standardized separately before constructing the state representation vector. For a spectral feature after fusion The fused feature matrix is: (1); in, Represents spectral characteristic variables; represents the color feature variable; n represents the total number of samples; m represents the number of spectral wavelengths. By using formula (1), the near-infrared spectral variable and the Lab low-dimensional color variable are spliced ​​together according to the sample dimension to form a unified feature matrix, integrating the molecular composition and appearance color information into the same space, completely depicting the state of the honey refining process, and providing standardized input for multimodal modeling.

[0049] S7: Establish a comprehensive characterization model for the honey refining process using PLS and XGBoost.

[0050] Based on the honey refining process state characterization vector constructed in step S6, a comprehensive characterization model for the honey refining process is established to characterize key quality attributes and process states during the honey refining process. Key quality attributes include glucose, fructose, moisture, pH, and 5-HMF, with 5-HMF being the preferred key quality marker for detection. Because the honey refining process data exhibits significant process continuity, samples show continuous evolution characteristics as the process progresses. Simultaneously, there are numerous near-infrared spectral variables with strong correlations between them, and color and spectral variables also show a certain coupling relationship. For this input structure characterized by strong collinearity, small sample size, and continuous reaction process, partial least squares (PLS) is adopted as the primary comprehensive characterization model. Partial least squares regression is more suitable for mapping multimodal inputs to a unified latent variable space and establishing stable comprehensive characterization relationships. (2); (3); in, This is a spectral-low-dimensional color feature fusion matrix. The latent variable score matrix, and For the load matrix, and Let Y be the residual matrix and Y be the response variable matrix. Using formulas (2) and (3), the fusion feature matrix is ​​decomposed into latent variable scores, loadings and residuals. The high-dimensional collinear multimodal data is projected onto the low-dimensional latent variable space to extract key information, reduce modeling complexity, and ensure the stability and reliability of linear representation. The latent variables are used to establish a linear mapping between fusion features and key quality attributes to achieve quantitative prediction of indicators such as glucose, fructose, and 5-HMF, and to capture the linear change law of components and quality during the honey refining process.

[0051] To better analyze the characterization information of the honey refining process, a nonlinear model is further introduced as a supplementary analysis model to examine whether key quality indicators rely more on the expressive power of the nonlinear model and to provide supplementary explanations for variable contributions. In this invention, the extreme gradient boosting model (XGBoost) is used to supplement the examination of the nonlinear information expressive power in the fused features and to provide a comparative basis for subsequent interpretive analysis. (4); in, The loss function used in this invention is the squared error loss; K represents the number of trees; Represents the k-th decision tree; Represents the measured value; Represents the predicted value; Represents the total loss; This is a regularization term used to control model complexity. (5); in, This indicates the number of leaf nodes in the tree. and For regularization parameters, an early stopping strategy was adopted during training based on the performance on the validation set to prevent model overfitting; This represents the sum of squared weights of the leaf nodes.

[0052] Formula (4) is used to measure the prediction bias by squared error, and the regularization term is superimposed to control the model complexity, taking into account both fitting accuracy and generalization ability, capturing the nonlinear relationship between fusion features and quality attributes, and supplementing the information not covered by PLS. Formula (5) is used to constrain the tree structure with the number of leaf nodes and the sum of squared weights to prevent overfitting, improve the robustness of the model under small sample and batch perturbation, and adapt to the nonlinear modeling needs of honey refining industry scenarios.

[0053] In terms of model evaluation, the coefficient of determination is used. The root mean square error (RMSE) and interquartile range (RPIQ) are used to evaluate the comprehensive characterization model.

[0054] Therefore, this invention employs a PLS comprehensive characterization model at the modeling level, with XGBoost as a supplementary analysis model: PLS is used to extract and characterize feature information during the honey refining process, while XGBoost is used to supplement the analysis of key quality attributes from a nonlinear perspective. This approach maintains the stability of the comprehensive characterization model under conditions of strong collinearity and continuous reaction processes, while also taking into account its ability to describe the nonlinearity of quality attributes.

[0055] As a further implementation, when it is necessary to enhance nonlinear compensation, XGBoost can be used to further correct the output deviation of the main representation model. However, the basic scheme of the present invention does not necessarily require strict residual correction.

