A tea quality evaluation method and system based on image analysis

Abstract

The application relates to the technical field of tea quality evaluation, in particular to a tea quality evaluation method and system based on image analysis. The method comprises the following steps: acquiring tea imaging data, extracting surface features of the tea based on the tea imaging data; reconstructing the tea imaging data to obtain structure information; acquiring processing parameters; identifying internal defects from the structure information based on the processing parameters to obtain internal quality features; judging the overall quality according to the processing parameters, the internal quality features and the surface features to obtain a judgment result; and outputting the quality grade of the tea according to the judgment result. The method solves the technical problem that the information gap between two-dimensional surface images and three-dimensional internal quality exists when the existing intelligent evaluation method based on image analysis evaluates the internal quality of tea and the visual changes caused by new processes, and the evaluation result has a significant deviation.

Description

Technical Field

[0001] This application relates to the technical field of tea quality evaluation. Specifically, it relates to a method and system for tea quality evaluation based on image analysis. Background Art

[0002] Currently, the evaluation of tea quality still mainly relies on manual sensory detection, which has problems such as low efficiency, inconsistent standards, and large subjective errors. For this reason, the industry has introduced intelligent evaluation methods and systems based on image analysis to achieve rapid determination in an objective and quantitative manner. However, with the continuous innovation of processing technologies and more refined quality requirements, the existing systems mainly rely on extracting appearance features from two-dimensional images on the surface of tea, and it is difficult to cover key internal quality attributes, such as the integrity of internal structures, the uniformity of heat distribution in the thickness direction, microscopic damage to cell tissues, and three-dimensional internal features such as internal mildew and insect damage. Many internal defects do not form obvious signals on the surface in the early stage and may not be exposed until storage or brewing, affecting the final quality and consumer experience. Since two-dimensional imaging cannot perceive information below the surface layer, the system may misjudge tea with internal hidden dangers as qualified; it may also wrongly downgrade or reject tea with excellent internal quality but an appearance that deviates from traditional standards due to subtle texture or color differences. Thus, an information gap is formed between two-dimensional appearance and three-dimensional internal quality, resulting in an increase in the deviation between the evaluation result and the actual sensory quality and consumer experience, and weakening the credibility and role of the intelligent evaluation system in comprehensive quality control.

[0003] In view of the above problems, there is an urgent need for improvement in the existing technology. Summary of the Invention

[0004] This application discloses a method and system for tea quality evaluation based on image analysis, aiming to solve the technical problem of the information gap between two-dimensional surface images and three-dimensional internal quality when the existing intelligent evaluation method based on image analysis evaluates the internal quality of tea and the visual changes brought by new processes, and there are significant deviations between the evaluation result and the actual sensory quality and consumer experience.

[0005] The technical solution of this application is as follows: In the first aspect, this application discloses a method for tea quality evaluation based on image analysis, including: Obtaining tea imaging data containing light intensity information and light direction information on a tea production line, and extracting surface features of the tea based on the tea imaging data; Processing the tea imaging data to reconstruct structural information of multiple layers of the tea along the thickness direction; Obtaining processing process parameters corresponding to the tea; Based on the processing process parameters, identifying the internal structural state, uniformity in the thickness direction, and internal defects of the tea from the structural information to obtain internal quality characteristics; The overall quality of tea is judged based on processing parameters, internal quality characteristics, and surface characteristics, and the judgment result is obtained. Based on the judgment results, the quality grade of the tea is output and fed back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

[0006] Furthermore, when processing the tea leaf imaging data to reconstruct the structural information of the tea leaves at multiple levels along the thickness direction, the method includes: The tea leaf imaging data was analyzed into multiple subaperture views; By comparing the local differences between multiple subaperture views, a deviation map characterizing spatial non-uniform distortion is obtained. Based on the deviation diagram, adjust the ray propagation path parameters used for reconstruction; Based on the adjusted light propagation path parameters, the tea leaf imaging data is reconstructed to obtain structural information of multiple layers of tea leaves along the thickness direction.

[0007] Furthermore, based on processing parameters, the internal structural state, thickness uniformity, and internal defects of tea leaves are identified from structural information to obtain internal quality characteristics. This process includes: In the structural information, the tea leaves are divided into multiple local regions; Obtain multiple sets of internal features for each local region; Based on the internal feature sets of multiple local regions, the overall internal feature statistical center value of tea is determined. Select at least one local region as the current local region, and determine at least one local region adjacent to the current local region as the surrounding local regions; Calculate the degree of difference between the internal feature set of the current local region and the internal feature set of the surrounding local regions, as well as the degree of deviation of the internal feature set of the current local region from the statistical center value of the overall internal features; Quantify the degree of anomaly in the current local area based on the degree of difference and deviation; Based on the degree of anomaly and a preset composite threshold, mark potential internal defect areas; Based on the processing parameters, the threshold for judging the degree of abnormality in potential internal defect areas is adjusted, and based on the adjusted threshold, the internal structural state, uniformity in the thickness direction, and internal defects of the tea are identified to obtain internal quality characteristics.

[0008] Furthermore, based on processing parameters, internal quality characteristics, and surface characteristics, the overall quality of the tea is judged to obtain a judgment result. This process includes: Based on the processing parameters, a combination of processing parameters is formed. The preset process parameter-quality standard mapping table is then queried to see if there is an entry corresponding to the combination of processing parameters. When there is an entry in the process parameter-quality standard mapping table that corresponds to the combination of processing process parameters, the internal quality characteristics and surface characteristics are matched and judged according to the quality standard corresponding to the entry to obtain the judgment result; When there is no entry in the process parameter-quality standard mapping table that corresponds to the combination of processing parameters, analyze the mutual influence between internal quality characteristics and surface characteristics. Temporary quality judgment rules are constructed based on mutual influence relationships, and the weights of internal quality characteristics and surface characteristics in the overall quality judgment are adjusted based on mutual influence relationships. Based on the adjusted weights and temporary quality judgment rules, a preliminary judgment is made on the overall quality of the tea, and a preliminary judgment result is obtained. The preliminary judgment results are compared with the characteristic distribution of historical high-quality tea to confirm the effectiveness of the provisional quality judgment rules and obtain the effectiveness confirmation results. Based on the validity confirmation results, the overall quality of the tea is judged, and the judgment result is obtained.

[0009] Furthermore, the preliminary judgment results are compared with the characteristic distribution of historical high-quality teas to confirm the effectiveness of the provisional quality judgment rule, thus obtaining a validity confirmation result. This process includes: Obtain the feature set corresponding to the preliminary judgment result. This feature set includes the internal quality characteristics and surface characteristics corresponding to the preliminary judgment result. Based on the processing parameters, the characteristic distribution of historical high-quality tea is adjusted to obtain a dynamic reference distribution that matches the processing parameters. Compare the similarity between the feature set and the dynamic reference distribution to identify deviation features in the feature set that deviate from the dynamic reference distribution; Analyze the potential impact of deviation characteristics on tea quality and determine whether deviation characteristics represent innovative quality. When a deviation characteristic is judged to represent innovative quality, the effectiveness confirmation level of the temporary quality judgment rule is increased; Based on the validity confirmation level, the validity of the provisional quality judgment rule is confirmed, and the validity confirmation result is obtained.

[0010] Furthermore, the potential impact of deviation characteristics on tea quality is analyzed, and it is determined whether the deviation characteristics represent innovative quality. This process includes: Obtain multi-dimensional data of the deviation features, including the spatial distribution, intensity variation, and correlation with surrounding structural elements of the deviation features in the structural information; Acquire quality factor data related to processing parameters, including the chemical composition, physical structure, and sensory evaluation indicators of tea leaves; Based on historical data, correlation analysis was performed on multiple dimensions of data and quality factor data to establish a mapping relationship between multiple dimensions of data and quality factor data. This historical data is a set of sample data formed during the production and quality control of historical tea batches. The mapping relationship is adjusted according to the processing parameters; Based on the adjusted mapping relationship, the degree of influence of the deviation feature on multiple quality factors in the quality factor data is calculated. Based on the degree of impact and the preset innovation quality judgment rules, determine whether the deviation characteristics represent innovation quality.

