Digital comprehensive evaluation system for tea quality based on multi-source data fusion

Abstract

The invention relates to the technical field of tea quality evaluation, and particularly discloses a tea quality digital comprehensive evaluation system based on multi-source data fusion, and the system comprises a multi-source data collection module which is used for obtaining multi-dimensional data of tea; the multi-dimensional feature extraction module is used for extracting chemical quality parameters, physical characteristic parameters and process correlation parameters and generating a comprehensive quality embedding vector; the digital evaluation model construction module is used for generating a quality grade division rule and a quality defect traceability path; and the tea quality comprehensive evaluation module is used for generating a multi-dimensional quality evaluation report according to the quality grade division rule and the quality defect traceability path. Through three technical breakthroughs of multi-source heterogeneous data fusion, intelligent algorithm collaborative analysis and dynamic feedback control, the method has remarkable advantages in the aspects of accuracy, efficiency, traceability, industrial application value and the like, and an innovative solution is provided for standardized production and quality control of tea leaves.

Description

Technical Field

[0001] The present invention relates to the technical field of tea quality evaluation, and in particular to a digital comprehensive evaluation system for tea quality based on multi-source data fusion. Background Art

[0002] As a globally important economic crop, tea quality evaluation is directly related to product added value, market competitiveness, and consumer trust. Traditional tea quality evaluation relies primarily on manual sensory evaluation, with professional tea tasters making subjective judgments based on appearance, aroma, tea soup color, flavor, and tea leaves. However, this method has significant limitations: First, sensory evaluation is easily influenced by the taster's experience, individual sensory differences, and environmental factors, resulting in unstable results; second, the evaluation process lacks quantitative standards, making objective comparisons across regions and time difficult, hindering the digitalization and standardization of the tea trade.

[0003] In recent years, some technologies have attempted to obtain the physical and chemical indicators or appearance characteristics of tea through a single detection method (such as near-infrared spectroscopy, gas chromatography-mass spectrometry, machine vision, etc.) to assist in quality evaluation. For example, near-infrared spectroscopy can quickly detect tea moisture, tea polyphenols and other components, and machine vision can analyze the color and morphological parameters of tea. However, a single data source can only reflect the local characteristics of quality and cannot fully cover the multi-dimensional attributes of tea quality (such as chemical composition, physical properties, processing technology correlation, etc.). In addition, the isolation of data between different detection technologies leads to a lack of systematicity in the comprehensive evaluation model.

[0004] As the digital transformation of agriculture accelerates, intelligent analysis technologies that integrate multi-source heterogeneous data (such as spectral data, sensor data, process parameters, and historical quality databases) are becoming a breakthrough. However, existing systems still have shortcomings in data fusion algorithms, multi-dimensional weight allocation, and model interpretability, which restrict the accuracy of evaluation results and the breadth of their application.

[0005] Therefore, it is urgent to provide a technical solution to solve the above problems. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides a digital comprehensive evaluation system for tea quality based on multi-source data fusion, which includes the following technical solutions:

[0007] Multi-source data acquisition module, used to obtain tea near-infrared spectral data, gas chromatography-mass spectrometry data, appearance morphology image data, planting environment dynamic data and processing parameter data;

[0008] A multi-dimensional feature extraction module is configured to extract chemical quality parameters using a partial least squares regression model based on the near-infrared spectral data and the gas chromatography-mass spectrometry data; perform morphological segmentation on the appearance morphological image data using a convolutional neural network to extract physical property parameters; construct process-related parameters by mining temporal association rules based on the processing parameter data and the dynamic data of the planting environment; dynamically assign feature weights to the chemical quality parameters, the physical property parameters, and the process-related parameters based on an attention mechanism, and perform feature fusion using a cascaded deep neural network to generate a comprehensive quality embedding vector;

[0009] A digital evaluation model construction module is used to train a random forest regression model to predict sensory scores based on historical sensory review data and a physical and chemical testing database, and to generate quality grading rules and quality defect traceability paths by combining a fuzzy comprehensive evaluation algorithm with the comprehensive quality embedding vector;

[0010] The comprehensive tea quality evaluation module is used to generate a multi-dimensional quality evaluation report containing sensory scores, quality grades and defect traceability information based on the quality grade classification rules and quality defect traceability paths.

