Tea leaf processing auxiliary method and system based on intelligent decision

Abstract

The invention discloses a tea processing auxiliary method and system based on intelligent decision making, and the method comprises the steps: carrying out the tea processing monitoring, obtaining tea processing monitoring data, carrying out the preprocessing of the obtained tea processing monitoring data, and obtaining the cross-modal tea processing monitoring information; analyzing a cross-modal association relationship of each monitoring data based on the cross-modal tea processing monitoring information, and constructing a multi-modal fusion feature to obtain a multi-modal tea processing fusion feature; constructing a tea leaf processing quality sensing model, inputting the multi-modal tea leaf processing fusion features into the trained tea leaf processing quality sensing model to sense the processing quality of the tea leaves of the current processing batch, and obtaining tea leaf processing quality sensing information; and judging whether the tea quality of the current processing batch meets the expectation or not based on the tea processing quality sensing information, and if not, optimizing the current processing scheme to carry out tea processing assistance. The sensing ability of the tea processing process is enhanced; and the final tea processing quality is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tea processing, and particularly to a tea processing assistance method and system based on intelligent decision-making. Background Art

[0002] In the traditional tea processing process, the regulation of key parameters such as the fixation temperature, rolling pressure, and fermentation humidity depends on manual experience, which has inherent defects such as low control accuracy, poor quality stability, and high energy consumption. Most of the existing automated equipment adopts a preset process curve or single-sensor feedback control, and it is difficult to adapt to the differences in the physical and chemical characteristics of fresh tea leaves and environmental dynamic disturbances, resulting in problems such as insufficient conversion of the internal components of tea leaves and unqualified morphological characteristics during the processing process. In recent years, although there have been research attempts to introduce machine vision to detect the color change of tea leaves or analyze the moisture content through near-infrared spectroscopy, each sensing data is processed in isolation, lacking dynamic correlation modeling between multi-modal feature fusion and process parameters, and it is impossible to construct a full-link decision-making system covering "raw material characteristics - processing status - finished product quality".

[0003] In addition, the existing technology lacks the ability of in-depth fusion and intelligent analysis of multi-source data (such as spectroscopy, images, environmental sensing). Most systems only rely on single-sensor feedback for local adjustment, and it is difficult to globally optimize the collaborative efficiency of the processing link. For example, in the withering and shaking process of oolong tea, although traditional equipment can maintain constant temperature and humidity, it cannot dynamically adjust the shaking frequency according to the degree of red edge formation and aroma release characteristics of tea leaves, which easily leads to insufficient or excessive fermentation. Such limitations make tea processing still face bottlenecks such as large quality fluctuations, high energy consumption, and low standardization, and there is an urgent need to achieve adaptive optimization of process parameters and precise control of the entire process through intelligent decision-making technology. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides a tea processing assistance method and system based on intelligent decision-making, and an important purpose thereof is to enhance the perception ability of the tea processing process and improve the final tea processing quality.

[0005] To achieve the above object, the first aspect of the present invention provides a tea processing assistance method based on intelligent decision-making, including: Performing tea processing monitoring on a target processing device to obtain tea processing monitoring data, and preprocessing the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; Analyzing the cross-modal correlation relationship of each monitoring data based on the cross-modal tea processing monitoring information, and constructing multi-modal fusion features to obtain multi-modal tea processing fusion features; Constructing a tea processing quality perception model, and inputting the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch to obtain tea processing quality perception information; Based on the tea processing quality perception information, determine whether the quality of the tea in the current processing batch meets the expectations. If not, optimize the current processing plan to assist in tea processing.