[0011] Furthermore, based on the degree of impact and the pre-defined rules for judging innovation quality, it is determined whether the deviation characteristics represent innovation quality. This process includes: Based on the processing parameters, retrieve the set of quality factor weights that match the processing parameters from the preset process-quality association database; When there is a set of quality factor weights that match the processing parameters in the process-quality correlation database, the degree of influence is weighted and summarized based on the set of quality factor weights to obtain the innovation quality judgment value. Based on the preset innovation quality discrimination rules and innovation quality judgment values, determine whether the deviation characteristics represent innovation quality; When there is no set of quality factor weights that match the processing parameters in the process-quality association database, the quality factors that show positive changes among multiple quality factors are identified based on the degree of influence. Temporary innovation quality discrimination rules are dynamically generated based on the positively changing quality factors. These temporary innovation quality discrimination rules include a priority judgment logic for the positively changing quality factors. Based on the temporary innovation quality judgment rules, determine whether the deviation characteristics represent innovation quality.

[0012] Furthermore, after determining whether the deviation characteristics represent innovation quality, the process also includes adaptive optimization of the priority judgment logic of the temporary innovation quality discrimination rule. This adaptive optimization includes: Record the deviation characteristics that are determined by the temporary innovative quality discrimination rules to represent innovative quality in the judgment results of each batch of tea as innovative deviation characteristics, and record the degree of influence of the quality factors corresponding to the innovative deviation characteristics; Based on the judgment results of multiple batches of tea, the verification results are obtained to characterize the accuracy of the identification of innovation deviation features, and the accuracy of the identification of innovation deviation features is determined based on the verification results; When the identification accuracy has a continuous deviation from the preset accuracy threshold for a preset number of consecutive batches, analyze the combination of quality factors that cause the continuous deviation and the degree of their impact. Based on the persistent deviation, and combined with the adjustment direction and magnitude of the priority judgment weights determined by the combination of quality factors and their degree of influence, the priority judgment weights corresponding to the combination of quality factors in the temporary innovation quality judgment rules are adjusted. The adjusted priority judgment weights are applied, and the recognition accuracy is continuously updated to achieve adaptive optimization of the priority judgment logic.

[0013] Furthermore, the analysis examines the combinations of quality factors that lead to persistent bias and the extent of their influence, including: In the batch corresponding to the persistent deviation, structural characterization indicators corresponding to multiple quality factors are extracted from the structural information based on the mapping relationship, forming quality factor characterization data of multiple quality factors. The quality factor characterization data includes the spatial distribution, intensity change and correlation degree between multiple quality factors of the structural characterization indicators used to characterize multiple quality factors in the structural information. Based on the processing parameters and the quality factor characterization data, the potential nonlinear interactions between multiple quality factors are identified and an interaction model is established. Based on the interaction model, the contribution of nonlinear interactions among multiple quality factors to persistence bias is quantified. Based on the degree of contribution, the key quality factor combinations that lead to persistent deviations and their nonlinear interaction patterns are identified, thus obtaining the quality factor combinations that lead to persistent deviations and their degree of influence.

[0014] Secondly, this application also discloses a tea quality evaluation system based on image analysis, comprising: The imaging module is used to acquire tea imaging data containing light intensity and light direction information from the tea production line, and to extract the surface features of the tea based on the tea imaging data; The calculation module is used to process the tea imaging data and reconstruct the structural information of the tea leaves at multiple levels along the thickness direction. The parameter acquisition module is used to acquire the processing parameters corresponding to tea leaves. The identification module is used to identify the internal structural state, uniformity in the thickness direction, and internal defects of tea leaves from structural information based on processing parameters, thereby obtaining internal quality characteristics. The judgment module is used to judge the overall quality of tea based on processing parameters, internal quality characteristics and surface characteristics, and obtain the judgment result; The output module is used to output the quality grade of the tea based on the judgment result, and feed the quality grade back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

[0015] Beneficial effects: The image analysis-based tea quality evaluation method disclosed in this application obtains tea imaging data containing light intensity and light direction information from the tea production line, extracts surface features of the tea based on the data, and processes the imaging data to reconstruct the structural information of the tea at multiple levels along the thickness direction, thereby overcoming the limitation of the prior art that only relies on two-dimensional surface images for evaluation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a tea quality evaluation method based on image analysis provided in this application.

[0017] Figure 2 A flowchart of a tea quality evaluation system based on image analysis provided in this application.

[0018] In the diagram: 1. Imaging module; 2. Calculation module; 3. Parameter acquisition module; 4. Recognition module; 5. Judgment module; 6. Output module. Detailed Implementation

[0019] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Reference Figure 1 This application proposes a tea quality evaluation method based on image analysis, including: S1000: Acquires tea imaging data containing light intensity and light direction information from the tea production line, and extracts surface features of the tea based on the tea imaging data; S2000: Process the tea leaf imaging data to reconstruct the structural information of the tea leaves at multiple levels along the thickness direction; S3000: Obtain the processing parameters corresponding to tea leaves; S4000: Based on processing parameters, it identifies the internal structural state, uniformity in the thickness direction, and internal defects of tea leaves from structural information to obtain internal quality characteristics. S5000: The overall quality of tea is judged based on processing parameters, internal quality characteristics, and surface characteristics, and the judgment result is obtained. S6000: Outputs the quality grade of tea based on the judgment result and feeds the quality grade back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

[0022] Tea leaf imaging data refers to raw image data acquired on a tea production line using specific imaging equipment. This data includes light intensity and light direction information. Light intensity information reflects the degree to which the tea leaves reflect or absorb incident energy and can be used to analyze color, luster, etc. Light direction information records changes in the direction of energy propagation, providing a basis for subsequent reconstruction of the structural information of multiple layers of the tea leaves along the thickness direction. For example, this information can be acquired using techniques such as multi-angle light source illumination or structured light projection.

[0023] Surface features refer to visual attributes extracted from tea imaging data that characterize the appearance quality of tea, including but not limited to the color uniformity, leaf integrity, tightness of the tea leaves, degree of bud and down exposure, and the presence of visible surface defects such as mold and insect damage. These surface features can be extracted using image processing algorithms such as edge detection, texture analysis, and color space conversion.

[0024] Structural information refers to the internal structural data of tea leaves along their thickness direction, obtained through depth processing and reconstruction of tea leaf imaging data. This is a three-dimensional or quasi-three-dimensional data representation that can reveal tissue structure, cell arrangement, pore distribution, and internal damage or foreign objects. The reconstruction process may involve techniques such as computational photography, tomography, or light field imaging to recover three-dimensional spatial information from two-dimensional projection data.

[0025] Processing parameters refer to the process control variables that affect the quality of tea during production, including but not limited to fixation temperature and time, rolling force and time, fermentation degree, drying temperature and time, etc.

[0026] Internal quality characteristics refer to the internal quality attributes identified based on structural information and processing parameters. These include the internal structural state of tea leaves (such as cell integrity and tissue density), uniformity in the thickness direction (such as uniform heating and moisture distribution), and internal defects (such as internal mold, insect infestation, hollowness, and breakage). These internal quality characteristics are difficult to obtain directly from traditional two-dimensional images, but they have a significant impact on core qualities such as brewing endurance, aroma, and taste.

[0027] Quality grade refers to the classification of tea based on the final judgment result. It is usually divided into special grade, first grade, second grade, etc., or more detailed classification according to specific standards.

[0028] A tea production line refers to the industrialized process of tea production from fresh leaf picking to finished product packaging, including steps such as fixation, rolling, fermentation, drying, and refining. This application serves as a quality control node, and its output quality grade is used to guide subsequent grading, packaging, and / or process adjustments, forming a closed-loop control system.

[0029] The implementation environment of this application typically includes: a testing station (e.g., located after the drying or refining process) integrating high-resolution imaging equipment (such as a multispectral camera, light field camera, or X-ray tomography equipment); a high-performance computing platform for performing image processing, 3D reconstruction, feature extraction, and quality judgment algorithms; and an interface connected to the production line control system for acquiring processing parameters and providing feedback on quality levels.

[0030] Firstly, regarding the acquisition of tea imaging data containing light intensity and direction information from the tea production line, and the extraction of surface features based on this imaging data, there are several possible approaches. One approach is to install a multispectral imaging system on the tea production line, using multiple light sources of different wavelengths and image sensors to acquire data containing rich light intensity information; and to employ polarization imaging technology to acquire light direction information, for example, by using a linear polarizer rotated to different angles for multiple imaging operations, and then calculating the degree of polarization image and the polarization angle image. Based on this imaging data, surface features such as color uniformity, leaf integrity, and leaf tightness can be extracted using methods such as color histogram analysis, texture feature extraction (e.g., gray-level co-occurrence matrix, local binary mode), and edge detection (e.g., Canny operator). Another approach is to use a structured light projection system to project a known pattern onto the tea surface and capture the deformed pattern to obtain three-dimensional surface morphology information, thereby indirectly obtaining light intensity and direction information; and then extracting surface texture, shape, and other features from the three-dimensional morphology data.