[0056] Step S8: Variable contribution analysis and mechanism consistency judgment of the hierarchical explanation framework.

[0057] After completing the comprehensive characterization model construction in step S7, this invention further conducts model interpretability analysis to determine whether the key variables that truly play a role in the model and their mechanisms of action are consistent with the physicochemical change pathways during the honey refining process, such as sugar conversion, moisture changes, browning intensification, and accumulation of thermal reaction products. This provides mechanistic support for the comprehensive characterization results. This invention employs a hierarchical interpretation framework, following a hierarchical analysis of global importance identification, global contribution direction analysis, and single-sample decision path analysis, to systematically analyze the key variables driving model predictions and their contribution directions to the prediction results. The hierarchical interpretation framework includes: VIP curves, SHAP summary plots, and single-sample force plots.

[0058] Variable projection importance (VIP) analysis of the PLS model, scoring analysis of fused features, and identification of key wavelength and key color variables that have a significant impact on the model output across the entire sample range: (6); in, Indicates the number of predictor variables. The number of latent variables. Indicates the first The sum of squares explained by each component for the response variable. Indicates the first The first component The weights of each variable; The modulus represents the weight of all variables for the a-th latent variable. Formula (6) quantifies the contribution weight of each variable to the latent variable and quality attributes, filters key wavelength and color variables, clarifies the driving factors for model prediction, and provides a globally interpretable basis for the PLS model.

[0059] The XGBoost model uses the SHAP method to quantify the contribution of each feature to a single prediction result. The SHAP summary plot analyzes the direction and strength of the contribution of key variables. (7); in, For the model's expected output, For the first The contribution value of each feature to the sample output; The total number of representative features; Representing the The model predicts the key quality attributes of each sample. Using formula (7) as the benchmark, the independent contribution of each feature to the prediction of a single sample is broken down, and the direction and intensity of the influence of the features are quantitatively displayed, so as to realize the local and global interpretability of the XGBoost model.

[0060] After completing the global variable screening and global contribution direction analysis, this invention further employs a single-sample approach to provide a local interpretation of representative samples, revealing how the prediction results of representative samples are jointly driven by key spectral and color variables. This analysis determines whether color modes supplement spectral modes or merely exist as redundant variables. The determination of the variable contribution structure shows a high degree of consistency with the physicochemical transformation pathways in the honey refining process, thereby further enhancing the reliability and interpretability of the proposed modeling strategy.

[0061] Step S9: Output the comprehensive characterization results.

[0062] The comprehensive characterization model output from step S7 is combined with the variable contribution analysis results from step S8 to obtain the comprehensive characterization results of the honey refining process. These results include key quality attribute results, process state characterization results, and variable contribution analysis results, which can be used for state assessment, process trend analysis, and quality control assistance in the honey refining process.

[0063] The honey refining experiment was conducted under the actual industrial production conditions of Hongjitang Pharmaceutical Factory. Approximately 30L of honey was weighed for each batch, sieved to remove dead bees and visible impurities, and then transferred to a 50L vacuum concentration refining tank for refining. During the refining process, the following parameters were controlled: tank temperature maintained at 85–105℃, vacuum degree controlled at 0.08 MPa (gauge pressure), and the refining process continued continuously for 70 minutes. Once the refining tank reached the preset operating conditions and entered the stable refining stage, samples were collected through the sampling port of the external circulation pipeline of the online honey refining detection system. Online spectral measurement was performed using a transmission flow cell with a 2mm optical path, connected to the spectrometer via a 600μm core optical fiber. This configuration enabled continuous online acquisition of spectral information during the honey refining process. One near-infrared spectral feature vector was obtained. Sampling was performed at 3-minute intervals in each honey refining batch, yielding approximately 22 samples per batch. A total of 7 honey refining batches were completed, resulting in a total of 157 samples.

[0064] Reference concentrations of glucose (GLU), fructose (FRU), and 5-hydroxymethylfurfural (5-HMF) in honey samples were determined using high-performance liquid chromatography (HPLC). Chromatographic analysis was performed on an Agilent 1260 HPLC system (Agilent Technologies). Moisture content was determined using refractive index, and pH was determined using a pH meter (FE30, Mettler Toledo).