[0006] In this solution, the tea processing monitoring data of the target processing equipment is obtained through tea processing monitoring, and the obtained tea processing monitoring data is preprocessed to obtain cross-modal tea processing monitoring information, specifically including: When the target processing equipment is performing tea processing, use the installed sensor array to monitor the tea processing of the target processing equipment to obtain tea processing monitoring data. The tea processing monitoring data includes processing environment monitoring data, processing equipment operation data, and processed tea monitoring data; Perform wavelet packet decomposition on the tea processing monitoring data using a preset wavelet basis function to obtain different sub-bands and calculate the energy entropy of each sub-band. Perform soft threshold filtering based on the calculated energy entropy to identify the effective feature band; Perform time series alignment processing on the tea processing monitoring data after filtering. Use median filtering to remove noise and Gaussian smoothing to eliminate motion blur. Introduce a sliding window algorithm to slide based on a preset window and calculate the mean and standard deviation of the data within the window. Combine the linear interpolation algorithm to perform data cleaning processing; Perform maximum-minimum normalization on the tea processing monitoring data after cleaning, map the continuous data to the [0,1] interval, and standardize the discrete data through binning processing to generate cross-modal tea processing monitoring information.

[0007] In this solution, analyze the cross-modal correlation relationship of each monitoring data based on the cross-modal tea processing monitoring information, and construct a multi-modal fusion feature to obtain a multi-modal tea processing fusion feature, specifically including: Obtain cross-modal tea processing monitoring information, obtain processed tea monitoring data through the cross-modal tea processing monitoring information and extract processed tea image monitoring data. Import the processed tea image monitoring data into the HSV color space to obtain the histogram distribution; Use the histogram distribution to calculate the weighted fusion of the local area hue mean and saturation variance to form the browning degree index. Subsequently, extract the contour of the largest connected region of the processed tea image through a morphological algorithm, calculate the ratio between the contour perimeter and the tea area as the curl index, and generate the first feature information in combination with the processed tea texture features; Extract near-infrared spectral data by obtaining processed tea monitoring data and calculate the first derivative to generate a near-infrared derivative spectrum. Input the near-infrared derivative spectrum into a pre-trained least squares regression model for tea chemical component inversion to generate a spectral domain feature vector of the processed tea, and obtain the second feature information; Introduce the multi-head attention mechanism, use the first feature information as the query vector, the second feature information as the key vector and value vector, calculate the scaled dot attention score for feature fusion to generate fused feature information; Extract the processing environment monitoring data and processing equipment operation data by using cross-modal tea processing monitoring information, and input them into the BiGRU network together with the fused feature information for multi-modal feature fusion to obtain the multi-modal tea processing fused feature.

[0008] In this solution, the extraction of the processing environment monitoring data and processing equipment operation data by using cross-modal tea processing monitoring information, and the input into the BiGRU network together with the fused feature information for multi-modal feature fusion specifically includes: In the time dimension, convert the data input into the BiGRU network into several time-series feature vectors, where each time-series feature vector contains the tea leaf texture details, spectral chemical components, equipment operation parameters, and processing environment parameters at the current moment; The forward propagation layer captures the immediate changes in the leaf state caused by the adjustment of the processing parameters through the update gate, and the backward uses the reset gate to filter the lagged changes in the leaf state caused by the adjustment of the processing parameters to generate time-series causal features; In the space dimension, perform spatial convolution on the data input into the BiGRU network by using the convolutional layer to obtain a convolutional feature map, and perform global average pooling on the convolutional feature map in the channel dimension to generate a spatial saliency weight vector; Obtain the time-series hidden state of the BiGRU, perform matrix multiplication with the spatial saliency weight vector, calculate the importance scores of each channel in the time-series context, and perform channel weighting on the original convolutional feature map to obtain a spatially responsive enhanced feature; Input the time-series causal features and the spatially responsive enhanced features into the fully connected layer for feature vector fusion to generate the multi-modal tea processing fused feature when the target processing equipment processes tea.