[0011] Furthermore, the multi-source data acquisition module is specifically used to:

[0012] Through spectral sensors, machine vision units, environmental monitoring terminals and processing parameter recording devices, the near-infrared spectral data, gas chromatography-mass spectrometry data, appearance morphological image data, planting environment dynamic data and processing parameter data of tea are obtained.

[0013] Furthermore, the multi-source data acquisition module is further configured to:

[0014] performing spectral noise filtering on the near-infrared spectral data and the gas chromatography-mass spectrometry data;

[0015] performing image feature normalization on the appearance morphology image data;

[0016] Performing spatiotemporal alignment on the dynamic data of the planting environment;

[0017] The processing parameter data is vectorized.

[0018] Furthermore, the physical property parameters include: color distribution uniformity, leaf integrity and texture complexity parameters; the chemical quality parameters include: quantitative indicators of tea polyphenols, amino acids and volatile components; the process-related parameters include: dynamic mapping relationship between fermentation temperature gradient, humidity gradient and changes in chemical indicators.

[0019] Furthermore, the tea quality comprehensive evaluation module is also used to:

[0020] Based on the multidimensional quality evaluation report, the processing parameter thresholds and environmental control strategies are adjusted in real time, and the heat map, quality grade distribution and process optimization suggestions of key indicators in the multidimensional quality evaluation report are dynamically displayed through an interactive interface.

[0021] Furthermore, the multi-source data acquisition module is also used to:

[0022] Dynamic filtering based on environmental interference factors is performed on the near-infrared spectral data, and the filtering function is: F(θ) represents the corrected near-infrared spectral data, S(θ) represents the original near-infrared spectral data, ΔT represents the ambient temperature fluctuation value, RH represents the relative humidity deviation coefficient, and α and β represent the environmental compensation factors.

[0023] Furthermore, an incremental learning mechanism is introduced into the digital evaluation model construction module, and the corresponding adaptive learning rate adjustment formula is: η t represents the learning rate at time t, η base is the baseline learning rate, E min is the historical minimum verification error, E t is the current verification error, is the gradient modulus of the loss function.

[0024] Furthermore, the tea quality comprehensive evaluation module is also used to:

[0025] An adversarial sample set containing noise spectra, blurred images and perturbed process parameters is constructed through generative adversarial networks to enhance model robustness.

[0026] Furthermore, the interactive interface integrates a digital twin model, maps the prediction results in the multi-dimensional quality evaluation report to the virtual processing production line, and dynamically optimizes the control strategy through a reinforcement learning algorithm.

[0027] On the other hand, a digital comprehensive evaluation method for tea quality based on multi-source data fusion is provided, as follows:

[0028] Obtain tea's near-infrared spectral data, gas chromatography-mass spectrometry data, appearance image data, planting environment dynamic data, and processing parameter data;

[0029] Based on the near-infrared spectral data and the gas chromatography-mass spectrometry data, chemical quality parameters are extracted using a partial least squares regression model; a convolutional neural network is used to perform morphological segmentation on the appearance morphological image data to extract physical property parameters; the processing parameter data and the planting environment dynamic data are combined to construct process-related parameters through temporal association rule mining; feature weights of the chemical quality parameters, the physical property parameters, and the process-related parameters are dynamically assigned according to an attention mechanism, and a cascaded deep neural network is used to perform feature fusion to generate a comprehensive quality embedding vector;

[0030] Based on historical sensory evaluation data and physical and chemical testing databases, a random forest regression model is trained to predict sensory scores. Fuzzy comprehensive evaluation algorithms are then combined with the comprehensive quality embedding vector to generate quality grading rules and quality defect traceability paths.

