Semi-refined tea grading method based on carbon baking parameters, medium and equipment

Abstract

The invention discloses a semi-refined tea grading method based on carbon baking parameters, a medium and equipment, and the method comprises the steps: collecting carbon baking process parameters and tea basic parameters, extracting thermodynamic characteristics and a Maillard reaction characteristic spectrum, and constructing a variety adaptability model; performing multi-modal fusion on the thermodynamic characteristics, the Maillard reaction characteristic spectrum and the variety adaptability parameters to generate a carbon baking quality prediction vector; performing preliminary grade judgment through a grading decision tree model, performing cascade fire compensation optimization on secondary tea leaves, and performing re-judgment; and finally, the grading result and the carbon baking parameters are stored in the block chain in an associated manner. According to the method, dynamic correlation analysis of carbon baking process parameters and tea quality is realized, the problem of inaccurate evaluation caused by separation of a traditional grading method and a carbon baking process is solved, scientificity and reliability of a grading result are remarkably improved, and meanwhile, data support is provided for process optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of tea processing, and in particular to a semi-refined tea grading method based on carbon roasting parameters, a medium and equipment. Background Art

[0002] Semi-refined tea is an intermediate state of fresh tea leaves, after undergoing primary processing (e.g., withering, fixing, rolling, and drying) but before completing refining (e.g., sorting, grading, blending, and roasting). Grading semi-refined tea can optimize quality consistency, facilitate subsequent refining, enhance the market value of the finished tea, and meet diverse consumer demands. Since tea grading is often completed before the carbonization roasting step, it fails to fully utilize the process data generated during subsequent processing. This fragmented evaluation approach can lead to prematurely assigning low grades to raw materials with potential for quality improvement. Furthermore, the lack of continuous monitoring and analysis of processing parameters makes it difficult to optimize process parameters to improve product quality. Furthermore, this static grading approach fails to provide effective feedback for subsequent process adjustments, limiting the scope for optimization throughout the production process. Establishing a dynamic link between processing and quality evaluation is a key issue to address in improving tea grading. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to propose a semi-refined tea grading method, medium and equipment based on carbon roasting parameters to solve the problem that the existing grading method is separated from the subsequent refining process, resulting in inaccurate quality evaluation and lack of basis for process optimization.

[0004] In order to achieve the above technical objectives, in a first aspect, the present application provides a semi-refined tea grading method based on carbon roasting parameters, comprising: Collect parameters of the carbon roasting process and basic parameters of tea leaves, including variety information, moisture content information and primary processing grade; Feature extraction of carbon baking process parameters was performed to obtain thermodynamic characteristics, including carbon baking uniformity index and key temperature duration; Furthermore, hyperspectral imaging information of tea leaves in the carbon roasting space is collected, the Maillard reaction characteristic spectrum within a preset wavelength range is extracted, and the spectral parameters of the tea leaves are calculated based on the Maillard reaction characteristic spectrum; Dynamically construct a variety adaptability model based on tea spectral parameters and tea basic parameters to obtain variety adaptability parameters; Thermodynamic characteristics, Maillard reaction characteristic spectrum and variety adaptability parameters are fused into a multimodal manner to generate a carbon roasting quality prediction vector. The multimodal fusion includes temperature-material transformation synergistic splicing and spatial inhomogeneity compensation. The carbon roasting quality prediction vector is input into the hierarchical decision tree model to obtain preliminary grade determination results, which include inferior, superior, and special grades. The hierarchical decision tree model is configured to trigger the cascade re-fire optimization process based on the cluster quality index; Cascaded re-fire optimization is performed on tea leaves that are initially rated as sub-grade to obtain optimized tea leaves. The optimized tea leaves are then re-graded to obtain updated grade results. The optimization strategy of the cascaded re-fire optimization is configured to use a material conversion model to combine and adjust the re-fire duration and temperature. The final grading result of the current tea is generated based on the preliminary grading results and the updated grading results. The final grading result is associated with the carbon roasting process parameters and stored in the blockchain to generate a grading report including grade distribution probability, process defect location and energy consumption assessment.

[0005] In some embodiments, the carbon baking process parameters include three-dimensional temperature field information and duration of the baking cage; The carbon baking process parameters are extracted to obtain thermodynamic characteristics. The thermodynamic characteristics include the carbon baking uniformity index and the duration of key temperatures, including: The ratio of the standard deviation to the mean of the temperature values ​​corresponding to each temperature measurement point is calculated based on the three-dimensional temperature field information of the baking cage to generate the carbon baking uniformity index; Furthermore, a temperature-time curve is constructed based on the three-dimensional temperature field information and duration of the roasting cage, and temperature platform intervals whose duration exceeds a preset time threshold are extracted from the curve. Each temperature platform interval corresponds to a temperature fluctuation range. The temperature platform interval whose temperature fluctuation range meets the preset fluctuation deviation range is recorded as a key platform, and the duration corresponding to the key platform is recorded as the key temperature duration. Furthermore, temperature gradient distribution information is generated based on the temperature values ​​of adjacent temperature measurement points, and abnormal temperature points in the temperature gradient distribution information that exceed a preset average temperature range are corrected by interpolation compensation based on a spatially weighted neighborhood to obtain corrected temperature field information; Thermodynamic characteristics are generated based on the carbon baking uniformity index, key temperature duration and corrected temperature field information.

[0006] In some embodiments, collecting hyperspectral imaging information of tea leaves in a carbon roasting space, extracting a Maillard reaction characteristic spectrum within a preset wavelength range, and calculating tea spectral parameters based on the Maillard reaction characteristic spectrum include: Collecting reflectance spectrum data of the tea leaves' surface through a hyperspectral imaging system, and selecting at least two characteristic bands within a preset band range, the characteristic bands including a first reaction characteristic band and a second reaction characteristic band; Perform absorbance conversion on the reflectance spectrum data of the first reaction characteristic band and the second reaction characteristic band to generate a time-absorbance change curve of the corresponding band; The time-absorbance change curve is smoothed to eliminate noise interference and extract characteristic points on the curve, including the absorbance maximum, inflection point and platform interval; According to the position and change trend of the characteristic points, the absorbance of the first reaction characteristic band and the absorbance of the second reaction characteristic band are calculated and the integral ratio is generated to generate the reaction progress index. The integral ratio adopts the trapezoidal numerical integration method; Monitor the rate of change of the reaction progress indicator over time. When the rate of change of the reaction progress indicator over time exceeds the set threshold, it is marked as a key turning point of the Maillard reaction. Based on the temperature and duration corresponding to the key turning point of the Maillard reaction, the cumulative amount of melanoidins and the dynamic change trend information of the aldehyde-ketone ratio are calculated. The cumulative amount of melanoidins is obtained by accumulating the absorbance changes after the key turning point of the Maillard reaction, and the dynamic change trend information of the aldehyde-ketone ratio is obtained by establishing a time series ratio curve. Tea spectral parameters are generated based on the accumulation of melanoidins, dynamic change trend information of aldehyde-ketone ratio and reaction progress indicators.

[0007] In some embodiments, a variety adaptability model is dynamically constructed based on tea spectral parameters and tea basic parameters to obtain variety adaptability parameters, including: Match the variety information with the preset variety feature library to obtain the reference substance conversion curve corresponding to the current variety information. The reference substance conversion curve includes the expected range of variation of the aldehyde-ketone ratio and the threshold value of the melanoidin accumulation amount; According to the dynamic change trend information of the aldehyde-ketone ratio in the tea spectral parameters, the dynamic time warping distance between the real-time aldehyde-ketone ratio curve and the reference substance conversion curve is calculated to generate a first matching index; and, calculating the relative deviation percentage between the current melanoidin accumulation amount and the expected value of the reference substance conversion curve based on the melanoidin accumulation amount in the tea spectral parameters, to generate a second matching index; The first matching index and the second matching index are weighted and fused to obtain a comprehensive matching index. The first matching weight corresponding to the dynamic change trend information of the aldehyde-ketone ratio increases with the increase of carbon baking temperature, and the second matching weight corresponding to the cumulative amount of melanoidin increases with the extension of the duration. Comparing the comprehensive matching index with the preset index threshold range, and generating temperature adjustment information based on the comparison result; and generating carbon baking amplitude variation information based on the moisture content information; Generate variety adaptability parameters based on temperature adjustment information and carbon roasting amplitude change information.