[0009] In this solution, the construction of the tea processing quality perception model, and the input of the multi-modal tea processing fused feature into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch specifically includes: Retrieve historical processing instances of different tea processing qualities based on historical data, and extract the historical tea processing features of each historical processing instance when performing tea processing according to the obtained historical processing instances to obtain historical tea processing feature information; Extract the tea processing processes corresponding to each historical processing instance, perform feature partitioning on the historical tea processing feature information to generate several tea processing feature subsets of different historical processing instances, and each tea processing feature subset corresponds to a different tea processing process; Associate the historical tea processing feature information after feature partitioning with the corresponding historical tea processing quality. Take the tea processing features as the first node and the tea processing quality as the second node, and define the tea processing process as the node label of the first node to establish a topological structure diagram; Construct a tea processing quality perception model based on the long short-term memory network and the graph neural network. Input the historical tea processing feature information after feature partitioning into the long short-term memory network to extract multi-scale time features of different processing qualities; Obtain the adjacency matrix through the established topological structure diagram, perform node aggregation and update based on the message passing mechanism to obtain the updated node representation, perform global pooling on the updated node representation of the entire graph to obtain spatial features, and fuse the multi-scale time features and spatial features to generate spatio-temporal fusion features; Input the spatio-temporal fusion features into the fully connected layer for tea processing quality perception, verify and adjust the parameters of the tea processing quality perception result through a preset validation set, and obtain a tea processing quality perception model that meets the expectations after iterative training; Obtain multi-modal tea processing fusion features, input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch, and obtain tea processing quality perception information.

[0010] In this solution, judge whether the quality of the tea in the current processing batch meets the expectations based on the tea processing quality perception information. If not, optimize the current processing plan for tea processing assistance, specifically including: Obtain the tea processing quality perception information, and generate a tea quality change perception curve of the current processing batch based on the tea processing quality perception information. The tea quality change perception curve represents the tea processing quality of the tea in the current processing batch at different timestamps; Obtain the tea quality requirements of the current processing batch, calculate the deviation between the tea processing quality corresponding to different timestamps and the tea quality requirements in combination with the tea quality change perception curve, and judge the calculated deviation with a preset tolerance range; If it is greater than the preset tolerance range, it means that the current tea processing control parameters do not match the tea quality requirements of the current processing batch, then obtain the preset processing plan of the current processing batch for processing plan optimization; Retrieve historical processing cases that meet the tea quality requirements of the current processed tea and the tea variety of the processed tea in the preset historical processing database, and extract the corresponding historical processing control plans of each historical processing case; Introduce the multi-objective grey wolf optimization algorithm. Initialize the population according to the historical processing control schemes corresponding to each historical processing case. Preset the objective function and set the constraint conditions. Calculate the objective function values of each individual in the initial grey wolf population through the objective function, and perform non-dominated sorting to divide all individuals in the population into different front levels. Perform congestion calculation on each front level to obtain the crowding distance. Select the individual with the largest crowding degree in each level as the leading wolf, and regard the remaining individuals as following wolves. Calculate the direction vector of the leading wolf in the decision space for position update. Repeat the iteration until the stop condition is met, and then output the optimal solution set. Generate several candidate processing optimization schemes based on the output optimal solution set. Conduct feasibility screening on each candidate processing optimization scheme, and select the optimal processing optimization scheme based on the feasibility screening results for tea processing assistance.

[0011] The second aspect of the present invention provides a tea processing assistance system based on intelligent decision-making. The system includes: a memory and a processor. The memory contains a program for the tea processing assistance method based on intelligent decision-making. When the program for the tea processing assistance method based on intelligent decision-making is executed by the processor, the following steps are implemented: Monitor the tea processing of the target processing equipment to obtain tea processing monitoring data, and preprocess the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information. Analyze the cross-modal correlation relationships of the monitoring data based on the cross-modal tea processing monitoring information, and construct multi-modal fusion features to obtain multi-modal tea processing fusion features. Construct a tea processing quality perception model, and input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perceive the processing quality of the tea in the current processing batch, and obtain tea processing quality perception information. Judge whether the quality of the tea in the current processing batch meets the expectation based on the tea processing quality perception information. If not, optimize the current processing scheme for tea processing assistance.