[0031] Secondly, regarding the processing of tea leaf imaging data to reconstruct the structural information of the tea leaves at multiple levels along the thickness direction, there are various ways to achieve this. One approach is based on the principle of light field imaging: capturing images of the tea leaves from different viewpoints and reconstructing the light field, calculating focused images of different internal depths from the multi-viewpoint images; for example, using a microlens array camera or a multi-camera array to acquire light field data. Another approach is to use optical coherence tomography (OCT) technology: using low-coherence light interference to perform tomographic scanning of the tea leaves, obtaining a series of cross-sectional images along the thickness direction, thereby forming structural information.

[0032] Furthermore, regarding the acquisition of processing parameters for tea, this can be collected by sensors and control systems in the tea production line and correlated with batches / samples. For example, temperature sensors in the fixation machine record the fixation temperature, timers in the rolling machine record the rolling time, temperature and humidity sensors in the fermentation room record fermentation environmental parameters, and temperature and wind speed sensors in the drying equipment record drying parameters. All of this data can be acquired in real time through a data acquisition system.

[0033] Next, regarding the step of identifying the internal structural state, thickness uniformity, and internal defects of tea leaves based on processing parameters to obtain internal quality characteristics, this can include: using image segmentation algorithms (such as thresholding, region growing, and deep learning segmentation models) to separate the tea leaf region from the background in the structural information; dividing the tea leaf into multiple local regions for local feature analysis; extracting internal feature sets from each local region, such as cell density, porosity, fiber arrangement direction, and abnormal structures (such as mold hyphae and insect holes), and obtaining them through morphological analysis, texture analysis, or deep learning feature extractors. This is then interpreted and judged in conjunction with processing parameters: for example, if processing parameters indicate uneven drying and heating, the structural information may show local cell structure damage or uneven moisture distribution; based on this, the internal structural state, thickness uniformity, and internal defects can be identified through preset rules or a trained classification model, thereby obtaining internal quality characteristics.

[0034] Then, regarding the step of judging the overall quality of tea based on processing parameters, internal quality characteristics, and surface characteristics, and obtaining the judgment result, a multi-feature fusion machine learning model can be used, such as Support Vector Machine (SVM), Random Forest, or neural network model, taking processing parameters, internal quality characteristics, and surface characteristics as input; the model establishes a mapping relationship by learning historical tea data (corresponding processing parameters, internal quality characteristics, surface characteristics, and final quality grade), and outputs a quality score or quality grade prediction for new samples.

[0035] Finally, regarding the step of outputting the tea quality grade based on the judgment result and feeding the quality grade back to the tea production line to guide subsequent grading, packaging, and / or process adjustment, the continuous quality score can be mapped into discrete quality grades such as premium, first-grade, and second-grade according to the interval, and sent to the execution unit through the production line control system interface: the grading equipment sorts the tea according to the quality grade; the packaging equipment selects packaging materials or labels according to the quality grade; and the process adjustment system automatically or semi-automatically adjusts the upstream processing parameters based on the feedback to optimize the tea quality.

[0036] The image analysis-based tea quality assessment method proposed in this application significantly improves the comprehensiveness, accuracy, and intelligence of tea quality assessment by introducing multi-dimensional imaging data, reconstructing internal structural information, and combining processing parameters for comprehensive judgment, and realizing production line feedback on quality grade. This provides technical support for quality control and quality improvement in the tea industry.

[0037] In another embodiment of this application, S2000 specifically includes the following steps: S2100: Resolves tea leaf imaging data into multiple subaperture views; S2200: Compare the local differences between multiple subaperture views to obtain a deviation map characterizing spatial non-uniform distortion; S2300: Adjust the ray propagation path parameters used for reconstruction based on the deviation diagram; S2400: Based on the adjusted light propagation path parameters, the tea leaf imaging data is reconstructed to obtain structural information of multiple layers of tea leaves along the thickness direction.

[0038] Specifically, decomposing tea leaf imaging data into multiple subaperture views involves using digital image processing techniques or computational optics to break down the raw tea leaf imaging data, which contains information about light intensity and direction, into a series of image subsets observed from different micro-viewpoints or local regions. Each subaperture view corresponds to a slightly different observation angle or light path, capturing subtle changes and distortions as light penetrates the tea leaf medium. This can be achieved, for example, by performing Fourier transforms, wavelet decomposition, or deep learning-based feature extraction on the raw imaging data.

[0039] The process of comparing local differences between multiple subaperture views to obtain a deviation map characterizing spatial non-uniform distortion involves comparing local features such as pixel values, textures, edges, or structures between subaperture views to quantify the spatial non-uniform distortion introduced by light propagation along different paths. This distortion can manifest as local misalignment, blurring, or uneven brightness. The deviation map can be a two-dimensional or three-dimensional data matrix, with its elements representing the degree of distortion at specific spatial locations or along light paths, thus providing a locatable and quantifiable basis for subsequent correction. For example, image registration and comparison techniques such as cross-correlation, structural similarity index (SSIM), or optical flow can be used to calculate these local differences.

[0040] In practical applications, adjusting the light propagation path parameters used for reconstruction based on the deviation map refers to using the spatial non-uniform distortion information reflected in the deviation map to correct the optical model parameters or algorithm parameters used for reconstruction. These parameters include the refractive index, scattering coefficient, propagation direction, and phase information of the light, to compensate for distortions caused by the uneven internal structure of the tea leaves or the external environment, making the reconstruction process closer to the actual light propagation process. For example, iterative optimization algorithms, machine learning models, or backpropagation algorithms based on physical models can be used to adjust these parameters according to the deviation map.

[0041] Furthermore, based on the adjusted light propagation path parameters, the tea leaf imaging data is reconstructed to obtain structural information of multiple layers along the thickness direction. This involves re-executing the image reconstruction algorithm after the light propagation path parameters have been corrected to eliminate or significantly reduce errors caused by spatial non-uniform distortion, resulting in more accurate and clearer structural information. This structural information can characterize the arrangement of tea leaf cells, the distribution of leaf veins, the size and distribution of internal pores, etc., thus providing a reliable data foundation for subsequent identification of internal quality features. For example, techniques such as tomography, light field reconstruction, or deep learning reconstruction networks can be used for reconstruction.

[0042] In another embodiment of this application, S4000 is further proposed to include: S4100: In the structural information, the tea leaves are divided into multiple local regions; S4200: Obtain multiple sets of internal features for each local region; S4300: Determine the overall internal feature statistical center value of tea based on the internal feature set of multiple local regions; S4400: Select at least one local region as the current local region, and determine at least one local region adjacent to the current local region as the surrounding local regions; S4500: Calculate the degree of difference between the internal feature set of the current local region and the internal feature set of the surrounding local regions, as well as the degree of deviation of the internal feature set of the current local region from the statistical center value of the overall internal features; S4600: Quantifies the degree of anomaly in the current local area based on the degree of difference and deviation; S4700: Mark potential internal defect areas based on the degree of anomaly and a preset composite threshold; S4800: Based on the processing parameters, adjust the threshold for judging the degree of abnormality in potential internal defect areas, and based on the adjusted threshold for judging the degree of abnormality, identify the internal structural state, uniformity in the thickness direction, and internal defects of the tea leaves to obtain internal quality characteristics.

[0043] Specifically, dividing tea leaves into multiple local regions refers to using image segmentation techniques, such as mesh partitioning, connected component analysis, or deep learning segmentation, to subdivide the reconstructed structural information of tea leaves along the thickness direction into several smaller, independently analyzable regions. This allows for refined and localized analysis of the internal structure of tea leaves and the capture of local anomalies.

[0044] In this context, obtaining multiple sets of internal features for each local region refers to extracting features from each local region. These features include, but are not limited to, density, porosity, texture features (such as gray-level co-occurrence matrix features), fiber arrangement direction, and cell structure integrity, in order to quantify the internal structural characteristics of each local region and provide a data basis for subsequent comparisons.

[0045] In practical applications, determining the overall internal characteristic statistical center value of tea based on the internal characteristic sets of multiple local regions refers to performing statistical analysis on the internal characteristic sets of all local regions, such as calculating the mean, median, or mode, to obtain a global reference value that represents the average or typical state of the internal structure of the entire tea leaf, which is used to assess the degree of deviation of local regions from the whole.