[0031] According to the quality grade classification rules and quality defect traceability path, a multidimensional quality evaluation report including sensory scores, quality grades and defect traceability information is generated.

[0032] Compared with the existing technology, the digital comprehensive evaluation system (method) of tea quality based on multi-source data fusion has the following beneficial effects:

[0033] 1. Comprehensive and Accurate Data Collection and Preprocessing: Utilizing spectral sensors, machine vision units, environmental monitoring terminals, and processing parameter recording devices, the system comprehensively acquires tea leaves' near-infrared spectral data, gas chromatography-mass spectrometry data, morphological image data, dynamic data on the growing environment, and processing parameter data. This multi-source data integration provides a solid foundation for comprehensive tea quality evaluation. Scientific preprocessing of the collected data, including spectral noise filtering, image feature normalization, spatiotemporal alignment, and vectorization, effectively removes noise and redundant information, improving data accuracy and reliability.

[0034] 2. Deep Fusion Feature Extraction and Weight Assignment: A partial least squares regression model is used to extract physical property parameters, such as color uniformity, leaf integrity, and texture complexity, from near-infrared spectral data and gas chromatography-mass spectrometry data. A convolutional neural network is used to perform morphological segmentation on the appearance morphological image data and extract chemical quality parameters, such as quantitative indicators of tea polyphenols, amino acids, and volatile components. Process-related parameters, such as the dynamic mapping relationship between fermentation temperature gradients, humidity gradients, and changes in chemical indicators, are constructed through temporal association rule mining, combining processing parameter data with dynamic data on the growing environment. Feature weights for chemical quality parameters, physical property parameters, and process-related parameters are dynamically assigned based on the attention mechanism. A cascaded deep neural network is used for feature fusion to generate a comprehensive quality embedding vector. This dynamic weight assignment mechanism ensures the flexibility and adaptability of the evaluation system, enabling it to accurately reflect the overall quality of tea based on the variations in different tea varieties and processing conditions.

[0035] 3. Construction and Application of Intelligent Evaluation Models: Based on historical sensory evaluation data and a physical and chemical testing database, a random forest regression model is trained to predict sensory scores. Fuzzy comprehensive evaluation algorithms and comprehensive quality embedding vectors are then combined to generate quality grading rules and quality defect traceability paths. This evaluation model, based on big data and machine learning, improves the accuracy and objectivity of tea quality evaluation. The comprehensive tea quality evaluation module can generate multidimensional quality evaluation reports in real time, including sensory scores, quality grades, and defect traceability information. It also dynamically displays heat maps of key indicators, quality grade distributions, and process optimization suggestions through an interactive interface. This real-time feedback mechanism helps processing companies promptly adjust processing parameters and environmental control strategies, improving tea processing quality and efficiency.

[0036] 4. System robustness and adaptability: Dynamic filtering based on environmental interference factors is implemented on near-infrared spectral data, effectively removing the impact of external factors such as ambient temperature and humidity on the spectral data, improving data stability and reliability. An incremental learning mechanism is introduced into the digital evaluation model construction module. The learning rate is dynamically adjusted through an adaptive learning rate adjustment formula, enabling the model to continuously learn and optimize as new data is continuously added, improving the model's adaptability and generalization capabilities. A generative adversarial network is used to construct an adversarial sample set containing noisy spectra, blurred images, and perturbed process parameters to enhance the model's robustness. This adversarial training method makes the model more robust and reliable in the face of abnormal data and noise.

[0037] 5. Digital Twin and Intelligent Optimization: The interactive interface integrates a digital twin model, mapping the predicted results in the multi-dimensional quality evaluation report to the virtual processing production line, enabling real-time interaction and synchronization between the physical and digital worlds. Through a reinforcement learning algorithm, the control strategy is dynamically optimized, automatically adjusting the processing parameters and environmental control strategies based on the predicted results and actual conditions on the virtual processing production line, thus realizing the intelligent and automated tea processing process.