[0065] In all batches, 5-HMF showed a clear upward trend with the progression of the honey refining process. Its concentration ranged from 4.82 to 601.13 mg / kg, with an average of 92.83 mg / kg. In most batches, 5-HMF remained at a low level in the early stages, while increasing rapidly in the middle and later stages, indicating a cumulative thermal reaction during honey refining. To further verify the differences between different refining stages, the samples were divided into three stages: early, middle, and late refining. Both within individual batches and in the overall data across all batches, the content of 5-HMF increased significantly from the early to the late stages (p < 0.001). This statistically significant accumulation phenomenon indicates that the thermal processing is the main driving factor for 5-HMF formation. The honey refining process is characterized by a gradual decrease in moisture and a significant accumulation of 5-HMF, while the main sugar components remain relatively stable. The strong coupling between moisture reduction and 5-HMF formation provides an important mechanistic basis for subsequent spectral analysis and image modeling studies.

[0066] Brightness parameters extracted through image analysis To assess the color changes during the honey refining process, the color of the honey samples gradually deepened from the early to the later stages. The honey color transitioned from light yellow to deep amber, providing direct visual evidence of the cumulative thermal reaction process. Image-extracted color parameters can effectively capture systematic changes related to the thermal reaction process. Simultaneously, The significant statistical coupling between color features and 5-HMF generation provides a clear mechanistic basis for introducing color features into multimodal regression models.

[0067] After acquiring spectral signals and image data, pairing processing based on the time synchronization principle is performed to form multimodal sample (S,I) data pairs. Various preprocessing steps are performed on the spectral data in parallel, and low-dimensional color features are extracted from the image data, with Lab color space features being the preferred method. The calculation... , and The average value of the three channels. Spectral characteristics. and low-dimensional color features Standardized fusion was performed to construct a spectral-color fusion matrix with 157 samples: (8); in, , ··· Characteristic values ​​representing 2075 wavelength points in the near-infrared spectrum.

[0068] To evaluate the contribution of image-extracted features to the prediction of key physicochemical indicators during honey refining, the linear model PLS and the nonlinear model XGBoost were evaluated using spectral features and spectral-color fusion features, respectively. The coefficient of determination was selected as the preferred evaluation metric. The root mean square error (RMSE) and interquartile range (IRR) performance ratios were compared. Results are shown in Table 1. From a mechanistic perspective, indicators closely related to the thermal reaction process (such as 5-HMF, GLU, and FRU) are more easily reflected in color changes, thus benefiting from the introduction of color features. Moisture content is directly related to strong absorption bands in the near-infrared spectrum, therefore spectral information itself plays a dominant role. As an indirect reaction indicator, pH seems to rely more on the expressive power of the nonlinear model than on additional color information.

[0069] Table 1: Performance comparison of models based on single spectral features and those based on spectral-Lab fusion features

[0070] As shown in Table 1, compared to single spectral features, fusing low-dimensional Lab color features improved the R², decreased the RMSE, and increased the RPIQ of multiple indicators under the PLS model. In particular, the key quality marker 5-HMF achieved an R² of 0.9803 in the PLS model. Multimodal fusion significantly improved prediction performance, indicating that the proposed solution can retain internal composition information while introducing appearance evolution information, thereby improving the comprehensive characterization of the honey refining process. For both modeling methods, the introduction of Lab features resulted in consistent performance improvements. Paired t-test results showed that the performance improvement of the multimodal model compared to the single-spectral model was statistically significant (p<0.05), indicating that color information provides independent and complementary predictive information in the quantitative prediction of 5-HMF. This result is also consistent with the physicochemical mechanism. The formation of 5-HMF is mainly driven by sugar dehydration and Maillard reactions, which are usually accompanied by the gradual browning of the honey system. The parameters in the Lab color space... (brightness), (Red-Green axis) and The yellow-blue axis is sensitive to these visual changes. Therefore, color features can capture information about macroscopic changes caused by the reaction, which complements the molecular vibrational information described by near-infrared spectroscopy, thereby enhancing the characterization of the overall features of the honey refining process.