[0046] Furthermore, selecting at least one local region as the current local region and determining at least one local region adjacent to the current local region as the surrounding local regions refers to establishing a local context by defining neighborhood relationships (e.g., 8 directly adjacent regions or regions determined by a certain distance threshold) to support local difference analysis.

[0047] Therefore, calculating the degree of difference between the internal feature set of the current local region and the internal feature set of the surrounding local regions, as well as the degree of deviation between the internal feature set of the current local region and the statistical center value of the overall internal feature set, refers to using mathematical measures to quantify neighborhood consistency and global deviation, such as calculating Euclidean distance, Mahalanobis distance, correlation coefficient, or statistical significance test; among them, the degree of difference is used to characterize the consistency between the current local region and the neighboring regions, and the degree of deviation is used to characterize the difference between the current local region and the typical level of the whole tea leaf.

[0048] Quantifying the degree of anomaly in a local area based on the degree of difference and deviation means integrating the degree of difference and deviation into a single anomaly index, such as by using weighted summation, fuzzy logic reasoning, or machine learning models. The higher the degree of anomaly, the more likely there is anomaly in the internal structure of the local area.

[0049] Marking potential internal defect areas based on the degree of anomaly and a preset composite threshold means initially marking local areas with an anomaly degree exceeding the preset composite threshold as potential internal defect areas; the preset composite threshold can be set based on historical data, expert experience, or standards for specific tea varieties.

[0050] One of the key innovations of this application is adjusting the threshold for judging the degree of abnormality of potential internal defect areas based on processing parameters. Different processing parameters (e.g., fixation temperature, rolling intensity, drying time, etc.) will change the normal distribution range of internal characteristics. For example, light rolling allows for higher porosity, while heavy rolling requires a denser structure. Therefore, by dynamically adjusting the threshold for judging the degree of abnormality through processing parameters, the defect identification standard can be matched with the current production background. The adjusted threshold is then used to finally identify the internal structural state, uniformity in the thickness direction, and internal defects of the tea leaves, so as to obtain more accurate and reliable internal quality characteristics.

[0051] In some preferred embodiments, the following specific example illustrates the situation: Suppose we are conducting a quality assessment on a batch of black tea produced using a specific rolling process. First, an imaging module acquires the imaging data of the black tea, which is then processed by a calculation module to reconstruct its structural information across multiple layers along the thickness direction. Next, in the recognition module, the structural information of the black tea is divided into 100 local regions, and internal features such as porosity, fiber density, and cell wall integrity are extracted from each local region, forming a set of internal features for each local region. Finally, the average value of porosity, fiber density, and cell wall integrity across all local regions is calculated as the statistical center value of the overall internal features of the entire batch of black tea.

[0052] Specifically, one local region is selected as the current local region, and eight neighboring local regions are identified. Then, the average difference between the internal feature set of the current local region and the internal feature sets of these eight surrounding local regions is calculated, as well as the deviation of the current local region's internal feature set from the statistical center value of the overall internal feature set. For example, if the porosity of the current local region is significantly higher than that of its surrounding regions and the overall average level, it indicates that there may be an anomaly in that region.

[0053] Based on the calculated degree of difference and deviation, the anomaly level of the current local area is quantified using a preset weighted model. For example, if the processing parameters of the black tea indicate that its rolling intensity is medium, the identification module will retrieve an anomaly judgment threshold applicable to medium-rolled black tea from a preset process-threshold database based on these parameters. This threshold will be stricter than the threshold for lightly rolled black tea, but more lenient than the threshold for heavily rolled black tea.

[0054] Finally, the degree of anomaly in the current local area is compared with an adjusted judgment threshold. If the degree of anomaly exceeds the adjusted threshold, the local area is identified as having an internal defect (e.g., excessive breakage or loose structure). By performing this analysis on all local areas, the internal structural state, thickness uniformity, and specific internal defects of the batch of black tea can be comprehensively identified, thereby obtaining accurate internal quality characteristics.

[0055] In another embodiment of this application, S5000 is further proposed to include: S5100: Based on the processing parameters, form a combination of processing parameters and query whether there is an entry corresponding to the combination of processing parameters in the preset process parameter-quality standard mapping table. S5200: When there is an entry in the process parameter-quality standard mapping table that corresponds to the combination of processing parameters, the internal quality characteristics and surface characteristics are matched and judged according to the quality standard corresponding to the entry to obtain the judgment result. S5300: When there is no entry in the process parameter-quality standard mapping table that corresponds to the combination of processing parameters, analyze the mutual influence between internal quality characteristics and surface characteristics. S5400: Construct temporary quality judgment rules based on mutual influence relationships, and adjust the weights of internal quality characteristics and surface characteristics in overall quality judgment based on mutual influence relationships; S5500: Based on the adjusted weights and temporary quality judgment rules, a preliminary judgment is made on the overall quality of the tea, and a preliminary judgment result is obtained; S5600: Compare the preliminary judgment results with the characteristic distribution of historical high-quality tea to confirm the validity of the provisional quality judgment rule and obtain the validity confirmation result; S5700: Based on the validity confirmation results, the overall quality of the tea is judged, and a judgment result is obtained.

[0056] Specifically, forming a processing parameter combination based on processing parameters refers to integrating all relevant processing parameters of the current batch of tea, such as fixation temperature, rolling time, fermentation degree, and drying method, into a unique identifier or vector, and then querying it in a preset process parameter-quality standard mapping table. The process parameter-quality standard mapping table is used to store the relationship between known processing parameter combinations and corresponding quality standards, so as to provide a fast and standard basis for quality judgment for common or verified tea production.

[0057] Specifically, when there is an entry in the process parameter-quality standard mapping table that corresponds to the combination of processing parameters, the internal quality characteristics and surface characteristics are matched and judged according to the quality standard corresponding to the entry. The judgment result means that when the query is matched, the system directly calls the preset quality standard in the entry. The quality standard includes specific requirements and thresholds for internal quality characteristics (such as internal structural state, uniformity, defects, etc.) and surface characteristics (such as color, shape, integrity, etc.). The matching judgment is completed and the judgment result is output by comparing the actual internal quality characteristics and surface characteristics of the current tea with the requirements and thresholds.

[0058] In practical applications, when there is no entry in the process parameter-quality standard mapping table corresponding to the combination of processing parameters, analyzing the interaction between internal quality characteristics and surface characteristics means that when encountering new or unknown combinations of processing parameters, the system no longer relies on preset standards, but analyzes the inherent connection and interaction between the current internal quality characteristics and surface characteristics of tea under specific process conditions. For example, a certain processing technology may cause the internal structure of tea to be loose, thereby affecting the uniformity of its surface color; or a certain surface defect may be associated with a specific internal structural anomaly, in order to reveal the formation mechanism of quality characteristics under the current process conditions.

[0059] Furthermore, constructing temporary quality judgment rules based on mutual influence relationships and adjusting the weights of internal quality characteristics and surface characteristics in the overall quality judgment based on these relationships means that the system dynamically generates temporary quality judgment rules based on the mutual influence relationships and simultaneously adjusts the weights of internal quality characteristics and surface characteristics in the overall quality judgment. For example, if a strong negative correlation is found between a certain internal defect and surface gloss, the weight of the internal defect is increased in the temporary quality judgment rules, and the weights of other characteristics are adjusted accordingly to more accurately reflect the actual quality status of the tea.

[0060] Therefore, based on the adjusted weights and temporary quality judgment rules, a preliminary judgment is made on the overall quality of the tea. The preliminary judgment result refers to the first evaluation and output of the preliminary judgment result using dynamically generated temporary quality judgment rules and adjusted weights. The preliminary judgment result reflects the comprehensive evaluation conclusion of the current tea under the internal and surface characteristics and their interactions.

[0061] The preliminary judgment results are compared with the characteristic distribution of historical high-quality teas to confirm the effectiveness of the provisional quality judgment rule. The effectiveness confirmation result means that the system compares the tea characteristics represented by the preliminary judgment results with the characteristic distribution of historical high-quality teas. The characteristic distribution of historical high-quality teas is used to characterize the characteristic range and pattern of industry-recognized high-quality teas. The effectiveness confirmation result obtained by comparison is used to evaluate the accuracy and effectiveness of the provisional quality judgment rule in identifying high-quality teas.