[0038] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0040] Figure 1 This is a structural diagram of the digital comprehensive evaluation system for tea quality based on multi-source data fusion. DETAILED DESCRIPTION

[0041] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Figure 1 FIG. 1 shows a structural diagram of an embodiment of a digital comprehensive evaluation system 100 for tea quality based on multi-source data fusion provided by the present invention. Figure 1 As shown, the system 100 includes: a multi-source data acquisition module 110, a multi-dimensional feature extraction module 120, a digital evaluation model construction module 130 and a tea quality comprehensive evaluation module 140;

[0043] The multi-source data acquisition module 110 is used to obtain tea leaves' near-infrared spectrum data, gas chromatography-mass spectrometry data, appearance morphology image data, planting environment dynamic data, and processing parameter data.

[0044] Among them, near-infrared spectral data refers to spectral reflection / absorption data in the wavelength range of 780-2500nm collected by a near-infrared spectrometer, which is used to detect the chemical composition of tea. Gas chromatography-mass spectrometry data refers to volatile organic compound composition and content data obtained by GC-MS (Gas Chromatography-Mass Spectrometry). Appearance morphological image data refers to images of tea shape, color, and texture collected by a high-resolution camera. Planting environment dynamic data refers to time-series monitoring data including temperature, humidity, and light intensity collected by environmental monitoring sensors. Processing parameter data refers to process control variables such as fermentation time, temperature, and rolling intensity collected by process parameter recorders or sensors. Specifically, for example, a near-infrared spectrometer (wavelength range 900-1700nm) and an industrial camera (resolution 2448×2048) are deployed on the Pu'er tea processing line. Environmental sensors (temperature and humidity sampling frequency 1Hz) collect data, and a PLC system records parameters such as withering temperature (150-180°C) and rolling pressure (50-80N). This technical solution enables the integrated analysis of multi-dimensional data on tea chemistry, physics, and process, improving the accuracy of comprehensive assessments.

[0045] The multi-dimensional feature extraction module 120 is used to extract chemical quality parameters based on the near-infrared spectral data and the gas chromatography-mass spectrometry data through a partial least squares regression model; use a convolutional neural network to perform morphological segmentation on the appearance morphological image data to extract physical characteristic parameters; combine the processing parameter data with the dynamic data of the planting environment to construct process-related parameters through time series association rule mining; dynamically allocate feature weights of the chemical quality parameters, the physical characteristic parameters and the process-related parameters according to the attention mechanism, use a cascaded deep neural network to perform feature fusion, and generate a comprehensive quality embedding vector.

[0046] Among them, the chemical quality parameters include but are not limited to: color distribution uniformity, leaf integrity and texture complexity parameters; the physical property parameters include but are not limited to: quantitative indicators of tea polyphenols, amino acids and volatile components; the process-related parameters include but are not limited to: dynamic mapping relationship between fermentation temperature gradient, humidity gradient and changes in chemical indicators. It should be noted that process-related parameters also include: withering, killing green, etc., which are not restricted here. Color distribution uniformity is: the standard deviation of image area hue calculated by HSV color space (threshold <15); leaf integrity is: the proportion of intact leaf area after morphological opening operation (threshold>85%); fermentation temperature gradient is: the correlation coefficient between the temperature change rate (unit ℃ / h) during fermentation and the tea polyphenol content.

[0047] Among them, the near-infrared spectral data is used as the input data of the partial least squares regression model, and the chemical quality parameters corresponding to the gas chromatography-mass spectrometry data are used as the output data of the partial least squares regression model. After the partial least squares regression model is trained using the training set and the trained partial least squares regression model is obtained, the currently acquired near-infrared spectral data is input into the partial least squares regression model to quickly obtain the chemical quality parameters.