[0008] In some embodiments, thermodynamic characteristics, Maillard reaction characteristic spectra, and variety adaptability parameters are multimodally fused to generate a carbon roasting quality prediction vector. The multimodal fusion includes temperature-substance conversion collaborative splicing and spatial heterogeneity compensation, including: The temperature field compensation coefficient is generated according to the carbon baking uniformity index, and the dynamic change trend information of the aldehyde-ketone ratio in the Maillard reaction characteristic spectrum is regionally corrected and recorded as the correction trend information; The temperature adjustment information in the variety adaptability parameters is coupled with the duration of the key temperature to obtain a time weighting factor. The time weighting factor is used as the weight value of the Maillard reaction characteristic spectrum at different carbon roasting stages and weighted to obtain a weighted reaction characteristic spectrum. Furthermore, based on the moisture content information, the temperature field compensation coefficient and the time weighting factor are dynamically scaled to generate a moisture content adaptation adjustment coefficient; A spatial grid mapping relationship is constructed, and the correction trend information, weighted response characteristic spectrum, and moisture content adaptation adjustment coefficient are integrated according to spatial position. In addition, abnormal grid points in the spatial grid mapping relationship are smoothed in the neighborhood. The smoothing process retains the temperature gradient characteristics while eliminating isolated noise points to obtain spatial grid information. The spatial grid information is expanded along the time dimension to generate a carbon roasting quality prediction vector that includes temperature-material transformation synergistic characteristics and spatial distribution characteristics.

[0009] In some embodiments, the carbon roasting quality prediction vector is input into a hierarchical decision tree model to obtain a preliminary grade determination result. The preliminary grade determination result includes inferior grade, superior grade, and special grade, including: A hierarchical decision tree model was established, which included a temperature-substance transformation synergistic feature branch, a spatial distribution feature branch, and a Maillard reaction integrity branch. The temperature-material conversion synergistic feature value in the carbon roasting quality prediction vector is input into the temperature-material conversion synergistic feature branch, and when the temperature-material conversion synergistic feature value is lower than the first feature threshold, it is judged as secondary; Input the spatial distribution characteristics in the carbon roasting quality prediction vector into the spatial distribution feature branch, calculate the quality uniformity index, and judge it as excellent when the quality uniformity index is lower than the second feature threshold; The Maillard reaction features in the carbon-roasted quality prediction vector are input into the Maillard reaction integrity branch. When the product of the slope of the aldehyde-ketone ratio curve and the cumulative amount of melanoidins exceeds the third characteristic threshold, it is judged as special grade. According to the variety information in the basic parameters of tea, the corresponding first characteristic threshold, second characteristic threshold and third characteristic threshold are extracted from the preset grading standard library; Performing time decay compensation on the first characteristic threshold, the second characteristic threshold, and the third characteristic threshold according to the actual carbon baking time in the carbon baking process parameters; Output includes preliminary grade determination results of inferior, superior and special grades; The hierarchical decision tree model is configured to trigger a cascading re-fire optimization process based on the clustering quality index, including: Calculate the clustering quality index of the preliminary grade judgment results. The clustering quality index is determined by the variance of the feature space distance between the special grade samples and the excellent grade samples. When the clustering quality index is lower than the preset clustering threshold, the cascaded supplementary fire optimization process is activated.

[0010] In some embodiments, the optimization strategy for cascade supplementary fire optimization is configured to generate a combination of supplementary fire duration and temperature using a material conversion model, including: A material transformation kinetics model is established. The input features of the material transformation kinetics model include the deviation value of the temperature-material transformation synergistic feature and the spatial distribution feature heterogeneity in the carbon roasting quality prediction vector. Calculate the optimal supplementary fire parameter combination based on the material transformation kinetics model, and generate the optimization strategy of cascade supplementary fire optimization based on the optimal supplementary fire parameter combination; Cascaded supplementary fire optimization is performed on tea leaves that are initially graded as inferior to obtain optimized tea leaves, and the optimized tea leaves are re-graded to obtain updated grade determination results, including: Re-collect and optimize the carbon roasting process parameters and hyperspectral imaging information of tea leaves, record them as optimized carbon roasting parameters and optimized imaging information, and update the carbon roasting quality prediction vector, record them as optimized quality prediction vector; The optimized quality prediction vector is input into the hierarchical decision tree model for secondary judgment to generate an updated grade judgment result.

[0011] In some embodiments, a final grading result of the current tea is generated based on the preliminary grading result and the updated grading result. The final grading result is associated with the carbon roasting process parameters and stored in the blockchain. A grading report containing the grade distribution probability, process defect location, and energy consumption assessment is generated, including: The weighted average of the preliminary special-grade ratio, preliminary superior-grade ratio, and preliminary sub-grade ratio in the preliminary grading results and the updated special-grade ratio, updated superior-grade ratio, and updated sub-grade ratio in the updated grading results is calculated to generate the final special-grade ratio, final superior-grade ratio, and final sub-grade ratio in the final grading results; Establish a correlation mapping relationship between the final classification results and the three-dimensional temperature field information of the baking cage and the duration of key temperatures in the carbon baking process parameters. The correlation mapping relationship includes the spatial correspondence between temperature field uniformity and quality grade distribution, and the temporal correspondence between key temperature duration and Maillard reaction integrity. The final grading results, carbon baking process parameters and associated mapping relationships are hashed to generate data blocks, which are then written to the blockchain network for distributed storage. The data blocks contain the grade distribution probability matrix, process defect location coordinates and carbon baking energy consumption statistical indicators. Based on the data blocks, a visual grading report is generated, which includes a three-dimensional temperature field distribution and a grade distribution superimposed thermal map, process defect improvement suggestions, and an energy efficiency optimization analysis report.

[0012] In a second aspect, the present invention further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0013] In a third aspect, the present invention further provides an electronic device comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect.

[0014] By adopting the above-mentioned technical solution, the present invention has the following beneficial effects compared with the prior art: the present invention discloses a semi-refined tea grading method, medium and equipment based on carbon roasting parameters. The method collects carbon roasting process parameters and basic tea parameters, extracts thermodynamic characteristics and Maillard reaction characteristic spectrum, and constructs a variety adaptability model; multimodally fuses the thermodynamic characteristics, Maillard reaction characteristic spectrum and variety adaptability parameters to generate a carbon roasting quality prediction vector; performs preliminary grade determination through a hierarchical decision tree model, and performs cascade supplementary fire optimization on secondary tea leaves before re-determination; and finally associates the grading results with the carbon roasting parameters and stores them in the blockchain. This method realizes the dynamic correlation analysis between carbon roasting process parameters and tea quality, solves the problem of inaccurate evaluation caused by the separation of traditional grading methods and carbon roasting processes, significantly improves the scientific nature and reliability of the grading results, and provides data support for process optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 is a method step diagram of steps S101 to S107 of the method described in the specific embodiment; Figure 2 is a method step diagram of steps S201 to S207 of the method described in the specific embodiment; Figure 3 It is a structural diagram of the electronic device described in the specific implementation method.

[0017] The reference numerals in the above drawings are described as follows: 1. Electronic equipment; 11. Memory; 12. Processor. DETAILED DESCRIPTION

[0018] The present invention will be described in further detail below with reference to the accompanying drawings and examples. It is particularly noted that the following examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Similarly, the following examples are only some embodiments of the present invention and are not intended to be exhaustive. All other embodiments obtained by those of ordinary skill in the art without creative effort are intended to fall within the scope of protection of the present invention.

[0019] See also Figure 1 In a first aspect, this embodiment provides a semi-refined tea grading method based on carbon roasting parameters, comprising: S101, collecting carbon roasting process parameters and basic tea parameters, wherein the carbon roasting process parameters include a three-dimensional temperature field distribution curve and duration of the roasting cage, and the basic tea parameters include variety information, moisture content information, and primary processing grade; S102, extracting features of carbon baking process parameters to obtain thermodynamic characteristics, including a carbon baking uniformity index and a key temperature duration, wherein the carbon baking uniformity index is obtained by calculating a spatial temperature standard deviation using infrared thermal imaging information; Furthermore, hyperspectral imaging information of tea leaves in the carbon roasting space is collected, a Maillard reaction characteristic spectrum within a preset wavelength range of 900-2500 nm is extracted, and tea spectral parameters are calculated based on the Maillard reaction characteristic spectrum, wherein the tea spectral parameters include the aldehyde-ketone ratio change rate and the melanoidin formation inflection point temperature; S103, dynamically constructing a variety adaptability model based on tea spectral parameters and tea basic parameters to obtain variety adaptability parameters, the variety adaptability parameters including a leaf edge burn compensation coefficient, an aroma retention rate threshold, and a water gradient parameter; S104. Performing multimodal fusion on the thermodynamic characteristics, Maillard reaction characteristic spectrum, and variety adaptability parameters to generate a carbon roasting quality prediction vector. The multimodal fusion includes temperature-substance conversion collaborative splicing and spatial inhomogeneity compensation. S105: Inputting the carbon roasting quality prediction vector into a hierarchical decision tree model to obtain a preliminary grade determination result, which includes inferior, superior, and special grade. The hierarchical decision tree model is configured to trigger a cascade supplementary firing optimization process based on the clustering quality index; S106. Cascaded re-fire optimization is performed on tea leaves that are initially rated as inferior to obtain optimized tea leaves. The optimized tea leaves are then re-graded to obtain an updated grade determination result. The optimization strategy of the cascaded re-fire optimization is configured to combine and adjust the re-fire duration and temperature using a material conversion model. S107. Generate the final grading result of the current tea based on the preliminary grading result and the updated grading result. Associate the final grading result with the carbon roasting process parameters and store them in the blockchain. Generate a grading report that includes the grade distribution probability, process defect location, and energy consumption assessment.