[0071] VIP analysis of PLS ​​model, such as Figure 2 As shown in the figure, the VIP distribution of the fusion model reveals that, apart from the key spectral regions, the color variable... , and It also shows a high VIP value, among which The contribution of color variables is particularly significant. In the fusion model, color variables make a crucial contribution to the construction of the latent variable structure. The formation of 5-HMF is accompanied by sugar dehydration and Maillard reactions, which cause the honey system to gradually darken. The decrease in value (indicating a decrease in brightness) and Increases in the values ​​(indicating enhanced red) are closely related to the reaction process. Therefore, color features not only introduce an additional dimension of information but also statistically align with the underlying chemical reaction pathways. The primary spectral regions emphasized by the PLS model are consistent with known near-infrared response bands, while the color features in the fused model possess independent and physically interpretable importance. Thus, this multimodal model does not simply combine unrelated variables but establishes a synergistic representation between molecular and visual information.

[0072] SHAP analysis of the XGBoost model, such as Figure 3 As shown, the variable contributions in the nonlinear model are further clarified under fusion conditions. It has become one of the most important variables in the ranking. The SHAP value distribution indicates that a lower SHAP value... The value (corresponding to a darker sample color) has a positive contribution to the predicted 5-HMF content, while a higher value... The value then shows a negative contribution, thus lowering the predicted value. This contribution direction is completely consistent with the actual physical phenomenon, that is, as the honey refining process progresses, the color gradually deepens and 5-HMF continuously accumulates. Furthermore, the variable... It also showed a significant contribution, which further supports the association between increased redness and the advancement of the reaction process.

[0073] like Figure 4 As shown, single-sample explanation and prediction path analysis, in the fusion model, In conjunction with several key wavelength variables, the predicted value shifts from the baseline to a lower final output value. Multiple variables collaboratively determine the final prediction, and the direction of each variable's contribution aligns with its physicochemical significance. In contrast, single-spectral models rely primarily on a few wavelength variables, making their predicted changes more susceptible to the dominance of local spectral regions. This difference indicates that the multimodal fusion model better reflects the actual reaction pathways in the honey refining process. As the thermal reaction progresses, changes in molecular structure occur within the system, manifesting not only as spectral changes but also as changes in visible color. By integrating these two dimensions of information, the model generates a more comprehensive predicted response.

[0074] Therefore, the advantages of multimodal models are not only reflected in the improvement of statistical performance indicators, but more importantly, their variable contribution structure is highly consistent with the physicochemical transformation pathways in the honey refining process, thereby further enhancing the reliability and interpretability of the proposed modeling strategy.

[0075] The results show that combining molecular-level spectral information with macroscopic changes in the appearance of thermal reactions can simultaneously improve the robustness of predictions and the ability to explain mechanisms. The proposed framework provides a practical and scalable paradigm for the intelligent monitoring of complex thermal processing systems and is expected to be extended to other heat-driven food and pharmaceutical manufacturing processes.

[0076] Figure 5 A honey refining process characterization system based on near-infrared spectroscopy and color characteristics is shown, comprising: The spectral image acquisition unit 501 is configured to acquire near-infrared spectral data of the sample in the flow cell and the color image corresponding to the sample, wherein the sample in the flow cell is the sample input through the external circulation pipeline during the honey refining process. The feature fusion construction unit 502 is configured to: extract low-dimensional color features from the color image, and couple and fuse the preprocessed near-infrared spectral data with the low-dimensional color features to construct a honey refining process state characterization vector; The dual-mode modeling and characterization unit 503 is configured to: establish a partial least squares regression model as the main characterization model based on the honey refining process state characterization vector, and introduce an extreme gradient boosting model as a nonlinear supplementary analysis model to obtain a comprehensive characterization result of the honey refining process; Mechanism consistency criterion unit 504 is configured to: based on the comprehensive characterization results of the honey refining process, use a hierarchical interpretation framework to perform variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model, and determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

[0077] It is understood that the aforementioned units can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effect of the embodiments of the present invention. The aforementioned units are based on logical functional division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of the present invention, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0078] According to another embodiment of the present invention, the system of this embodiment can be constructed by running a computer program (including program code) capable of performing the steps involved in the corresponding method of the present invention on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the aforementioned computing device through the computer-readable recording medium, and run therein.