[0048] Among them, the convolutional neural network adopts the U-Net architecture, and the temporal association rules adopt the TP-Apriori algorithm. For example, a rule was discovered during the black tea fermentation process: fermentation temperature gradient ≥ 2°C / h (lasting 3 hours) and ambient humidity < 60% → theaflavins content increased by 15% (confidence level 0.82).

[0049] The digital evaluation model construction module 130 is used to train a random forest regression model to predict sensory scores based on historical sensory review data and physical and chemical testing databases, and to generate quality grade classification rules and quality defect traceability paths by combining a fuzzy comprehensive evaluation algorithm with the comprehensive quality embedding vector.

[0050] The default tree depth for the random forest regression model is 8, but it can be adjusted based on actual conditions, with no limit set here. Historical sensory evaluation data includes historically acquired data on five tea-related indicators: appearance, aroma, flavor, tea soup color, and tea leaf base. The physical and chemical testing database includes historically acquired near-infrared spectral data, gas chromatography-mass spectrometry data, appearance and morphological image data, planting environment dynamics data, and processing parameter data.

[0051] The tea quality comprehensive evaluation module 140 is used to generate a multi-dimensional quality evaluation report including sensory scores, quality grades and defect traceability information according to the quality grade classification rules and quality defect traceability paths.

[0052] For example, the sensory score in the multi-dimensional quality evaluation report of a certain tea is 92.4, the quality grade is A, and the defect traceability information is: ① The rolling pressure exceeds the limit (+12%); ② The fermentation humidity drops sharply (-22%).

[0053] In an optional manner, the multi-source data acquisition module 110 is specifically configured to:

[0054] Through spectral sensors, machine vision units, environmental monitoring terminals and processing parameter recording devices, the near-infrared spectral data, gas chromatography-mass spectrometry data, appearance morphological image data, planting environment dynamic data and processing parameter data of tea are obtained.

[0055] In an optional manner, the multi-source data acquisition module 110 is further configured to:

[0056] performing spectral noise filtering on the near-infrared spectral data and the gas chromatography-mass spectrometry data;

[0057] performing image feature normalization on the appearance morphology image data;

[0058] Performing spatiotemporal alignment on the dynamic data of the planting environment;

[0059] The processing parameter data is vectorized.

[0060] In an optional manner, the tea quality comprehensive evaluation module 140 is further configured to:

[0061] Based on the multidimensional quality evaluation report, the processing parameter thresholds and environmental control strategies are adjusted in real time, and the heat map, quality grade distribution and process optimization suggestions of key indicators in the multidimensional quality evaluation report are dynamically displayed through an interactive interface.

[0062] The heat map is a spatial distribution diagram of key indicators generated based on kernel density estimation (bandwidth 0.5). Process optimization suggestions: Parameter adjustment directions are calculated using gradient descent (learning rate 0.01). For example, if the withering temperature fluctuation exceeds ±3°C, the system automatically generates PID control parameter adjustment suggestions (such as adjusting the proportional coefficient Kp from 1.2 to 1.5).

[0063] In an optional manner, the multi-source data acquisition module 110 is further configured to:

[0064] Dynamic filtering based on environmental interference factors is performed on the near-infrared spectral data, and the filtering function is: F(θ) represents the corrected near-infrared spectral data, S(θ) represents the original near-infrared spectral data, ΔT represents the ambient temperature fluctuation value, RH represents the relative humidity deviation coefficient, and α and β represent the environmental compensation factors.

[0065] In an optional manner, an incremental learning mechanism is introduced into the digital evaluation model construction module 130, and the corresponding adaptive learning rate adjustment formula is: η t represents the learning rate at time t, η base is the baseline learning rate, E min is the historical minimum verification error, E t is the current verification error, is the gradient modulus of the loss function.

[0066] In an optional manner, the tea quality comprehensive evaluation module 140 is further configured to:

[0067] An adversarial sample set containing noise spectra, blurred images and perturbed process parameters is constructed through generative adversarial networks to enhance model robustness.