[0020] In step S101, carbon roasting process parameters are collected using a distributed temperature sensor array. The three-dimensional temperature distribution curve of the roasting cage represents the dynamic changes in the spatial thermal field during carbon roasting, and the duration records the duration of each temperature zone. Among the basic tea parameters, variety information can be obtained through a variety identification system, and moisture content can be measured using a near-infrared moisture meter. Preferably, the primary processing grade is determined based on a combination of sensory evaluation (manual evaluation) and instrumental testing (e.g., moisture content, color, etc.) after primary processing, encompassing both subjective quality evaluation and objective quantitative indicators. These parameters collectively constitute the fundamental data source for tea quality evaluation.

[0021] In step S102, thermodynamic signatures are extracted using an infrared thermal imaging system. The carbon roasting uniformity index is calculated by calculating the standard deviation of the temperature in each region of the roasting cage, reflecting the stability of the thermal field distribution. The duration of critical temperature is calculated by calculating the cumulative duration of exposure within a specific temperature range. A hyperspectral imaging system captures the Maillard reaction spectrum in the 900-2500nm band. The rate of change in the aldehyde-ketone ratio indicates the progress of the glucosamine reaction, while the inflection point temperature for melanoidin formation indicates the critical state of the browning reaction. These characteristic parameters collectively reflect the physical and chemical changes in tea leaves during the carbon roasting process.

[0022] In step S103, a variety adaptability model is constructed using a machine learning algorithm. The leaf edge burn compensation coefficient is calculated based on the variety's cuticle thickness, the aroma retention threshold is set based on the variety's volatile components, and the water gradient parameter reflects the differences in water loss rates among different varieties. The model dynamically adjusts these parameters to adapt to the carbon roasting characteristics of different tea varieties.

[0023] In step S104, multimodal fusion preferably employs a feature-level fusion strategy. Temperature-substance transformation collaborative stitching aligns thermodynamic parameters with spectral parameters in a time series, thereby establishing a precise correspondence between substance transformation and temperature changes during processing. Spatial inhomogeneity compensation corrects parameter differences between the edge and center of the roasting cage using a weighted algorithm, thereby eliminating the impact of temperature distribution differences within the roasting cage on tea quality and ensuring overall processing uniformity. This process generates a carbon roasting quality prediction vector containing multidimensional features.

[0024] In step S105, a hierarchical decision tree model uses a pre-trained classifier. Preferably, a clustering quality index is calculated by the ratio of intra-class distance to inter-class distance. When this index falls below a set threshold, the cascaded re-firing optimization process is triggered. The model outputs preliminary judgment results for three categories: inferior, excellent, and exceptional.

[0025] In step S106, the cascaded tea-replenishing optimization is preferably implemented by a material transformation model, which establishes the temperature-time-material transformation relationship based on the Arrhenius equation and generates an optimized parameter combination for the secondary tea. The tea leaves after the tea-replenishing need to go through the complete grading and evaluation process again.

[0026] In step S107, blockchain storage can utilize distributed ledger technology. Grade distribution probabilities are derived from historical data statistics, process defect location is based on parameter deviation analysis, and energy consumption assessment is based on a combination of temperature field data and duration calculations. These grade distribution probabilities, process defect location, and energy consumption assessment together form a traceable, graded quality report.

[0027] This embodiment realizes closed-loop control of carbon roasting process and quality grading through three core links: multi-source data collection, feature fusion and dynamic optimization. Among them, thermodynamic characteristics reflect the state of the processing environment, spectral parameters characterize the degree of material transformation, and the variety adaptability model ensures the personalization of evaluation standards, and finally forms a scientific and objective grading decision-making system. The cascaded fire replenishment mechanism provides a secondary optimization opportunity for quality improvement, and blockchain technology ensures the authenticity and traceability of the data. This embodiment organically integrates the three-dimensional temperature field distribution, the critical state of material transformation and the variety characteristic parameters to form a closed-loop control system, which not only ensures the uniformity and controllability of the carbon roasting process, but also ensures the traceability of the grading results through blockchain storage, significantly improving the scientificity and reliability of the grading of semi-refined tea.

[0028] In some embodiments, the carbon baking process parameters include three-dimensional temperature field information and duration of the baking cage; The carbon baking process parameters are extracted to obtain thermodynamic characteristics. The thermodynamic characteristics include the carbon baking uniformity index and the duration of key temperatures, including: The ratio of the standard deviation to the mean of the temperature values ​​corresponding to each temperature measurement point is calculated based on the three-dimensional temperature field information of the baking cage to generate the carbon baking uniformity index; Furthermore, a temperature-time curve is constructed based on the three-dimensional temperature field information and duration of the roasting cage, and temperature platform intervals whose duration exceeds a preset time threshold are extracted from the curve. Each temperature platform interval corresponds to a temperature fluctuation range. The temperature platform interval whose temperature fluctuation range meets the preset fluctuation deviation range is recorded as a key platform, and the duration corresponding to the key platform is recorded as the key temperature duration. Furthermore, temperature gradient distribution information is generated based on the temperature values ​​of adjacent temperature measurement points, and abnormal temperature points in the temperature gradient distribution information that exceed a preset average temperature range are corrected by interpolation compensation based on a spatially weighted neighborhood to obtain corrected temperature field information; Thermodynamic characteristics are generated based on the carbon baking uniformity index, key temperature duration and corrected temperature field information.

[0029] In this embodiment, the three-dimensional temperature field information of the roasting cage refers to a collection of data reflecting the spatial temperature distribution within the roasting cage, collected in real time by a distributed array of temperature sensors. This data is used to reconstruct the three-dimensional temperature field using a grid of temperature measurement points. Duration represents the complete process cycle from the start to the end of carbon roasting. This duration, combined with the temporal sequence of the temperature field information, forms the basis for constructing a temperature-time curve.

[0030] The carbon baking uniformity index is achieved by calculating the ratio of the standard deviation of the temperature values ​​at each measuring point to the mean. This index is used to quantify the uniformity of the temperature distribution inside the baking cage. The lower the ratio, the better the spatial consistency of the temperature field.

[0031] The temperature-time curve is a continuous curve formed by associating time-series temperature data with duration coordinates, in which the preset time threshold is pre-set according to the optimal carbon roasting process requirements for the tea category and is used to screen temperature platforms with process significance. The temperature platform interval refers to the continuous period in the curve where the temperature fluctuation remains within a specific range. Its identification is achieved by calculating the slope change between adjacent sampling points. The key platform refers to the platform interval that meets the duration threshold requirements and the temperature fluctuation range is controlled within the preset deviation range. The corresponding duration reflects the stability of the core process stage.

[0032] The temperature gradient distribution information is vector field data generated by analyzing the temperature difference between adjacent temperature measurement points, which is used to characterize the spatial trend of temperature changes. The preset average temperature range is set according to historical process data, and the determination of abnormal temperature points is achieved by comparing the statistical characteristics of the temperature of the point with that of the surrounding neighborhood. The interpolation compensation correction of the spatial weighted neighborhood is to calculate the replacement value based on the spatial distribution weight of the effective temperature measurement points around the abnormal point, and its weight distribution is inversely proportional to the spatial distance between the temperature measurement points. The corrected temperature field information is an optimized temperature distribution model obtained by eliminating the interference of abnormal temperature measurement data. Together with the carbon baking uniformity index and the duration of key temperature, it constitutes a thermodynamic characteristic system that reflects the quality of the carbon baking process.

[0033] This embodiment uses multi-dimensional feature extraction to convert raw temperature data into thermodynamic features that can be quantified to evaluate process quality. The three dimensions of spatial uniformity, temporal stability, and data reliability complement each other, providing data support for standardized control of the carbon baking process.

[0034] See also Figure 2 In some embodiments, hyperspectral imaging information of tea leaves in a carbon roasting space is collected, a Maillard reaction characteristic spectrum within a preset wavelength range is extracted, and tea spectral parameters are calculated based on the Maillard reaction characteristic spectrum, including: S201, collecting reflectance spectrum data of the tea surface through a hyperspectral imaging system, and selecting at least two characteristic bands within a preset band range, wherein the characteristic bands include a first reaction characteristic band and a second reaction characteristic band; S202, performing absorbance conversion on the reflectance spectrum data of the first reaction characteristic band and the second reaction characteristic band to generate a time-absorbance change curve of the corresponding bands; S203, smoothing the time-absorbance change curve to eliminate noise interference and extracting characteristic points on the curve, the characteristic points including the absorbance maximum, inflection point and platform interval; S204, calculating the integral ratio of the absorbance of the first reaction characteristic band to the absorbance of the second reaction characteristic band based on the position and change trend of the characteristic points to generate a reaction progress index, wherein the integral ratio adopts a trapezoidal numerical integration method; S205, monitoring the rate of change of the reaction progress indicator over time, and marking it as a key turning point of the Maillard reaction when the rate of change of the reaction progress indicator over time exceeds a set threshold; S206. Calculating the cumulative amount of melanoidins and the dynamic change trend information of the aldehyde-ketone ratio based on the temperature and duration corresponding to the key turning point of the Maillard reaction, the cumulative amount of melanoidins is obtained by accumulating the absorbance changes after the key turning point of the Maillard reaction, and the dynamic change trend information of the aldehyde-ketone ratio is obtained by establishing a time series ratio curve; S207. Generate tea spectrum parameters according to the melanoidin accumulation amount, the aldehyde-ketone ratio dynamic change trend information, and the reaction progress index.