[0062] Finally, based on the validity confirmation results, the overall quality of the tea is judged. The judgment result refers to the final judgment made by the system based on the validity confirmation results: if the validity confirmation results show that the provisional quality judgment rule is reliable, the preliminary judgment result is adopted or refined to form the final judgment result; if the validity is insufficient, further analysis or manual intervention is triggered to ensure the accuracy of the judgment result.

[0063] In another embodiment of this application, S5600 is further proposed to include: S5610: Obtain the feature set corresponding to the preliminary judgment result. The feature set includes the internal quality characteristics and surface characteristics corresponding to the preliminary judgment result. S5620: Based on processing parameters, the characteristic distribution of historical high-quality tea is adjusted to obtain a dynamic reference distribution that matches the processing parameters; S5630: Compare the similarity between the feature set and the dynamic reference distribution to identify the deviation features in the feature set that deviate from the dynamic reference distribution; S5640: Analyze the potential impact of deviation characteristics on tea quality and determine whether deviation characteristics represent innovative quality. S5650: When a deviation characteristic is judged to represent innovative quality, the validity confirmation level of the provisional quality judgment rule is increased; S5660: Based on the validity confirmation level, confirm the validity of the provisional quality judgment rule and obtain the validity confirmation result.

[0064] Specifically, the feature set corresponding to the preliminary judgment result refers to the comprehensive data representation of the internal quality characteristics and surface characteristics of the tea obtained after making a preliminary quality judgment on the tea. Among them, the internal quality characteristics can be understood as the information such as the internal structural state, uniformity in the thickness direction, and internal defects of the tea identified by processing and reconstructing the tea imaging data; the surface characteristics refer to the externally visible features extracted from the tea imaging data, such as color, texture, and shape.

[0065] This method involves adjusting the characteristic distribution of historically high-quality teas based on processing parameters to obtain a dynamic reference distribution that matches these parameters. The aim is to align the characteristic distribution of historically high-quality teas with the specific production conditions of current teas, creating a more comparable benchmark. For example, different processing parameters such as fermentation temperature, roasting time, or rolling intensity can lead to subtle differences in the internal structure or surface characteristics of even high-quality teas. By adjusting these processing parameters, misjudgments caused by differences in processing techniques can be avoided.

[0066] In practical applications, comparing the similarity between a feature set and a dynamic reference distribution, and identifying deviation features that deviate from the dynamic reference distribution, involves quantifying the differences between the current tea characteristics and the dynamic reference distribution using statistical methods or machine learning algorithms, and marking deviation features as those exceeding a preset threshold. For example, methods such as calculating Euclidean distance, cosine similarity, or using cluster analysis can be employed to assess similarity.

[0067] Furthermore, analyzing the potential impact of deviation characteristics on tea quality and determining whether these deviations represent innovative quality aims to distinguish between normal fluctuations / defects and new characteristics with positive significance. For example, a new processing technique may cause certain internal structural characteristics of tea to differ from those of traditional high-quality tea, but this difference may result in a better taste or aroma. In this case, it is necessary to determine whether the deviation characteristic constitutes a quality improvement rather than a defect.

[0068] When a deviation feature is judged to represent innovative quality, the validity confirmation level of the temporary quality judgment rule is increased. The purpose is to improve the credibility and adoption priority of the temporary quality judgment rule in subsequent judgments, so that the system can recognize and absorb new quality manifestations.

[0069] Therefore, based on the validity confirmation level, the validity of the provisional quality judgment rule is confirmed, and the validity confirmation result is obtained. The purpose is to form a rule conclusion after dynamic verification, which provides a basis for the final tea quality judgment. The higher the validity confirmation level, the stronger the ability of the provisional quality judgment rule to handle novel quality characteristics, and the more reliable the corresponding judgment result.

[0070] In another embodiment of this application, S5640 further includes: S5641: Obtain multi-dimensional data of the deviation features, including the spatial distribution of the deviation features in the structural information, intensity changes, and the degree of correlation with surrounding structural elements; S5642: Obtain quality factor data related to processing parameters. Quality factor data includes the chemical composition, physical structure, and sensory evaluation indicators of tea. S5643: Based on historical data, conduct correlation analysis between multi-dimensional data and quality factor data, and establish a mapping relationship between multi-dimensional data and quality factor data. Historical data refers to the set of sample data formed during the production and quality control of historical tea batches. S5644: Adjust the mapping relationship according to the processing parameters; S5645: Based on the adjusted mapping relationship, calculate the degree of influence of the deviation feature on multiple quality factors in the quality factor data; S5646: Based on the degree of impact and the preset innovation quality discrimination rules, determine whether the deviation characteristics represent innovation quality.

[0071] Specifically, obtaining multi-dimensional data on deviation features refers to analyzing structural information to extract quantitative indicators that characterize the deviation features. This multi-dimensional data includes the spatial distribution, intensity variation, and correlation with surrounding structural elements of the deviation feature within the structural information. This forms a comprehensive description of the deviation feature and provides input for subsequent quality impact analysis. Spatial distribution includes its position, shape, size, and relative position to other structural regions along the tea leaf thickness direction; intensity variation includes abnormal fluctuations in its grayscale value, color saturation, or texture features in the image; and correlation with surrounding structural elements includes whether the deviation feature is closely connected to or interacts with specific cellular tissues, vessels, or fibrous structures.

[0072] Obtaining quality factor data related to processing parameters refers to collecting quantifiable indicators directly related to tea quality. This quality factor data includes the chemical composition, physical structure, and sensory evaluation indicators of tea, providing an objective quality evaluation benchmark. Chemical composition includes the content of tea polyphenols, amino acids, caffeine, and aromatic substances; physical structure includes the density, porosity, toughness, and breakage rate of tea leaves; sensory evaluation indicators include the type and intensity of aroma, the richness and mellowness of flavor, the brightness of the liquor color, and the uniformity of the infused leaves. This quality factor data is typically obtained through professional laboratory testing and sensory evaluation.

[0073] In practical applications, correlation analysis based on historical data and quality factor data is used to establish a mapping relationship between the data and the quality factors. This involves modeling and learning from the sample data set formed by the historical tea batch production and quality control process. The historical data set contains structural information under different processing parameters, multi-dimensional data of deviation features, and corresponding quality factor data. By training the historical data set with machine learning algorithms such as regression analysis, neural networks, and support vector machines, a quantitative model from structural features to changes in quality factors is established to reveal the correlation between internal structural features and final quality performance.

[0074] Furthermore, adjusting the mapping relationship based on processing parameters means dynamically modifying the established mapping relationship by considering the differences in the impact of different processing parameters on the structural formation and quality factor transformation pathways, so that the mapping relationship can reflect the quality patterns under specific process conditions. For example, under high-temperature blanching, certain structural features may have a more significant impact on the formation of aromatic substances, while under low-temperature baking, their impact on amino acid retention may be greater.

[0075] Therefore, based on the adjusted mapping relationship, calculating the degree of influence of deviation features on multiple quality factors of quality factor data means inputting the multi-dimensional data of the currently detected deviation features into the mapping model after the processing parameters have been adjusted, and predicting the direction, magnitude or expected amount of change of the deviation features on each quality factor, so as to quantify the potential contribution or damage of deviation features to tea quality.

[0076] Ultimately, judging whether a deviation characteristic represents innovative quality, based on the degree of impact and the preset innovative quality discrimination rules, involves comparing the degree of impact with the preset innovative quality discrimination rules to obtain an objective and consistent judgment conclusion. The preset innovative quality discrimination rules can be thresholds, logical conditions, or scoring standards. For example, if a deviation characteristic leads to a significant increase in a specific positive quality factor (such as floral aromatic substances) without negatively impacting other key quality factors, it is judged as innovative quality, in order to identify new tea qualities with market potential or unique value.

[0077] In some preferred embodiments, the following specific example illustrates the situation: Suppose that in a certain batch of tea production, an off-target feature is identified through image analysis: a unique microporous structure exists inside the tea leaves, with pore size distribution and connectivity significantly different from traditional tea. To determine whether this microporous structure represents innovative quality, multiple dimensions of data on the microporous structure are first acquired, including its spatial distribution along the thickness of the tea leaves (e.g., mainly concentrated in the leaf mesophyll rather than the veins), the statistical distribution of pore size, variations in pore wall strength (e.g., thinner or thicker pore walls), and its correlation with surrounding cell tissues. Simultaneously, quality factor data related to the processing parameters of this batch of tea (e.g., specific withering and fermentation times) are acquired, including the content of tea polyphenols, amino acids, and caffeine, as well as the aroma type and intensity (e.g., whether unique floral and fruity aromas appear), flavor richness and mellowness (e.g., whether it is more mellow and sweet), and liquor brightness as assessed by professional tasters.