[0068] The noise spectrum is simulated device anomaly data with Gaussian noise (μ = 0, σ = 0.1) added, and the blurred image is a degraded image generated by Gaussian filtering (kernel size 15×15, σ = 5). For example, when generating adversarial examples, a ±3nm wavelength offset is added to the spectral data, and random occlusion (10%-20% area) is added to the image data.

[0069] In an optional manner, the interactive interface integrates a digital twin model, maps the prediction results in the multi-dimensional quality evaluation report to the virtual processing production line, and dynamically optimizes the control strategy through a reinforcement learning algorithm.

[0070] The digital twin model is a virtual processing line built in Unity3D (with a time synchronization error of <50ms). The reinforcement learning algorithm is the PPO algorithm (with a clipping range of 0.2 and a discount factor γ of 0.99). For example, the digital twin model simulates a gradual increase in fermentation temperature from 28°C to 32°C (with a step size of 1°C / hour), and Q-learning is used to select the optimal temperature ramp curve.

[0071] The technical solution of this embodiment has the following beneficial effects compared to the prior art:

[0072] 1. Comprehensive and Accurate Data Collection and Preprocessing: Utilizing spectral sensors, machine vision units, environmental monitoring terminals, and processing parameter recording devices, the system comprehensively acquires tea leaves' near-infrared spectral data, gas chromatography-mass spectrometry data, morphological image data, dynamic data on the growing environment, and processing parameter data. This multi-source data integration provides a solid foundation for comprehensive tea quality evaluation. Scientific preprocessing of the collected data, including spectral noise filtering, image feature normalization, spatiotemporal alignment, and vectorization, effectively removes noise and redundant information, improving data accuracy and reliability.

[0073] 2. Deep Fusion Feature Extraction and Weight Assignment: A partial least squares regression model is used to extract physical property parameters, such as color uniformity, leaf integrity, and texture complexity, from near-infrared spectral data and gas chromatography-mass spectrometry data. A convolutional neural network is used to perform morphological segmentation on the appearance morphological image data and extract chemical quality parameters, such as quantitative indicators of tea polyphenols, amino acids, and volatile components. Process-related parameters, such as the dynamic mapping relationship between fermentation temperature gradients, humidity gradients, and changes in chemical indicators, are constructed through temporal association rule mining, combining processing parameter data with dynamic data on the growing environment. Feature weights for chemical quality parameters, physical property parameters, and process-related parameters are dynamically assigned based on the attention mechanism. A cascaded deep neural network is used for feature fusion to generate a comprehensive quality embedding vector. This dynamic weight assignment mechanism ensures the flexibility and adaptability of the evaluation system, enabling it to accurately reflect the overall quality of tea based on the variations in different tea varieties and processing conditions.

[0074] 3. Construction and Application of Intelligent Evaluation Models: Based on historical sensory evaluation data and a physical and chemical testing database, a random forest regression model is trained to predict sensory scores. Fuzzy comprehensive evaluation algorithms and comprehensive quality embedding vectors are then combined to generate quality grading rules and quality defect traceability paths. This evaluation model, based on big data and machine learning, improves the accuracy and objectivity of tea quality evaluation. The comprehensive tea quality evaluation module can generate multidimensional quality evaluation reports in real time, including sensory scores, quality grades, and defect traceability information. It also dynamically displays heat maps of key indicators, quality grade distributions, and process optimization suggestions through an interactive interface. This real-time feedback mechanism helps processing companies promptly adjust processing parameters and environmental control strategies, improving tea processing quality and efficiency.

[0075] 4. System robustness and adaptability: Dynamic filtering based on environmental interference factors is implemented on near-infrared spectral data, effectively removing the impact of external factors such as ambient temperature and humidity on the spectral data, improving data stability and reliability. An incremental learning mechanism is introduced into the digital evaluation model construction module. The learning rate is dynamically adjusted through an adaptive learning rate adjustment formula, enabling the model to continuously learn and optimize as new data is continuously added, improving the model's adaptability and generalization capabilities. A generative adversarial network is used to construct an adversarial sample set containing noisy spectra, blurred images, and perturbed process parameters to enhance the model's robustness. This adversarial training method makes the model more robust and reliable in the face of abnormal data and noise.