[0035] In step S201, the preset wavelength range of the hyperspectral imaging system is determined by the Maillard reaction characteristics. Reflectance spectral data refers to optical information on the tea leaf surface collected by the hyperspectral imaging system. The first and second characteristic wavelengths correspond to characteristic absorption peaks of products at different stages of the Maillard reaction. These characteristic wavelengths are selected based on the spectral properties of the characteristic compounds produced during the Maillard reaction.

[0036] In step S202, absorbance conversion is a mathematical process of converting reflectance data into absorbance. The time-absorbance change curve is used to characterize the change pattern of absorbance in a specific band over time, which can intuitively reflect the reaction process.

[0037] In step S203, smoothing is performed using a digital filtering algorithm to eliminate spectral noise and ensure data reliability. The absorbance maximum represents the peak moment of the reaction rate, the inflection point reflects the transition characteristics of the reaction mechanism, and the plateau indicates the stage when the reaction tends to stabilize. Accurately extracting these characteristic points is crucial for subsequent analysis.

[0038] In step S204, the integral ratio refers to the calculation of the ratio of the areas under the absorbance curves of two characteristic bands, and the trapezoidal numerical integration method is used to improve the calculation accuracy.

[0039] In step S205 , the threshold is set to be a critical judgment value determined according to the kinetic characteristics of the Maillard reaction, and is used to accurately identify the key turning point of the Maillard reaction.

[0040] In step S206, the cumulative amount of melanoidins refers to the total amount of reaction products obtained by accumulating the absorbance changes, and the dynamic change trend information of the aldehyde-ketone ratio refers to the reaction characteristics reflected by establishing a time series ratio curve.

[0041] In step S207 , the tea spectral parameters refer to a combination of quantitative indicators that comprehensively reflect the progress of the Maillard reaction, providing data support for subsequent analysis.

[0042] This embodiment uses hyperspectral imaging technology to dynamically monitor the Maillard reaction during the carbon roasting process of tea leaves, and uses the spectral characteristic changes of characteristic bands to characterize the reaction process. By selecting the first reaction characteristic band and the second reaction characteristic band, establishing a time-absorbance change curve, and extracting key characteristic points such as the absorbance maximum, inflection point, and platform interval, accurate tracking of the Maillard reaction process is achieved. The reaction process index is calculated using the integral ratio, and the key turning points of the reaction are accurately identified in combination with the set threshold, and then the cumulative amount of melanoidins and the dynamic change trend of the aldehyde-ketone ratio are calculated. This solution uses non-contact optical detection means to achieve real-time and quantitative monitoring of the Maillard reaction during carbon roasting, providing objective and reliable spectral parameter indicators for tea carbon roasting process control, effectively solving the subjective problem of traditional manual experience judgment, and significantly improving the scientificity and accuracy of carbon roasting process control. This embodiment uses spectral analysis of characteristic bands to achieve dynamic monitoring and quantitative characterization of the Maillard reaction during tea carbon roasting, providing reliable optical detection means for carbon roasting process control.

[0043] In some embodiments, a variety adaptability model is dynamically constructed based on tea spectral parameters and tea basic parameters to obtain variety adaptability parameters, including: Match the variety information with the preset variety feature library to obtain the reference substance conversion curve corresponding to the current variety information. The reference substance conversion curve includes the expected range of variation of the aldehyde-ketone ratio and the threshold value of the melanoidin accumulation amount; According to the dynamic change trend information of the aldehyde-ketone ratio in the tea spectral parameters, the dynamic time warping distance between the real-time aldehyde-ketone ratio curve and the reference substance conversion curve is calculated to generate a first matching index; and, calculating the relative deviation percentage between the current melanoidin accumulation amount and the expected value of the reference substance conversion curve based on the melanoidin accumulation amount in the tea spectral parameters, to generate a second matching index; The first matching index and the second matching index are weighted and fused to obtain a comprehensive matching index. The first matching weight corresponding to the dynamic change trend information of the aldehyde-ketone ratio increases with the increase of carbon baking temperature, and the second matching weight corresponding to the cumulative amount of melanoidin increases with the extension of the duration. Compare the comprehensive matching index with the preset index threshold range, generate temperature adjustment information according to the comparison result, generate a temperature acceleration instruction when the comprehensive matching index is lower than the lower limit threshold of the preset index threshold range, and generate a constant temperature maintenance instruction when the comprehensive matching index is higher than the upper limit threshold of the preset index threshold range; Furthermore, carbon baking amplitude change information is generated according to the moisture content information, and the adjustment amplitude is increased when the moisture content is high, and the adjustment amplitude is reduced when the moisture content is low; Generate variety adaptability parameters based on temperature adjustment information and carbon roasting amplitude change information.

[0044] In this embodiment, the variety adaptability model is a dynamic parameter adjustment system established based on the characteristics of tea varieties. It achieves process optimization by matching real-time monitoring data with benchmark parameters. The preset variety characteristic library contains the material transformation characteristic data of different tea varieties under standard carbon roasting conditions. Among them, the benchmark material conversion curve represents the aldehyde-ketone ratio change trajectory and melanoidin accumulation trend of a specific variety under ideal process conditions. The expected range of aldehyde-ketone ratio change is obtained through historical process data statistics, and the melanoidin accumulation threshold is pre-set according to the quality requirements of the variety.

[0045] The real-time aldehyde-ketone ratio curve is the actual reaction progress data obtained through hyperspectral monitoring. The dynamic time warping distance is used to quantify the morphological difference between the actual curve and the baseline curve. This distance calculation is achieved through the dynamic time warping algorithm, which can eliminate the influence of nonlinear deformation on the time axis. The first matching index reflects the process compliance of the change in the aldehyde-ketone ratio. The smaller the value, the closer the actual reaction process is to the ideal state. The current melanoidin accumulation refers to the real-time measurement value obtained by spectral analysis. The relative deviation percentage is obtained by calculating the difference between the measured value and the baseline expected value. The generated second matching index is used to evaluate whether the melanoidin production progress meets the standard.

[0046] Weighted fusion is a process of dynamically weighted summation of two matching indicators according to the characteristics of the process stages. Among them, the first matching weight is positively correlated with the carbon baking temperature, reflecting the dominance of the aldehyde and ketone reaction in the high-temperature stage; the second matching weight is positively correlated with the duration, reflecting the time lag characteristics of the cumulative effect of melanoidins.

[0047] Preferably, the comprehensive matching index is obtained through linear weighted calculation, and the preset index threshold range is determined according to the variety characteristics and quality requirements. The lower limit threshold triggers the temperature acceleration instruction to promote the reaction rate, and the upper limit threshold triggers the constant temperature maintenance instruction to stabilize the reaction state.

[0048] Roasting amplitude variation information refers to the process parameter correction dynamically adjusted based on the tea's initial moisture content. This adjustment logic follows the negative correlation between moisture content and heat transfer efficiency, and quantitative control of the adjustment amplitude is achieved through a proportional coefficient. Temperature adjustment information and roasting amplitude variation information together constitute the variety adaptability parameter, which is used through a feedback control system to achieve real-time dynamic adjustment of the roasting process.

[0049] This example establishes a dual-matching evaluation mechanism, dynamically comparing tea spectral parameters with reference substance conversion curves corresponding to the current variety. Combined with a moisture content compensation mechanism, this approach enables customized control of the carbon roasting process for each variety. The dynamic time warping algorithm addresses the temporal nonlinearity of the reaction process, while the dual-weighting mechanism ensures a focused assessment of different process stages. The resulting variety adaptability parameters provide a precise basis for controlling the carbon roasting process.