[0078] Next, based on a large amount of data from historical tea batches, a mapping relationship between multiple dimensions of microporous structure data and various quality factors will be established. For example, a deep learning model will be trained to learn the influence of different microporous structure features on chemical composition and sensory performance. Since the current batch adopts specific withering and fermentation processes, the mapping relationship will be adjusted according to these processing parameters to reflect the specificity of the influence of microporous structure on quality under the process conditions. For example, under specific fermentation conditions, a certain microporous structure may be more conducive to the formation of specific aromatic substances.

[0079] Then, the multi-dimensional data of the unique microporous structure of the current batch of tea is input into the adjusted mapping model to calculate the degree of influence of the microporous structure on various quality factors. For example, the model predicts that the microporous structure will lead to a slight decrease in tea polyphenol content, but a significant increase in amino acid content, and a significant enhancement of the floral and fruity aroma of the tea, while making the taste more mellow. Finally, according to the preset innovative quality discrimination rules (for example, the rule stipulates that if the amino acid content increases by more than 10% and a unique floral and fruity aroma appears, while there is no obvious negative impact, it is judged as innovative quality), the unique microporous structure is judged to represent innovative quality. In this way, this application can objectively and quantitatively identify the innovative quality brought about by specific structural deviations, providing strong support for the research and development and marketing of tea products.

[0080] In another embodiment of this application, S5646 further includes: S5646-1: Based on the processing parameters, retrieve a set of quality factor weights that match the processing parameters from a preset process-quality association database; S5646-2: When there is a set of quality factor weights that match the processing parameters in the process-quality association database, the degree of influence is weighted and summarized based on the set of quality factor weights to obtain the innovation quality judgment value. S5646-3: Based on the preset innovation quality discrimination rules and innovation quality judgment values, determine whether the deviation characteristics represent innovation quality; S5646-4: When there is no set of quality factor weights that match the processing parameters in the process-quality association database, identify the quality factors that show positive changes among multiple quality factors based on the degree of influence. S5646-5: Dynamically generate temporary innovation quality discrimination rules based on positively changing quality factors. The temporary innovation quality discrimination rules include priority judgment logic for positively changing quality factors. S5646-6: Based on the temporary innovation quality judgment rules, determine whether the deviation characteristics represent innovation quality.

[0081] Specifically, when determining whether a deviation characteristic represents innovative quality, the process first retrieves a set of quality factor weights that match the current processing parameters from a pre-defined process-quality association database. This process-quality association database can be understood as a knowledge base storing the importance weights of different processing parameters and their corresponding quality factors. Its purpose is to differentiate the degree of influence of various quality factors under different processing conditions. For example, for black tea with a specific fermentation process, aroma quality factors may have higher weights, while for a certain type of green tea, freshness quality factors may be more critical.

[0082] Specifically, when a set of quality factor weights matching the current processing parameters can be found in the process-quality correlation database, the degree of influence is weighted and summarized based on this set of weights to obtain a comprehensive innovation quality judgment value. This weighted summarization process aims to more accurately reflect the contribution of each quality factor to innovation quality under a specific process background. Subsequently, this innovation quality judgment value is compared with preset innovation quality discrimination rules to make a final judgment. The preset innovation quality discrimination rules can be a series of thresholds, logical conditions, or machine learning models used to define which judgment value range or pattern represents innovation quality.

[0083] In practical applications, when the process-quality association database does not contain a set of quality factor weights that match the current processing parameters, it usually means that the current processing parameters are novel, unrecorded, or have a unique combination. In this case, the system will identify quality factors showing positive changes among multiple quality factors based on the calculated degree of influence. A positive change means that the quality factor exhibits a positive and better-than-usual trend under the influence of deviation characteristics. For example, the content of a certain aroma component increases significantly, or the intensity of a certain bitterness decreases significantly.

[0084] Furthermore, based on these quality factors exhibiting positive changes, the system dynamically generates temporary innovation quality discrimination rules. These rules aim to provide a flexible judgment mechanism for currently unknown process conditions. Their core lies in including a priority judgment logic for quality factors showing positive changes, assigning these positively changing quality factors higher judgment priority or weight. For example, if a specific aroma factor is found to have significantly improved, even if other factors show little change, it may be prioritized as an innovative quality. Finally, based on these temporary innovation quality discrimination rules, a judgment is made as to whether the deviation characteristics represent innovative quality.

[0085] In some preferred embodiments, assuming a batch of tea is produced under a novel fermentation process, the imaging data is analyzed and a deviation feature is identified. This deviation feature has a significant positive impact on the floral aroma quality factor and sweetness quality factor of the tea, but has no significant impact on the bitterness quality factor.

[0086] Specifically, the system first obtains the processing parameters of the new fermentation process.

[0087] As a specific implementation method, if a set of quality factor weights matching the novel fermentation process exists in the preset process-quality association database—for example, if the database indicates that under similar fermentation processes, the weight of the floral aroma quality factor is 0.6, the weight of the sweetness quality factor is 0.3, and the weight of the bitterness quality factor is 0.1—the system will weight and summarize the influence of the deviation feature on each quality factor according to these weights to obtain an innovative quality judgment value. For example, if the deviation feature causes the floral aroma influence to be 0.8, the sweetness influence to be 0.7, and the bitterness influence to be 0.1, then the innovative quality judgment value = 0.8 * 0.6 + 0.7 * 0.3 + 0.1 * 0.1 = 0.48 + 0.21 + 0.01 = 0.7. If the preset innovative quality discrimination rule stipulates that a judgment value greater than 0.6 indicates innovative quality, then the deviation feature is judged to represent innovative quality.

[0088] For example, if the pre-defined process-quality association database does not contain a set of quality factor weights matching the novel fermentation process, the system will identify quality factors showing positive changes—namely, floral aroma and sweetness—based on the degree of influence of the deviation feature on the quality factors. Subsequently, the system will dynamically generate a temporary innovative quality discrimination rule. This rule may include the logic that if either floral aroma or sweetness shows a significant positive change, it will be preferentially judged as an innovative quality. According to this temporary rule, since both floral aroma and sweetness show significant positive changes, this deviation feature will be judged as representing innovative quality.

[0089] By judging the two scenarios, the solution proposed in this application can flexibly and accurately identify the innovative qualities brought about by the new process, and can make effective judgments even in the absence of historical data support.

[0090] In another embodiment of this application, it is further proposed that after determining whether the deviation feature represents innovation quality, the priority judgment logic of the temporary innovation quality discrimination rule is adaptively optimized. The adaptive optimization includes: S5646-7: Record the deviation characteristics that are determined by the temporary innovative quality discrimination rules to represent innovative quality in the judgment results of each batch of tea as innovative deviation characteristics, and record the degree of influence of the quality factors corresponding to the innovative deviation characteristics; S5646-8: Based on the judgment results of multiple batches of tea, obtain the verification results used to characterize the accuracy of innovation deviation feature identification, and determine the accuracy of innovation deviation feature identification based on the verification results; S5646-9: When the identification accuracy has a continuous deviation from the preset accuracy threshold for a preset number of consecutive batches, analyze the combination of quality factors that cause the continuous deviation and the degree of their influence. S5646-10: Based on the persistence deviation, and combined with the adjustment direction and adjustment range of the priority judgment weights determined by the combination of quality factors and their degree of influence, the priority judgment weights corresponding to the combination of quality factors in the temporary innovation quality discrimination rules are adjusted. S5646-11: Apply the adjusted priority judgment weights and continuously update the recognition accuracy to achieve adaptive optimization of the priority judgment logic.

[0091] Specifically, adaptive optimization refers to an automated, continuous learning process in which the system dynamically adjusts its internal parameters, especially the priority judgment weights in temporary innovation quality discrimination rules, based on its performance in practical applications, in order to continuously improve the accuracy and reliability of innovation quality identification.

[0092] The record includes the determination of each batch of tea, the deviation characteristics that are identified as representing innovative quality by the temporary innovative quality discrimination rules, and the degree of influence of the quality factors corresponding to the innovative deviation characteristics. The purpose is to accumulate traceable optimization data. The record may include the description of the innovative deviation characteristics, their position in the structural information, the intensity changes, and the degree of influence of the innovative deviation characteristics on the quality factors such as the chemical composition, physical structure and sensory evaluation of the tea.