[0076] 5. Digital Twin and Intelligent Optimization: The interactive interface integrates a digital twin model, mapping the predicted results in the multi-dimensional quality evaluation report to the virtual processing production line, enabling real-time interaction and synchronization between the physical and digital worlds. Through a reinforcement learning algorithm, the control strategy is dynamically optimized, automatically adjusting the processing parameters and environmental control strategies based on the predicted results and actual conditions on the virtual processing production line, thus realizing the intelligent and automated tea processing process.

[0077] The technical solution of this embodiment achieves the transformation and upgrading of tea quality evaluation from "experience-driven" to "data-driven" through three major technological breakthroughs: multi-source heterogeneous data fusion, intelligent algorithm collaborative analysis, and dynamic feedback control. It has significant advantages in accuracy, efficiency, traceability and industrial application value, and provides an innovative solution for standardized production and quality control of tea.

[0078] In another aspect, the present invention provides an embodiment of a digital comprehensive evaluation method for tea quality based on multi-source data fusion, the method comprising the following steps:

[0079] S1. Acquire tea leaves' near-infrared spectral data, gas chromatography-mass spectrometry data, appearance image data, planting environment dynamic data, and processing parameter data;

[0080] S2. Extract chemical quality parameters using a partial least squares regression model based on the near-infrared spectral data and the gas chromatography-mass spectrometry data; perform morphological segmentation on the appearance morphological image data using a convolutional neural network to extract physical property parameters; construct process-related parameters by mining temporal association rules based on the processing parameter data and the dynamic data of the planting environment; dynamically assign feature weights to the chemical quality parameters, the physical property parameters, and the process-related parameters according to an attention mechanism, perform feature fusion using a cascaded deep neural network, and generate a comprehensive quality embedding vector;

[0081] S3. Based on historical sensory evaluation data and a physical and chemical testing database, a random forest regression model is trained to predict sensory scores. The model then combines a fuzzy comprehensive evaluation algorithm with the comprehensive quality embedding vector to generate quality grading rules and quality defect traceability paths.

[0082] S4. Generate a multidimensional quality evaluation report including sensory scores, quality grades, and defect traceability information based on the quality grade classification rules and quality defect traceability paths.

[0083] It should be noted that the beneficial effects of the digital comprehensive evaluation method for tea quality based on multi-source data fusion provided in the above embodiment are the same as the beneficial effects of the digital comprehensive evaluation system for tea quality based on multi-source data fusion 100, which will not be repeated here.

[0084] The above description is merely a preferred embodiment of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present invention.

[0085] It should be noted that the terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and to define a specific order or precedence. Where appropriate, the order used for similar objects may be interchanged, such that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0086] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A digital comprehensive evaluation system for tea quality based on multi-source data fusion, characterized by: include: Multi-source data acquisition module, used to obtain tea near-infrared spectral data, gas chromatography-mass spectrometry data, appearance morphology image data, planting environment dynamic data and processing parameter data; A multi-dimensional feature extraction module is configured to extract chemical quality parameters using a partial least squares regression model based on the near-infrared spectral data and the gas chromatography-mass spectrometry data; perform morphological segmentation on the appearance morphological image data using a convolutional neural network to extract physical property parameters; construct process-related parameters by mining temporal association rules based on the processing parameter data and the dynamic data of the planting environment; dynamically assign feature weights to the chemical quality parameters, the physical property parameters, and the process-related parameters based on an attention mechanism, and perform feature fusion using a cascaded deep neural network to generate a comprehensive quality embedding vector; A digital evaluation model construction module is used to train a random forest regression model to predict sensory scores based on historical sensory review data and a physical and chemical testing database, and to generate quality grading rules and quality defect traceability paths by combining a fuzzy comprehensive evaluation algorithm with the comprehensive quality embedding vector; The comprehensive tea quality evaluation module is used to generate a multi-dimensional quality evaluation report containing sensory scores, quality grades and defect traceability information based on the quality grade classification rules and quality defect traceability paths.

2. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The multi-source data acquisition module is specifically used for: Through spectral sensors, machine vision units, environmental monitoring terminals and processing parameter recording devices, the near-infrared spectral data, gas chromatography-mass spectrometry data, appearance morphological image data, planting environment dynamic data and processing parameter data of tea are obtained.

3. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 2 is characterized in that: The multi-source data acquisition module is further specifically used for: performing spectral noise filtering on the near-infrared spectral data and the gas chromatography-mass spectrometry data; performing image feature normalization on the appearance morphology image data; Performing spatiotemporal alignment on the dynamic data of the planting environment; The processing parameter data is vectorized.

4. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The physical property parameters include: color distribution uniformity, leaf integrity and texture complexity parameters; the chemical quality parameters include: quantitative indicators of tea polyphenols, amino acids and volatile components; the process-related parameters include: dynamic mapping relationship between fermentation temperature gradient, humidity gradient and changes in chemical indicators.

5. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The tea quality comprehensive evaluation module is also used for: Based on the multidimensional quality evaluation report, the processing parameter thresholds and environmental control strategies are adjusted in real time, and the heat map, quality grade distribution and process optimization suggestions of key indicators in the multidimensional quality evaluation report are dynamically displayed through an interactive interface.

6. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 2 is characterized in that: The multi-source data acquisition module is also used for: Dynamic filtering based on environmental interference factors is performed on the near-infrared spectral data, and the filtering function is: F(θ) represents the corrected near-infrared spectral data, S(θ) represents the original near-infrared spectral data, ΔT represents the ambient temperature fluctuation value, RH represents the relative humidity deviation coefficient, and α and β represent the environmental compensation factors.

7. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The incremental learning mechanism is introduced into the digital evaluation model construction module, and the corresponding adaptive learning rate adjustment formula is: η t represents the learning rate at time t, η base is the baseline learning rate, E min is the historical minimum verification error, E t is the current verification error, is the gradient modulus of the loss function.

8. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The tea quality comprehensive evaluation module is also used for: An adversarial sample set containing noise spectra, blurred images and perturbed process parameters is constructed through generative adversarial networks to enhance model robustness.

9. The digital comprehensive evaluation system for tea quality based on multi-source data fusion according to claim 1 is characterized in that: The interactive interface integrates a digital twin model, maps the prediction results in the multi-dimensional quality evaluation report to the virtual processing production line, and dynamically optimizes the control strategy through a reinforcement learning algorithm.

10. A digital comprehensive evaluation method for tea quality based on multi-source data fusion, characterized in that: include: Obtain tea's near-infrared spectral data, gas chromatography-mass spectrometry data, appearance image data, planting environment dynamic data, and processing parameter data; Based on the near-infrared spectral data and the gas chromatography-mass spectrometry data, chemical quality parameters are extracted using a partial least squares regression model; a convolutional neural network is used to perform morphological segmentation on the appearance morphological image data to extract physical property parameters; the processing parameter data and the planting environment dynamic data are combined to construct process-related parameters through temporal association rule mining; feature weights of the chemical quality parameters, the physical property parameters, and the process-related parameters are dynamically assigned according to an attention mechanism, and a cascaded deep neural network is used to perform feature fusion to generate a comprehensive quality embedding vector; Based on historical sensory evaluation data and physical and chemical testing databases, a random forest regression model is trained to predict sensory scores. Fuzzy comprehensive evaluation algorithms are then combined with the comprehensive quality embedding vector to generate quality grading rules and quality defect traceability paths. According to the quality grade classification rules and quality defect traceability path, a multidimensional quality evaluation report including sensory scores, quality grades and defect traceability information is generated.

Images