[0050] In some embodiments, thermodynamic characteristics, Maillard reaction characteristic spectra, and variety adaptability parameters are multimodally fused to generate a carbon roasting quality prediction vector. The multimodal fusion includes temperature-substance conversion collaborative splicing and spatial heterogeneity compensation, including: The temperature field compensation coefficient is generated according to the carbon baking uniformity index, and the dynamic change trend information of the aldehyde-ketone ratio in the Maillard reaction characteristic spectrum is regionally corrected and recorded as the correction trend information; The temperature adjustment information in the variety adaptability parameters is coupled with the duration of the key temperature to obtain a time weighting factor. The time weighting factor is used as the weight value of the Maillard reaction characteristic spectrum at different carbon roasting stages and weighted to obtain a weighted reaction characteristic spectrum. Furthermore, based on the moisture content information, the temperature field compensation coefficient and the time weighting factor are dynamically scaled to generate a moisture content adaptation adjustment coefficient; A spatial grid mapping relationship is constructed, and the correction trend information, weighted response characteristic spectrum, and moisture content adaptation adjustment coefficient are integrated according to spatial position. In addition, abnormal grid points in the spatial grid mapping relationship are smoothed in the neighborhood. The smoothing process retains the temperature gradient characteristics while eliminating isolated noise points to obtain spatial grid information. The spatial grid information is expanded along the time dimension to generate a carbon roasting quality prediction vector that includes temperature-material transformation synergistic characteristics and spatial distribution characteristics.

[0051] In the present embodiment, the temperature field compensation coefficient is a correction factor obtained by converting the carbon baking uniformity index, and its value is negatively correlated with the uniformity index. It is used to perform regional specific correction on the dynamic change trend information of the aldehyde-ketone ratio, thereby eliminating the influence of uneven temperature distribution on the reaction process. That is, the lower the carbon baking uniformity index, the more uneven the temperature distribution, and the greater the temperature field compensation coefficient, thereby enhancing the regional correction intensity of the dynamic change trend of the aldehyde-ketone ratio. The generated correction trend information characterizes the reaction process data after spatial compensation.

[0052] The time weighting factor is a dynamic weight value obtained by nonlinearly coupling the temperature adjustment information with the duration of the critical temperature. The calculation process takes into account the synergistic effect of the temperature adjustment amplitude and the action time. Furthermore, the calculation process of the time weighting factor includes: Normalize the temperature adjustment information (temperature acceleration or constant temperature maintenance instructions) and convert it into an adjustment intensity coefficient; The regulation intensity coefficient is multiplied by the duration of the critical temperature to reflect the cumulative effect of temperature regulation time; The S-type function is used to map the weight value to the interval [0,1] to ensure that the weight value conforms to the probability distribution characteristics, and the high temperature and long time stage will obtain a higher weight.

[0053] The time weighting factor serves as the weighting coefficient of the Maillard reaction characteristic spectrum, which enables the reaction characteristics of different carbon roasting stages to be differentiated. The final weighted reaction characteristic spectrum reflects the material transformation state after process control.

[0054] The moisture content adaptation adjustment coefficient is a scaling factor that dynamically adjusts the temperature field compensation coefficient and the time weighting factor through moisture content information. The adjustment direction is positively correlated with the moisture content, ensuring that the process parameters adapt to the differences in the initial state of the tea leaves. Preferably, the dynamic adjustment process of the moisture content adaptation adjustment coefficient is as follows: when the initial moisture content of the tea leaves is high, the temperature field compensation coefficient (strengthening the temperature correction strength) and the time weighting factor (extending the weight of the key temperature effect) are proportionally amplified to compensate for the heat conduction lag effect caused by high moisture; otherwise, the coefficient is appropriately reduced. The adjustment range is based on a preset moisture content-adjustment coefficient mapping table, which is established through historical process data to ensure that tea leaves with different moisture contents can obtain matching thermodynamic parameter configurations.

[0055] The spatial grid mapping relationship is a three-dimensional spatial coordinate system established, and its grid division is consistent with the distribution of measurement points of the roasting cage temperature field. Preferably, when the correction trend information, weighted reaction characteristic spectrum and moisture content adaptation adjustment coefficient are integrated according to spatial position, a grid interpolation algorithm is used to ensure data spatial alignment. Furthermore, the specific process of the grid interpolation algorithm is as follows: The carbon baking area is divided into uniform two-dimensional grid cells, and each grid node stores parameters such as correction trend, weighted characteristic spectrum and moisture content coefficient; The bicubic spline interpolation method is used to calculate the integrated value of any position inside the grid based on the parameter values ​​of adjacent nodes, ensuring smooth transition of different parameters in space. Gradient detection is used to eliminate edge mutations that may be caused by interpolation, so that the temperature compensation parameters are continuously distributed in the three-dimensional space including the plane position and process time.

[0056] The identification of abnormal grid points is based on local statistical characteristics. Preferably, it is achieved by calculating the standard deviation and mean deviation of each grid point from its 8 neighbors. When the data of a point deviates from the neighborhood mean by more than 3 times the standard deviation, it is marked as abnormal.

[0057] Preferably, the neighborhood smoothing process adopts an anisotropic filtering algorithm. During the smoothing process, the anisotropic filtering enhances the filtering intensity along the temperature gradient direction (suppresses outliers), while maintaining isotropic smoothness in the uniform area. The filter kernel size is dynamically adjusted with the local variation coefficient, and isolated noise points are eliminated while retaining the temperature gradient characteristics. The generated spatial grid information contains complete spatial distribution characteristics.

[0058] The carbon roast quality prediction vector is a high-dimensional feature vector generated by expanding spatial grid information along the time dimension. Its temperature-substance transformation synergy reflects the relationship between thermodynamic parameters and reaction progress, and its spatial distribution reflects the three-dimensional distribution of process parameters. The carbon roast quality prediction vector is generated through tensor expansion operations, resulting in a multidimensional data structure that provides comprehensive characterization for quality prediction.

[0059] Preferably, the carbon-roast quality prediction vector is generated using a high-order tensor expansion. This involves expanding the three-dimensional tensor (spatial position, time, and parameter type) composed of multiple process parameters, such as temperature field, moisture content, and time series, into multiple two-dimensional matrices by modality. The principal component features of each modality are extracted using alternating least squares, and ultimately fused into a prediction vector that incorporates spatial distribution patterns, time-varying characteristics, and parameter coupling relationships. This prediction vector is decoded using a convolutional neural network, outputting the probability distribution of quality indicators such as moisture uniformity and caramelization degree.

[0060] This embodiment achieves deep coupling of thermodynamic environmental parameters, material transformation characteristics, and variety adaptability by establishing a multi-source data fusion framework. The temperature field compensation mechanism solves the problem of spatial heterogeneity, the dynamic weighting strategy reflects the differences in process stages, the adaptive adjustment of moisture content ensures the generalization capability of parameters, and the spatial grid processing ensures the spatial consistency of features. The resulting carbon roasting quality prediction vector combines the three-dimensional characteristics of time, space, and material transformation, providing high-precision feature input for subsequent quality grading. The fusion process of this embodiment strictly follows the laws of physical and chemical changes, and achieves quantitative characterization of the carbon roasting process through a data-driven approach.

[0061] In some embodiments, the carbon roasting quality prediction vector is input into a hierarchical decision tree model to obtain a preliminary grade determination result. The preliminary grade determination result includes inferior grade, superior grade, and special grade, including: A hierarchical decision tree model was established, which included a temperature-substance transformation synergistic feature branch, a spatial distribution feature branch, and a Maillard reaction integrity branch. The temperature-material conversion synergistic feature value in the carbon roasting quality prediction vector is input into the temperature-material conversion synergistic feature branch, and when the temperature-material conversion synergistic feature value is lower than the first feature threshold, it is judged as secondary; Input the spatial distribution characteristics in the carbon roasting quality prediction vector into the spatial distribution feature branch, calculate the quality uniformity index, and judge it as excellent when the quality uniformity index is lower than the second feature threshold; The Maillard reaction features in the carbon-roasted quality prediction vector are input into the Maillard reaction integrity branch. When the product of the slope of the aldehyde-ketone ratio curve and the cumulative amount of melanoidins exceeds the third characteristic threshold, it is judged as special grade. According to the variety information in the basic parameters of tea, the corresponding first characteristic threshold, second characteristic threshold and third characteristic threshold are extracted from the preset grading standard library; Performing time decay compensation on the first characteristic threshold, the second characteristic threshold, and the third characteristic threshold according to the actual carbon baking time in the carbon baking process parameters; Outputting preliminary grade determination results including inferior, superior, and special grades, and the preliminary grade determination results are accompanied by feature threshold comparison data of each branch; The hierarchical decision tree model is configured to trigger a cascading re-fire optimization process based on the clustering quality index, including: Calculate the clustering quality index of the preliminary grade judgment results. The clustering quality index is determined by the variance of the feature space distance between the special grade samples and the excellent grade samples. When the clustering quality index is lower than the preset clustering threshold, the cascaded supplementary fire optimization process is activated.

[0062] In this embodiment, the hierarchical decision tree model is a tea quality grading algorithm framework based on multi-branch conditional judgment. Its temperature-substance transformation synergistic feature branch is used to evaluate the matching degree between thermodynamic parameters and substance transformation, the spatial distribution feature branch is used to detect the spatial uniformity of quality indicators, and the Maillard reaction integrity branch is used to quantify the completion degree of key reactions.