[0093] In practical applications, based on the judgment results of multiple batches of tea, a verification result is obtained to characterize the accuracy of identifying innovative deviation features. The accuracy of identifying innovative deviation features is determined based on the verification result. The verification result can come from human expert review, market feedback data, consumer evaluation, or comparative analysis with known high-quality tea characteristics. When the rule determines that it is an innovative deviation feature and the verification confirms that it has a positive quality improvement effect, it is counted as one accurate identification.

[0094] When the recognition accuracy deviates from the preset accuracy threshold for a preset number of consecutive batches, the system triggers an optimization mechanism. The preset number of batches can be set to, for example, 3 to 5 batches, and the preset accuracy threshold can be set to, for example, 90% recognition accuracy. The preset number of consecutive deviations is used to characterize the situation where the recognition accuracy is lower than the preset accuracy threshold multiple times in a row, so as to avoid false triggering caused by a single fluctuation.

[0095] Furthermore, when a persistent deviation is detected, the system will analyze the combination of quality factors that cause the persistent deviation and the degree of their impact in order to locate the source of misjudgment or omission; for example, it may identify that a certain innovation deviation feature is accompanied by a slight bitter taste under certain processing conditions, and the current provisional innovation quality discrimination rules fail to fully reflect this negative impact.

[0096] Therefore, based on the persistence bias, and combined with the adjustment direction and magnitude of the priority judgment weights determined by the combination of quality factors and their degree of influence, the priority judgment weights corresponding to the combination of quality factors in the temporary innovation quality judgment rules are adjusted. The adjustment direction includes increasing or decreasing the weight of a certain quality factor, and the adjustment magnitude is determined by its contribution to the persistence bias. For example, if bitterness is a key factor, then the negative weight of bitterness in the judgment logic is increased.

[0097] Finally, the adjusted priority judgment weights are applied, and the recognition accuracy is continuously updated to achieve adaptive optimization of the priority judgment logic; that is, the new weights are used to perform subsequent judgments, and the closed-loop evaluation is continued through the review results, triggering the analysis-adjustment-application iteration again when necessary.

[0098] The solution proposed in this application continuously monitors the accuracy of the identification of temporary innovation quality judgment rules, and analyzes the combination of quality factors that cause the persistent deviation and the degree of their influence when persistent deviation occurs. This allows for targeted adjustment of the priority judgment weights, enabling the temporary innovation quality judgment rules to learn and correct themselves from practical applications, thereby reducing misjudgments and omissions.

[0099] In another embodiment of this application, it is further proposed to analyze the combination of quality factors that lead to persistent deviations and the degree of their influence, including: S5646-91: In the batch corresponding to the persistent deviation, based on the mapping relationship, the structural characterization index corresponding to multiple quality factors is extracted from the structural information to form quality factor characterization data of multiple quality factors. The quality factor characterization data includes the spatial distribution, intensity change and correlation degree between multiple quality factors of the structural characterization index used to characterize multiple quality factors in the structural information. S5646-92: Based on the processing parameters and the quality factor characterization data, identify the potential nonlinear interactions between multiple quality factors and establish an interaction model. S5646-93: Based on the interaction model, quantify the contribution of nonlinear interactions among multiple quality factors to persistence bias; S5646-94: Based on the degree of contribution, determine the combination of key quality factors that lead to persistent deviations and their nonlinear interaction patterns, and obtain the combination of quality factors that lead to persistent deviations and their degree of influence.

[0100] Specifically, in batches corresponding to persistent deviations, the mapping relationship established in the implementation method is first used to extract structural characterization indicators corresponding to multiple quality factors (e.g., chemical composition, physical structure, sensory evaluation indicators, etc.) from the reconstructed structural information of tea leaves along multiple layers in the thickness direction. These structural characterization indicators are then integrated to form quality factor characterization data. This data not only includes the spatial distribution and intensity variations of each quality factor in the structural information but also covers the degree of interrelationship among multiple quality factors at the structural level. For example, the enrichment regions of specific chemical components in the tea cell walls, the density variations of cellulose bundles, and the correlation between these features and the overall compactness of the tea leaves can be extracted.

[0101] Furthermore, considering the significant impact of processing parameters on the interactions of quality factors, this application identifies potential nonlinear interactions among multiple quality factors based on quality factor characterization data and establishes corresponding interaction models. These interaction models characterize the synergistic or antagonistic effects between quality factors under specific processing conditions (such as baking temperature and fermentation time). For example, machine learning algorithms (such as neural networks, support vector machines, or ensemble learning models) can be used to construct interaction models to reveal how a specific ratio of polyphenols to amino acids nonlinearly affects the aroma formation of tea at a particular baking temperature.

[0102] Based on this, the established interaction model is used to quantify the contribution of nonlinear interactions among multiple quality factors to persistence bias. This quantification can be achieved through model interpretation techniques (such as SHAP values, LIME values, etc.) to assess the contribution of each nonlinear interaction term to persistence bias. For example, the quantitative contribution of a nonlinear combination of two specific quality factors (such as chlorophyll content and tea polyphenol oxidation level) to the tea color assessment bias under specific processing parameters can be calculated.

[0103] Finally, based on the quantified contribution levels, the key quality factor combinations leading to persistent deviations and their nonlinear interaction patterns are determined, thus obtaining the quality factor combinations causing persistent deviations and their degree of influence. This determination includes not only the identification of key quality factors but also their nonlinear interaction mechanisms and their specific impact on persistent deviations. For example, it can be determined that the combination of high rolling pressure and low fermentation temperature, through its specific nonlinear interaction pattern, leads to persistent deviations in the assessment of the body and flavor of tea.

[0104] The proposed solution extracts quality factor characterization data from structural information, introduces process parameters to constrain nonlinear interactions for identification and modeling, and further quantifies the contribution of nonlinear interactions to persistent deviations. This enables precise localization of the root causes of persistent deviations, thus providing a fine-grained, data-driven basis for subsequent rule optimization.

[0105] Reference Figure 2 The specific embodiments of this application also disclose a tea quality evaluation system based on image analysis, including: Imaging module 1 is used to acquire tea imaging data containing light intensity and light direction information on the tea production line, and to extract the surface features of the tea based on the tea imaging data; Calculation module 2 is used to process the tea imaging data and reconstruct the structural information of the tea leaves at multiple levels along the thickness direction. Parameter acquisition module 3 is used to acquire the processing parameters corresponding to tea leaves; The identification module 4 is used to identify the internal structural state, uniformity in the thickness direction, and internal defects of tea leaves from the structural information based on the processing parameters, so as to obtain the internal quality characteristics. Judgment module 5 is used to judge the overall quality of tea based on processing parameters, internal quality characteristics and surface characteristics, and obtain the judgment result; Output module 6 is used to output the quality grade of tea based on the judgment result, and feed the quality grade back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

[0106] The image analysis-based tea quality assessment system of this application acquires tea imaging data containing light intensity and light direction information and extracts surface features. At the same time, it processes the imaging data to reconstruct the structural information of multiple layers of tea along the thickness direction, thereby comprehensively capturing the internal and external quality characteristics of tea.

[0107] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for evaluating tea quality based on image analysis, characterized in that, include: Acquire tea imaging data containing light intensity and light direction information from the tea production line, and extract surface features of the tea based on the tea imaging data; The tea leaf imaging data is processed to reconstruct the structural information of the tea leaves at multiple layers along the thickness direction; Obtain the processing parameters for tea leaves; Based on the processing parameters, the internal structural state, thickness uniformity, and internal defects of the tea leaves are identified from the structural information to obtain internal quality characteristics. The overall quality of the tea is judged based on the processing parameters, internal quality characteristics, and surface characteristics, and a judgment result is obtained. The quality grade of the tea is output based on the judgment result, and the quality grade is fed back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

2. The tea quality evaluation method based on image analysis according to claim 1, characterized in that, The tea leaf imaging data is processed to reconstruct the structural information of the tea leaves at multiple levels along the thickness direction, including: The tea leaf imaging data is parsed into multiple subaperture views; By comparing the local differences among the multiple subaperture views, a deviation map characterizing spatial non-uniform distortion is obtained; Based on the deviation diagram, adjust the light propagation path parameters used for reconstruction; Based on the adjusted light propagation path parameters, the tea leaf imaging data is reconstructed to obtain structural information of multiple layers of tea leaves along the thickness direction.