[0079] This implementation also proposes a honey refining process characterization system based on near-infrared spectroscopy and color characteristics, including: a honey refining device; and an external circulation pipeline connected to the honey refining device for transporting samples during the honey refining process. A flow cell, which is located on the external circulation pipeline, is used to hold the sample to be tested; And a processor, configured to execute the aforementioned method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics.

[0080] Figure 6 A computer device is shown, which includes a processor 601, a communication interface 602, and a computer-readable storage medium 603. The processor 601, communication interface 602, and computer-readable storage medium 603 can be connected via a bus or other means.

[0081] The communication interface 602 is used to receive and send data. The computer-readable storage medium 603 can be stored in the memory of the electronic device. The computer-readable storage medium 603 is used to store computer programs, which include program instructions. The processor 601 is used to execute the program instructions stored in the computer-readable storage medium 603.

[0082] The processor 601 is the computing and control core of an electronic device. It is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to achieve the corresponding method flow or corresponding function.

[0083] Processor 601 is configured to perform the following procedure: Near-infrared spectral data and corresponding color images of the samples in the flow cell were obtained. The samples in the flow cell were samples input through the external circulation pipeline during the honey refining process. Low-dimensional color features are extracted from color images, and preprocessed near-infrared spectral data are coupled and fused with low-dimensional color features to construct a state characterization vector for the honey refining process. Based on the state representation vector of the honey refining process, a partial least squares regression model is established as the main representation model, and an extreme gradient boosting model is introduced as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. Based on the comprehensive characterization results of the honey refining process, a hierarchical interpretation framework was used to conduct variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model to determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

[0084] This invention also provides a computer-readable storage medium, which is a memory device in an electronic device for storing programs and data. It is understood that the computer-readable storage medium here may include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space for storing the processing system of the electronic device.

[0085] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory; alternatively, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0086] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to perform the following process: Near-infrared spectral data and corresponding color images of the samples in the flow cell were obtained. The samples in the flow cell were samples input through the external circulation pipeline during the honey refining process. Low-dimensional color features are extracted from color images, and preprocessed near-infrared spectral data are coupled and fused with low-dimensional color features to construct a state characterization vector for the honey refining process. Based on the state representation vector of the honey refining process, a partial least squares regression model is established as the main representation model, and an extreme gradient boosting model is introduced as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. Based on the comprehensive characterization results of the honey refining process, a hierarchical interpretation framework was used to conduct variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model to determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

[0087] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0088] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital cable) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics, characterized in that, Includes the following processes: Near-infrared spectral data and corresponding color images of the samples in the flow cell were obtained. The samples in the flow cell were samples input through the external circulation pipeline during the honey refining process. Low-dimensional color features are extracted from color images, and preprocessed near-infrared spectral data are coupled and fused with low-dimensional color features to construct a state characterization vector for the honey refining process. Based on the state representation vector of the honey refining process, a partial least squares regression model is established as the main representation model, and an extreme gradient boosting model is introduced as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. Based on the comprehensive characterization results of the honey refining process, a hierarchical interpretation framework was used to conduct variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model to determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

2. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 1, characterized in that, The hierarchical interpretation framework is used to conduct variable contribution analysis on the principal representation model, including: Calculate the variable projection importance value of each variable in the main representation model, and identify the key wavelength variables and key color variables that have a significant impact on the prediction of key quality attributes within the overall sample range; The process of calculating the projected importance value of variables is as follows: the sum of squares of the explanations of each latent variable component to the response variable is calculated, and the weights of each variable in each component are combined to obtain the weighted sum of the contributions of each variable to all latent variable components, thereby quantifying the importance of each variable in constructing a unified latent variable space.

3. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 1, characterized in that, A hierarchical interpretation framework is used to perform variable contribution analysis on a nonlinear supplementary analysis model, including: The SHAP method is used to quantify the contribution of each feature in the nonlinear supplementary analysis model to a single prediction result, and a SHAP summary plot is generated. Based on the generated SHAP summary diagram, analyze the direction and intensity of the contribution of key wavelength variables and key color variables to the prediction results of key quality attributes. The calculation logic for the contribution value is as follows: based on the expected output of the model, calculate the marginal contribution of the model output when each feature exists or not, and obtain the independent contribution of each feature to the final predicted value by arranging and combining the average marginal contribution of all feature subsets.

4. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 1, characterized in that, Key quality attributes include glucose content, fructose content, moisture content, pH value, and 5-hydroxymethylfurfural content; Partial least squares regression model was used to characterize the linear trends of glucose content, fructose content, moisture content, pH value and 5-hydroxymethylfurfural content; The extreme gradient boosting model is used to detect whether there are deviations in glucose content, fructose content, moisture content, pH value, and 5-hydroxymethylfurfural content caused by nonlinear coupling between internal composition state variables and external thermal reaction state variables.

5. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 4, characterized in that, To determine the consistency between the model's prediction mechanism and the physicochemical reaction pathways in the honey refining process, including: Based on the variable projection importance values ​​and SHAP summary plot, the set of key variables driving the model prediction is determined; Analyze whether the contribution direction of the key color variables in the key variable set is consistent with the increasing browning trend caused by sugar dehydration and Maillard reaction during honey refining; If the key color variable representing a decrease in brightness contributes positively to the prediction of 5-hydroxymethylfurfural content, and the key color variable representing an increase in redness contributes positively to the prediction of 5-hydroxymethylfurfural content, then the model prediction mechanism is considered to be consistent with the physicochemical reaction pathway.

6. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 1, characterized in that, The hierarchical explanation framework also includes single-sample decision path analysis, which includes: By selecting representative samples and using single samples, we aim to demonstrate how the prediction results of representative samples are jointly driven by key wavelength variables and key color variables. By comparing the variable contribution distributions of the single-spectral model and the multimodal fusion model on representative samples, we can determine whether low-dimensional color features are used to supplement missing information in near-infrared spectral data or exist as redundant variables.

7. The method for characterizing the honey refining process based on near-infrared spectroscopy and color characteristics as described in claim 1, characterized in that, The output of the comprehensive characterization results includes: Quantitative predictions of key quality attributes based on the output of the main characterization model and the nonlinear supplementary analysis model; Based on the hierarchical interpretation framework, the report on the contribution analysis of key variables includes the global importance ranking and local contribution direction of key wavelength variables and key color variables to key quality attributes. The process status credibility evaluation generated based on the consistency judgment results is used to guide the status judgment, process trend analysis and quality control assistance of the honey refining process.

8. A honey refining process characterization system based on near-infrared spectroscopy and color characteristics, characterized in that, include: The spectral image acquisition unit is configured to acquire near-infrared spectral data of the sample in the flow cell and the color image corresponding to the sample, wherein the sample in the flow cell is the sample input through the external circulation pipeline during the honey refining process; The feature fusion construction unit is configured to: extract low-dimensional color features from the color image, and couple and fuse the preprocessed near-infrared spectral data with the low-dimensional color features to construct a honey refining process state representation vector; The dual-mode modeling and representation unit is configured to: establish a partial least squares regression model as the main representation model based on the state representation vector of the honey refining process, and introduce an extreme gradient boosting model as a nonlinear supplementary analysis model to obtain a comprehensive representation result of the honey refining process. The mechanism consistency criterion unit is configured to: based on the comprehensive characterization results of the honey refining process, use a hierarchical interpretation framework to perform variable contribution analysis on the main characterization model and the nonlinear supplementary analysis model, and determine the consistency between the model prediction mechanism and the physicochemical reaction pathway in the honey refining process.

9. A honey refining process characterization system based on near-infrared spectroscopy and color characteristics, characterized in that, include: Honey refining device; An external circulation pipeline, which is connected to the honey refining device, is used to transport samples during the honey refining process; A flow cell, which is located on the external circulation pipeline, is used to hold the sample to be tested; And a processor configured to perform the process of the honey refining process characterization method based on near-infrared spectral and color characteristics as described in any one of claims 1-7.

10. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the honey refining process characterization method based on near-infrared spectral and color characteristics as described in any one of claims 1 to 7.

Images