[0063] The temperature-material transformation synergistic characteristic value is a scalar parameter extracted from the carbon roasting quality prediction vector. Its value reflects the coupling strength between the temperature field and the material transformation process. The first characteristic threshold is the classification boundary value determined based on the statistics of historical high-quality tea samples.

[0064] The quality uniformity index is a comprehensive evaluation parameter calculated through spatial distribution characteristics. It is quantified by analyzing the spatial variation coefficient of parameters such as moisture content and degree of caramelization. The second characteristic threshold represents the lower limit of allowable quality fluctuation.

[0065] The slope of the aldehyde-ketone ratio curve in the Maillard reaction integrity branch refers to the instantaneous rate of change in the concentration of aldehydes and ketones during the reaction process. The cumulative amount of melanoidins is an indicator of pigment production extracted through near-infrared spectral characteristics. The product of the two reflects the reaction kinetics. The third characteristic threshold corresponds to the reaction completion benchmark required for premium tea.

[0066] The above-mentioned hierarchical decision tree model adopts a three-level progressive judgment structure to achieve strict quality grading. When the temperature-substance conversion synergistic characteristic value is lower than the first characteristic threshold, the model directly judges it as secondary and terminates the subsequent branch evaluation, which indicates that there is a serious mismatch between the thermodynamic environment and the material conversion process. Only when the temperature-substance conversion synergistic characteristic value meets the standard will the model enter the spatial distribution characteristic branch evaluation, and judge whether it meets the excellent grade standard through the quality uniformity index; if the index is lower than the second characteristic threshold, it will terminate at the excellent grade judgment, reflecting that although the thermodynamic conditions are qualified, there is a spatial heterogeneity problem. Finally, only samples that pass the first two levels of judgment will enter the Maillard reaction integrity branch, and the special grade qualification will be determined by the product of the slope of the aldehyde-ketone ratio curve and the cumulative amount of melanoidins. This mechanism ensures that special grade tea must meet the three core requirements of thermodynamic matching, spatial uniformity and reaction integrity at the same time. Through the cascade judgment logic and the establishment of a strict access mechanism, accurate distinction of quality grades is achieved.

[0067] The preset grading standard library refers to a database that stores the grading parameters corresponding to different tea varieties, and its threshold data can be obtained through variety-specific process test calibration.

[0068] Time attenuation compensation refers to the dynamic adjustment of the characteristic threshold according to the actual carbon baking time. The compensation coefficient is negatively correlated with the carbon baking time and is used to eliminate the influence of characteristic signal attenuation caused by long-term carbon baking.

[0069] The feature threshold comparison data contains the deviation percentage between the actual feature value of each branch and the threshold, which is used to assist the interpretability analysis of grade determination.

[0070] The clustering quality index (CQI) is a measure of model discriminative effectiveness calculated through the dispersion of feature spatial distribution. It is quantified by calculating the Mahalanobis distance variance between premium and superior grade samples in the three-dimensional feature space of temperature-material transformation synergy, spatial distribution characteristics, and Maillard reaction characteristics. The preset clustering threshold is a stability boundary value determined based on performance verification during the model training phase. A value below the preset clustering threshold indicates fuzzy boundaries in the classification results. The cascaded re-firing optimization process is a process adjustment mechanism triggered for batches with insufficient clustering quality, re-optimizing carbon roasting parameters through feedback control.

[0071] This embodiment achieves refined grading through a three-level progressive judgment structure, with temperature-substance transformation synergy serving as the foundational screening criteria, spatial uniformity as the basis for intermediate quality determination, and Maillard reaction integrity as the determinant of top-level quality. The grading process incorporates a variety-adaptive dynamic threshold mechanism and time-decay compensation to ensure that the judgment criteria align with process realities. A clustered quality monitoring mechanism ensures the reliability of grading results and provides a data basis for subsequent process optimization.

[0072] In some embodiments, the optimization strategy for cascade supplementary fire optimization is configured to generate a combination of supplementary fire duration and temperature using a material conversion model, including: A material transformation kinetics model is established. The input features of the material transformation kinetics model include the deviation value of the temperature-material transformation synergistic feature and the spatial distribution feature heterogeneity in the carbon roasting quality prediction vector. The optimal supplementary fire parameter combination is calculated based on the material conversion kinetics model, and the optimization strategy of cascade supplementary fire optimization is generated based on the optimal supplementary fire parameter combination. The optimal supplementary fire parameter combination calculated based on the material conversion kinetics model includes: For areas where the temperature-matter conversion synergistic characteristic value is lower than the threshold, the amount of supplementary fire temperature increase and the amount of action time extension are generated; For areas with uneven spatial distribution characteristics, calculate the turning frequency adjustment coefficient and heat energy redistribution plan; For samples with incomplete Maillard reaction, determine the optimal heating rate curve during the re-firing stage; Cascaded supplementary fire optimization is performed on tea leaves that are initially graded as inferior to obtain optimized tea leaves, and the optimized tea leaves are re-graded to obtain updated grade determination results, including: Re-collect and optimize the carbon roasting process parameters and hyperspectral imaging information of tea leaves, record them as optimized carbon roasting parameters and optimized imaging information, and update the carbon roasting quality prediction vector, record them as optimized quality prediction vector; The optimized quality prediction vector is input into the hierarchical decision tree model for secondary judgment to generate an updated grade judgment result; When the update level determination result is still secondary and the number of supplementary fire times has not reached the upper limit, iteratively executing the cascade supplementary fire optimization process; The parameter adjustment records of each re-firing operation and the corresponding quality change data are recorded, and the parameter weights of the material conversion kinetics model are dynamically updated.

[0073] In this embodiment, the optimization strategy for cascaded supplementary fire optimization is generated using a material transformation kinetics model, which uses the temperature-material transformation synergistic characteristic deviation and spatial distribution characteristic nonuniformity as core input features. The temperature-material transformation synergistic characteristic deviation refers to the difference between the actual characteristic value and the standard characteristic value, while the spatial distribution characteristic nonuniformity is calculated as the percentage deviation between the quality uniformity index and the standard value.

[0074] The optimal combination of supplementary firing parameters is a process adjustment plan calculated for different types of quality defects. Its generation process adopts the principle of differentiated processing: for areas where the temperature-material transformation synergistic characteristic value is lower than the threshold, the material transformation is enhanced by increasing the supplementary firing temperature and extending the action time; for areas with uneven spatial distribution, the material movement trajectory is changed by adjusting the roasting frequency coefficient, and the temperature field distribution is optimized in conjunction with the heat energy redistribution plan; for samples with incomplete Maillard reaction, the reaction kinetics process is controlled by the optimal heating rate curve.

[0075] Optimized tea refers to the improved product after undergoing the cascaded re-firing optimization process. Its quality is assessed using newly collected optimized carbon roasting parameters and hyperspectral imaging data. The optimized carbon roasting parameters include newly added temperature-time curves and roasting operation records from the re-firing phase. The optimized imaging data is used to extract updated material distribution characteristics. The optimized quality prediction vector is a secondary evaluation feature set constructed by integrating the newly collected data. This feature set, when input into a hierarchical decision tree model, generates an updated grade judgment result, which is used to verify the optimization effect.

[0076] The upper limit for re-firing is a safety threshold set based on the tea's heat-resistance characteristics. Parameter adjustment records include temperature corrections, duration adjustments, and changes to the roasting schedule for each re-firing operation. Quality change data is obtained by comparing the differences in eigenvalues ​​before and after optimization. This data is used to dynamically update the parameter weights in the material transformation kinetics model. This update mechanism achieves adaptive optimization of the model through a machine learning algorithm.

[0077] This embodiment establishes a mapping relationship between defect characteristics and process parameters to accurately generate targeted re-firing strategies. It employs an iterative optimization mechanism to ensure quality improvement, and dynamically updates model parameters to maintain the adaptability of the optimization strategy. This transforms the traditional empirical re-firing operation into a quantitative control process based on the laws of material transformation, providing a scientific basis for refined control of the carbon baking process.

[0078] In some embodiments, a final grading result of the current tea is generated based on the preliminary grading result and the updated grading result. The final grading result is associated with the carbon roasting process parameters and stored in the blockchain. A grading report containing the grade distribution probability, process defect location, and energy consumption assessment is generated, including: The weighted average of the preliminary special-grade ratio, preliminary superior-grade ratio, and preliminary sub-grade ratio in the preliminary grading results and the updated special-grade ratio, updated superior-grade ratio, and updated sub-grade ratio in the updated grading results is calculated to generate the final special-grade ratio, final superior-grade ratio, and final sub-grade ratio in the final grading results; Establish a correlation mapping relationship between the final classification results and the three-dimensional temperature field information of the baking cage and the duration of key temperatures in the carbon baking process parameters. The correlation mapping relationship includes the spatial correspondence between temperature field uniformity and quality grade distribution, and the temporal correspondence between key temperature duration and Maillard reaction integrity. The final grading results, carbon baking process parameters and associated mapping relationships are hashed to generate data blocks, which are then written to the blockchain network for distributed storage. The data blocks contain the grade distribution probability matrix, process defect location coordinates and carbon baking energy consumption statistical indicators. Based on the data blocks, a visual grading report is generated, which includes a three-dimensional temperature field distribution and a grade distribution superimposed thermal map, process defect improvement suggestions, and an energy efficiency optimization analysis report.