3. The tea quality evaluation method based on image analysis according to claim 1, characterized in that, Based on the processing parameters, the internal structural state, thickness uniformity, and internal defects of the tea leaves are identified from the structural information to obtain internal quality characteristics, including: In the structural information, the tea leaves are divided into multiple local regions; Obtain multiple sets of internal features for each of the local regions; Based on the internal feature sets of multiple local regions, the overall internal feature statistical center value of tea is determined; Select at least one of the local regions as the current local region, and determine at least one local region adjacent to the current local region as the surrounding local region; Calculate the degree of difference between the internal feature set of the current local region and the internal feature set of the surrounding local regions, and the degree of deviation between the internal feature set of the current local region and the statistical center value of the overall internal features; Based on the degree of difference and the degree of deviation, the anomaly level of the current local region is quantified; Based on the degree of anomaly and a preset composite threshold, potential internal defect areas are marked; Based on the processing parameters, the threshold for judging the degree of abnormality in potential internal defect areas is adjusted, and based on the adjusted threshold, the internal structural state, uniformity in the thickness direction, and internal defects of the tea are identified to obtain the internal quality characteristics.

4. The tea quality evaluation method based on image analysis according to claim 1, characterized in that, Based on the processing parameters, internal quality characteristics, and surface characteristics, the overall quality of the tea is judged, and the judgment result is obtained, including: Based on the processing parameters, a processing parameter combination is formed, and the preset process parameter-quality standard mapping table is queried to see if there is an entry corresponding to the processing parameter combination. When there is an entry in the process parameter-quality standard mapping table that corresponds to the combination of processing process parameters, the internal quality characteristics and the surface characteristics are matched and judged according to the quality standard corresponding to the entry to obtain the judgment result; When there is no entry in the process parameter-quality standard mapping table that corresponds to the combination of processing process parameters, analyze the mutual influence relationship between the internal quality characteristics and the surface characteristics. Based on the mutual influence relationship, a temporary quality judgment rule is constructed, and the weights of the internal quality features and the surface features in the overall quality judgment are adjusted based on the mutual influence relationship. Based on the adjusted weights and temporary quality judgment rules, a preliminary judgment is made on the overall quality of the tea, and a preliminary judgment result is obtained. The preliminary judgment results are compared with the characteristic distribution of historical high-quality tea to confirm the effectiveness of the provisional quality judgment rule and obtain the effectiveness confirmation result. Based on the validity confirmation results, the overall quality of the tea is judged, and the judgment result is obtained.

5. The tea quality evaluation method based on image analysis according to claim 4, characterized in that, The preliminary judgment results are compared with the characteristic distribution of historical high-quality tea to confirm the effectiveness of the provisional quality judgment rule, and the effectiveness confirmation results are obtained, including: Obtain the feature set corresponding to the preliminary judgment result, wherein the feature set includes the internal quality feature and the surface feature corresponding to the preliminary judgment result; Based on the processing parameters, the characteristic distribution of the historical high-quality tea is adjusted to obtain a dynamic reference distribution that matches the processing parameters. Compare the similarity between the feature set and the dynamic reference distribution to identify deviation features in the feature set that deviate from the dynamic reference distribution; Analyze the potential impact of the deviation characteristics on tea quality and determine whether the deviation characteristics represent innovative quality; When the deviation feature is determined to represent innovative quality, the validity confirmation level of the temporary quality judgment rule is increased; Based on the validity confirmation level, the validity of the temporary quality judgment rule is confirmed, and the validity confirmation result is obtained.

6. The tea quality evaluation method based on image analysis according to claim 5, characterized in that, Analyze the potential impact of the deviation characteristics on tea quality, and determine whether the deviation characteristics represent innovative quality, including: Obtain multi-dimensional data of the deviation feature, including the spatial distribution, intensity variation, and correlation degree of the deviation feature with surrounding structural elements in the structural information; Acquire quality factor data related to processing parameters, including the chemical composition, physical structure, and sensory evaluation indicators of tea leaves; Based on historical data, correlation analysis is performed between multiple dimensions of data and the quality factor data to establish a mapping relationship between multiple dimensions of data and the quality factor data. The historical data is a set of sample data formed during the production and quality control of historical tea batches. The mapping relationship is adjusted according to the processing parameters. Based on the adjusted mapping relationship, the degree of influence of the deviation feature on multiple quality factors in the quality factor data is calculated; Based on the degree of influence and the preset innovation quality discrimination rules, determine whether the deviation feature represents innovation quality.

7. The tea quality evaluation method based on image analysis according to claim 6, characterized in that, Based on the degree of influence and the preset innovation quality discrimination rules, determine whether the deviation feature represents innovation quality, including: Based on the processing parameters, retrieve a set of quality factor weights that match the processing parameters from a preset process-quality association database; When the process-quality association database contains a set of quality factor weights that match the processing parameters, the degree of influence is weighted and summarized based on the set of quality factor weights to obtain an innovation quality judgment value. Based on the preset innovation quality discrimination rules and the innovation quality judgment value, it is determined whether the deviation feature represents innovation quality; When there is no set of quality factor weights that match the processing parameters in the process-quality association database, the quality factors that show positive changes among multiple quality factors are identified based on the degree of influence. Temporary innovation quality discrimination rules are dynamically generated based on positively changing quality factors, and the temporary innovation quality discrimination rules include priority judgment logic for positively changing quality factors. Based on the aforementioned temporary innovation quality discrimination rules, determine whether the deviation characteristics represent innovation quality.

8. The tea quality evaluation method based on image analysis according to claim 7, characterized in that, After determining whether the deviation feature represents innovation quality, the method further includes adaptively optimizing the priority judgment logic of the temporary innovation quality discrimination rule. This adaptive optimization includes: Record the deviation characteristics that are determined by the temporary innovative quality discrimination rule to represent innovative quality in the judgment results of each batch of tea as innovative deviation characteristics, and record the degree of influence of the quality factors corresponding to the innovative deviation characteristics; Based on the judgment results of multiple batches of tea, a verification result is obtained to characterize the accuracy of the identification of the innovation deviation feature, and the identification accuracy of the innovation deviation feature is determined based on the verification result; When the recognition accuracy has a continuous deviation from the preset accuracy threshold for a preset number of consecutive batches, the combination of quality factors that cause the continuous deviation and its degree of influence are analyzed. Based on the persistent deviation, and combined with the adjustment direction and magnitude of the priority judgment weights determined by the combination of quality factors and their degree of influence, the priority judgment weights corresponding to the combination of quality factors in the temporary innovation quality discrimination rule are adjusted. The adjusted priority judgment weights are applied, and the recognition accuracy is continuously updated to achieve adaptive optimization of the priority judgment logic.

9. The tea quality evaluation method based on image analysis according to claim 8, characterized in that, The analysis includes the combination of quality factors that lead to the persistent deviation and the degree of their influence, including: In the batch corresponding to the persistent deviation, structural characterization indicators corresponding to multiple quality factors are extracted from the structural information based on the mapping relationship to form quality factor characterization data of multiple quality factors. The quality factor characterization data includes the spatial distribution, intensity change and correlation degree between the structural characterization indicators used to characterize multiple quality factors in the structural information. Based on the processing parameters, the potential nonlinear interactions between multiple quality factors are identified and an interaction model is established based on the quality factor characterization data. Based on the interaction model, the contribution of the nonlinear interaction between multiple quality factors to the persistence deviation is quantified. Based on the degree of contribution, the combination of key quality factors that lead to the persistent deviation and their nonlinear interaction patterns are determined, thereby obtaining the combination of quality factors that lead to the persistent deviation and their degree of influence.

10. A tea quality evaluation system based on image analysis, characterized in that, include: An imaging module is used to acquire tea imaging data containing light intensity and light direction information from the tea production line, and to extract the surface features of the tea based on the tea imaging data; The calculation module is used to process the tea imaging data and reconstruct the structural information of the tea leaves at multiple layers along the thickness direction. The parameter acquisition module is used to acquire the processing parameters corresponding to tea leaves. The identification module is used to identify the internal structural state, uniformity in the thickness direction, and internal defects of tea leaves from the structural information based on the processing parameters, thereby obtaining internal quality characteristics. The judgment module is used to judge the overall quality of tea based on the processing parameters, the internal quality characteristics and the surface characteristics, and to obtain the judgment result; The output module is used to output the quality grade of the tea based on the judgment result, and to feed the quality grade back to the tea production line to guide subsequent grading, packaging and / or process adjustment.

Images