[0079] In this embodiment, the final grading result refers to the final evaluation of tea quality generated by integrating the preliminary judgment and the optimized judgment data, among which the preliminary special grade proportion, preliminary superior grade proportion and preliminary secondary grade proportion in the preliminary grade judgment result reflect the quality distribution under the original carbon roasting state, and the updated special grade proportion, updated superior grade proportion and updated secondary grade proportion in the updated grade judgment result represent the quality improvement effect after supplementary fire optimization.

[0080] Correlation mapping establishes a traceable relationship between grading results and carbon roasting process parameters. Specifically, this involves analyzing the spatial uniformity of the three-dimensional temperature field and the regional matching of tea grade distribution, as well as verifying the correlation between temperature continuity during key process periods and the degree of Maillard reaction completion. This correlation mapping provides a basis for analyzing the process-driven causes of grading results.

[0081] The final grading results, carbon baking process parameters and associated mapping relationships are hashed to generate an unalterable data block, which includes the grade distribution probability matrix, process defect location coordinates and carbon baking energy consumption statistical indicators. These data are stored in a distributed manner to ensure the traceability of the grading process.

[0082] In the generation of visual grading reports, preferably, the three-dimensional temperature field distribution and grade distribution superimposed heat map are used to realize bivariate heat map rendering through spatial interpolation technology; the process defect improvement suggestion plan automatically generates an optimization plan based on defect location and parameter association; the energy efficiency optimization analysis report provides energy-saving suggestions by analyzing energy consumption data and quality-output ratio.

[0083] This embodiment has established a verification and traceability system for grading results, ensuring the authenticity of grading data through blockchain technology, and using visual analysis tools to achieve process traceability of grading conclusions. This embodiment generates accurate final grading results by integrating the weighted calculation of preliminary grade determination results and updated grade determination results, and establishes an associated mapping relationship between the final grading results and the three-dimensional temperature field information and key temperature duration of the carbon baking process parameters, thereby achieving traceability analysis of quality grades and process parameters; the grading results, process parameters, and associated mapping relationships are stored through blockchain to ensure that the data cannot be tampered with and is traceable; the visual grading report generated based on the data block, combined with the superimposed heat map of the temperature field and grade distribution, process defect improvement suggestions, and energy efficiency optimization analysis, forms a complete closed loop of quality assessment and process optimization, significantly improving the reliability and guiding value of the grading results.

[0084] In a second aspect, this embodiment further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0085] The computer program involved in this embodiment can be stored in a computer device readable storage medium, which includes but is not limited to a disk, a magnetic tape, a magnetic card, a floppy disk, a flash memory, an optical disc, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can achieve similar or equivalent functions to the storage media listed above, such as DNA, RNA, proteins and other units with information storage capabilities. In a specific embodiment, the storage medium involved can be one of the above-mentioned media types or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiment can be stored in a single medium in a centralized manner or in a distributed manner in multiple media. The memory containing the computer device readable storage medium can be a non-volatile memory or a random access memory. These computer device readable storage media can be built into the device or can be connected to the device involved in the embodiment as an external device or part of an external device. In some embodiments, a memory having a computer-readable storage medium is deployed locally; in other embodiments, a solution of deploying the memory away from the processor may also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form, or can be designed as training data, which can be integrated and reorganized through model training and implicitly stored in the parameter state of a deep neural network or other machine learning model.

[0086] See also Figure 3 In a third aspect, this embodiment further provides an electronic device 1, comprising a memory 11 and a processor 12, wherein the memory 11 is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor 12 to implement the method described in the first aspect.

[0087] The processor described in this embodiment can be implemented by hardware, firmware, software or a combination thereof, and can use a circuit, a single or multiple application-specific integrated circuits (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or at least one of a microprocessor. It also includes other physical, biological or chemical structures that can achieve similar or equivalent functions to the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in each embodiment of the present application, or any combination of the steps mentioned therein.

[0088] Different from the existing technology, the above technical solution has the following beneficial effects: The present invention establishes a complete semi-refined tea grading system by integrating carbon roasting process parameters, tea spectral parameters and variety adaptability parameters. The three-dimensional temperature field distribution curve of the roasting cage is collected by an infrared thermal imaging system, and the Maillard reaction characteristic spectrum is extracted by combining the hyperspectral imaging system. The carbon roasting quality prediction vector is generated by multimodal fusion; the carbon roasting quality prediction vector is analyzed based on the hierarchical decision tree model, and the preliminary grade judgment results including special grade, superior grade and secondary grade are output, and the process of secondary tea is improved through the cascade supplementary fire optimization process; finally, the grading results are associated with the carbon roasting process parameters and stored in the blockchain, and a visual grading report containing a three-dimensional temperature field distribution and a grade distribution superimposed heat map is generated. The above technical solution realizes the precise association between carbon roasting process parameters and tea quality grading, replaces traditional empirical judgment with a data-driven approach, ensures the objectivity and traceability of the grading results, and provides reliable technical support for the quality control of semi-refined tea.

[0089] The above descriptions are only some embodiments of the present invention and do not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A semi-refined tea classification method based on carbon roasting parameters, characterized in that: include: Collecting carbon roasting process parameters and basic tea parameters, including tea variety information, moisture content information, and primary processing grade; Extracting characteristics of the carbon baking process parameters to obtain thermodynamic characteristics, wherein the thermodynamic characteristics include a carbon baking uniformity index and a key temperature duration; Furthermore, hyperspectral imaging information of tea leaves in the carbon roasting space is collected, a Maillard reaction characteristic spectrum within a preset wavelength range is extracted, and spectral parameters of the tea leaves are calculated based on the Maillard reaction characteristic spectrum; Dynamically constructing a variety adaptability model based on the tea spectral parameters and tea basic parameters to obtain variety adaptability parameters; The thermodynamic characteristics, Maillard reaction characteristic spectrum and variety adaptability parameters are subjected to multimodal fusion to generate a carbon roasting quality prediction vector, wherein the multimodal fusion includes temperature-material transformation collaborative splicing and spatial heterogeneity compensation; Inputting the carbon roasting quality prediction vector into a hierarchical decision tree model to obtain a preliminary grade determination result, wherein the preliminary grade determination result includes inferior grade, superior grade, and special grade. The hierarchical decision tree model is configured to trigger a cascade supplementary fire optimization process according to the clustering quality index; Cascaded re-fire optimization is performed on tea leaves that are initially graded as inferior to obtain optimized tea leaves, and the optimized tea leaves are re-graded to obtain an updated grade determination result. The optimization strategy of the cascaded re-fire optimization is configured to combine and adjust the re-fire duration and temperature using a material conversion model; The final grading result of the current tea is generated based on the preliminary grading results and the updated grading results. The final grading result is associated with the carbon roasting process parameters and stored in the blockchain to generate a grading report including grade distribution probability, process defect location and energy consumption assessment.

2. The semi-refined tea classification method based on carbon roasting parameters according to claim 1, characterized in that: The carbon baking process parameters include the three-dimensional temperature field information and duration of the baking cage; The carbon baking process parameters are subjected to feature extraction to obtain thermodynamic characteristics, which include a carbon baking uniformity index and a key temperature duration, including: The ratio of the standard deviation to the mean of the temperature values ​​corresponding to each temperature measurement point is calculated based on the three-dimensional temperature field information of the baking cage to generate the carbon baking uniformity index; Furthermore, a temperature-time curve is constructed based on the three-dimensional temperature field information and the duration of the roasting cage, and temperature platform intervals whose duration exceeds a preset time threshold are extracted from the curve, each of which corresponds to a temperature fluctuation range. The temperature platform interval whose temperature fluctuation range satisfies the preset fluctuation deviation range is recorded as a key platform, and the duration corresponding to the key platform is recorded as the key temperature duration. Furthermore, temperature gradient distribution information is generated based on the temperature values ​​of adjacent temperature measurement points, and abnormal temperature points in the temperature gradient distribution information that exceed a preset average temperature range are corrected by interpolation compensation based on a spatially weighted neighborhood to obtain corrected temperature field information; The thermodynamic characteristics are generated according to the carbon baking uniformity index, the key temperature duration and the corrected temperature field information.

3. The semi-refined tea classification method based on carbon roasting parameters according to claim 1, characterized in that: Collecting hyperspectral imaging information of tea leaves in a carbon roasting space, extracting Maillard reaction characteristic spectrum within a preset band range, and calculating tea spectral parameters based on the Maillard reaction characteristic spectrum, including: Collecting reflectance spectrum data of the tea leaves' surface through a hyperspectral imaging system, and selecting at least two characteristic bands within a preset band range, wherein the characteristic bands include a first reaction characteristic band and a second reaction characteristic band; Performing absorbance conversion on the reflectance spectrum data of the first reaction characteristic band and the second reaction characteristic band to generate time-absorbance change curves of the corresponding bands; Smoothing the time-absorbance change curve to eliminate noise interference, and extracting characteristic points on the curve, wherein the characteristic points include the absorbance maximum, inflection point and platform interval; According to the position and change trend of the characteristic points, the absorbance of the first reaction characteristic band and the absorbance of the second reaction characteristic band are calculated and integrated to generate a reaction progress index, wherein the integral ratio is calculated using a trapezoidal numerical integration method; monitoring the rate of change of the reaction progress indicator over time, and marking it as a key turning point of the Maillard reaction when the rate of change of the reaction progress indicator over time exceeds a set threshold; According to the temperature and duration corresponding to the key turning point of the Maillard reaction, the cumulative amount of melanoidin and the dynamic change trend information of the aldehyde-ketone ratio are calculated, wherein the cumulative amount of melanoidin is obtained by accumulating the change in absorbance after the key turning point of the Maillard reaction, and the dynamic change trend information of the aldehyde-ketone ratio is obtained by establishing a time series ratio curve; The tea spectrum parameters are generated according to the melanoidin accumulation amount, the aldehyde-ketone ratio dynamic change trend information and the reaction progress index.

4. The semi-refined tea classification method based on carbon roasting parameters according to claim 1, characterized in that: A variety adaptability model is dynamically constructed based on the tea spectral parameters and tea basic parameters to obtain variety adaptability parameters, including: Matching the variety information with a preset variety feature library to obtain a reference substance conversion curve corresponding to the current variety information, wherein the reference substance conversion curve includes an expected range of variation of the aldehyde-ketone ratio and a threshold value of melanoidin accumulation; Calculating the dynamic time warping distance between the real-time aldehyde-ketone ratio curve and the reference substance conversion curve based on the dynamic change trend information of the aldehyde-ketone ratio in the tea spectral parameters to generate a first matching index; and, calculating the relative deviation percentage between the current melanoidin accumulation amount and the expected value of the reference substance conversion curve based on the melanoidin accumulation amount in the tea spectral parameters, to generate a second matching index; The first matching index and the second matching index are weightedly fused to obtain a comprehensive matching index, wherein the first matching weight corresponding to the dynamic change trend information of the aldehyde-ketone ratio increases with increasing carbon baking temperature, and the second matching weight corresponding to the cumulative amount of melanoidin increases with increasing duration; Comparing the comprehensive matching index with the preset index threshold range, and generating temperature adjustment information based on the comparison result; and generating carbon baking amplitude change information according to the moisture content information; The variety adaptability parameter is generated according to the temperature adjustment information and the carbon roasting amplitude change information.

5. The semi-refined tea classification method based on carbon roasting parameters according to claim 1, characterized in that: The thermodynamic characteristics, Maillard reaction characteristic spectrum and variety adaptability parameters are subjected to multimodal fusion to generate a carbon roasting quality prediction vector. The multimodal fusion includes temperature-material transformation collaborative splicing and spatial heterogeneity compensation, including: generating a temperature field compensation coefficient according to the carbon baking uniformity index, and performing regional correction on the dynamic change trend information of the aldehyde-ketone ratio in the Maillard reaction characteristic spectrum, which is recorded as correction trend information; The temperature adjustment information in the variety adaptability parameter is coupled with the duration of the key temperature to obtain a time weighting factor, and the time weighting factor is used as the weight value of the Maillard reaction characteristic spectrum at different carbon roasting stages and weighted to obtain a weighted reaction characteristic spectrum; and, dynamically scaling the temperature field compensation coefficient and the time weighting factor according to the moisture content information to generate a moisture content adaptation adjustment coefficient; Constructing a spatial grid mapping relationship, integrating the correction trend information, the weighted response characteristic spectrum, and the moisture content adaptation adjustment coefficient according to spatial position, and performing neighborhood smoothing processing on abnormal grid points in the spatial grid mapping relationship, wherein the smoothing processing retains the temperature gradient characteristics while eliminating isolated noise points to obtain spatial grid information; The spatial grid information is expanded along the time dimension to generate a carbon roasting quality prediction vector including temperature-material conversion synergistic characteristics and spatial distribution characteristics.

6. The method for grading semi-refined tea based on carbon roasting parameters according to claim 1, wherein: The carbon roasting quality prediction vector is input into the hierarchical decision tree model to obtain a preliminary grade determination result. The preliminary grade determination result includes inferior grade, superior grade and special grade, including: Establishing a hierarchical decision tree model, wherein the hierarchical decision tree model includes a temperature-substance conversion synergistic feature branch, a spatial distribution feature branch, and a Maillard reaction integrity branch; Inputting the temperature-material conversion synergistic feature value in the carbon roasting quality prediction vector into the temperature-material conversion synergistic feature branch, and determining it as secondary when the temperature-material conversion synergistic feature value is lower than a first feature threshold; Inputting the spatial distribution characteristics in the carbon roasting quality prediction vector into the spatial distribution characteristic branch, calculating the quality uniformity index, and determining it as excellent when the quality uniformity index is lower than the second characteristic threshold; Inputting the Maillard reaction feature in the carbon-roasted quality prediction vector into the Maillard reaction integrity branch, and determining that the product is of special grade when the product of the slope of the aldehyde-ketone ratio curve and the cumulative amount of melanoidins exceeds a third feature threshold; According to the variety information in the basic parameters of the tea leaves, the corresponding first characteristic threshold, second characteristic threshold and third characteristic threshold are extracted from a preset grading standard library; performing time decay compensation on the first characteristic threshold, the second characteristic threshold, and the third characteristic threshold according to the actual carbon baking time in the carbon baking process parameter; Output includes preliminary grade determination results of inferior, superior and special grades; The hierarchical decision tree model is configured to trigger a cascaded fire optimization process based on the clustering quality index, including: Calculating the clustering quality index of the preliminary grade determination results, wherein the clustering quality index is determined by the feature space distance variance between the special grade samples and the excellent grade samples; When the clustering quality index is lower than a preset clustering threshold, the cascaded supplementary firing optimization process is activated.

7. The method for grading semi-refined tea based on carbon roasting parameters according to claim 1, wherein: The optimization strategy of the cascade supplementary fire optimization is configured to generate a combination of supplementary fire duration and temperature using a material conversion model, including: Establishing a material conversion kinetics model, wherein the input features of the material conversion kinetics model include a temperature-material conversion synergistic feature deviation value and a spatial distribution feature non-uniformity in the carbon roasting quality prediction vector; Calculating an optimal supplementary fire parameter combination according to the material conversion kinetics model, and generating an optimization strategy for the cascade supplementary fire optimization according to the optimal supplementary fire parameter combination; Performing cascaded supplementary fire optimization on tea leaves that are initially graded as secondary to obtain optimized tea leaves, and re-grading the optimized tea leaves to obtain an updated grade determination result, including: Recollecting the optimized carbon roasting process parameters and hyperspectral imaging information of the tea leaves, recording them as optimized carbon roasting parameters and optimized imaging information, and updating the carbon roasting quality prediction vector, recording them as optimized quality prediction vector; The optimized quality prediction vector is input into the hierarchical decision tree model for secondary determination to generate the updated grade determination result.

8. The method for grading semi-refined tea based on carbon roasting parameters according to claim 1, wherein: The final grading result of the current tea is generated based on the preliminary and updated grading results. The final grading result is associated with the carbon roasting process parameters and stored in the blockchain. A grading report is generated that includes the grade distribution probability, process defect location, and energy consumption assessment, including: Calculate the weighted average of the preliminary special-grade ratio, preliminary superior-grade ratio, and preliminary sub-grade ratio in the preliminary grade determination results and the updated special-grade ratio, updated superior-grade ratio, and updated sub-grade ratio in the updated grade determination results to generate the final special-grade ratio, final superior-grade ratio, and final sub-grade ratio in the final grading results; Establishing a correlation mapping relationship between the final classification result and the three-dimensional temperature field information of the baking cage and the key temperature duration in the carbon baking process parameters, wherein the correlation mapping relationship includes a spatial correspondence between temperature field uniformity and quality grade distribution, and a temporal correspondence between the key temperature duration and Maillard reaction integrity; The final grading result, the carbon baking process parameters and the associated mapping relationship are hashed to generate a data block, and the data block is written into the blockchain network for distributed storage. The data block includes a grade distribution probability matrix, process defect location coordinates and carbon baking energy consumption statistical indicators; Based on the data blocks, a visual grading report is generated, including a three-dimensional temperature field distribution and a grade distribution superimposed thermal map, a process defect improvement suggestion plan, and an energy efficiency optimization analysis report.

9. A computer-readable storage medium storing computer program instructions, characterized in that: The computer program instructions implement the method according to any one of claims 1 to 8 when executed by a processor.

10. An electronic device comprising a memory and a processor, characterized in that: The memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 8.

Images