[0012] The present invention discloses a tea processing assistance method and system based on intelligent decision-making, including: monitoring tea processing to obtain tea processing monitoring data, preprocessing the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; analyzing the cross-modal correlation relationship of each monitoring data based on the cross-modal tea processing monitoring information, and constructing multi-modal fusion features to obtain multi-modal tea processing fusion features; constructing a tea processing quality perception model, inputting the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch, and obtaining tea processing quality perception information; judging whether the quality of the tea of the current processing batch meets the expectation based on the tea processing quality perception information, and if not, optimizing the current processing plan to assist in tea processing. Enhance the perception ability of the tea processing process and improve the final tea processing quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments or exemplary examples of the present invention, the following will briefly introduce the drawings required for use in the embodiments or exemplary descriptions. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the drawings shown.

[0014] Figure 1 It is a flowchart of a tea processing assistance method based on intelligent decision-making provided by an embodiment of the present invention; Figure 2 It is a flowchart of an intelligent decision-making and assistance method for a tea processing process provided by an embodiment of the present invention; Figure 3 It is a block diagram of a tea processing assistance system based on intelligent decision-making provided by an embodiment of the present invention; The realization, functional characteristics and advantages of the purpose of the present invention will be further described in conjunction with the embodiments and with reference to the drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] In order to be able to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0017] Figure 1Flowchart of a tea processing assistance method based on intelligent decision-making provided by an embodiment of the present invention; As Figure 1 shown, the present invention provides a flowchart of a tea processing assistance method based on intelligent decision-making, including: S102, monitor the tea processing of the target processing equipment to obtain tea processing monitoring data, and preprocess the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; S104, analyze the cross-modal correlation relationship of each monitoring data based on the cross-modal tea processing monitoring information, and construct a multi-modal fusion feature to obtain a multi-modal tea processing fusion feature; S106, construct a tea processing quality perception model, input the multi-modal tea processing fusion feature into the trained tea processing quality perception model to perceive the processing quality of the tea in the current processing batch, and obtain tea processing quality perception information; S108, judge whether the quality of the tea in the current processing batch meets the expectation based on the tea processing quality perception information. If not, optimize the current processing plan to assist in tea processing.

[0018] Further, in a preferred embodiment of the present invention, the step of monitoring the tea processing of the target processing equipment to obtain tea processing monitoring data and preprocessing the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information specifically includes: When the target processing equipment is performing tea processing, use the installed sensor array to monitor the tea processing of the target processing equipment to obtain tea processing monitoring data, where the tea processing monitoring data includes processing environment monitoring data, processing equipment operation data, and processed tea monitoring data; Perform wavelet packet decomposition on the tea processing monitoring data using a preset wavelet basis function to obtain different subbands and calculate the energy entropy of each subband, and perform soft threshold filtering based on the calculated energy entropy to identify effective feature bands; Perform time series alignment processing on the tea processing monitoring data after filtering, use median filtering to remove noise and Gaussian smoothing to eliminate motion blur, introduce a sliding window algorithm to slide based on a preset window and calculate the mean and standard deviation of the data within the window, and combine a linear interpolation algorithm to perform data cleaning processing; Perform maximum-minimum normalization on the tea processing monitoring data after cleaning, map continuous data to the [0,1] interval, and standardize discrete data through binning to generate cross-modal tea processing monitoring information.

[0019] It should be noted that the monitoring data of the entire tea processing process is collected in real time through the deployed multi-source sensor network. This network integrates high-precision infrared thermal imagers, multi-spectral imaging units, three-axis vibration sensors, and temperature and humidity sensing modules, and synchronously captures processing environment parameters (such as the three-dimensional temperature field distribution inside the drum, environmental humidity gradient, etc.), equipment operating states (such as motor current fluctuations, mechanical drive shaft rotation speed time-series waveforms, etc.), and tea leaf body characteristics (including leaf surface shrinkage, near-infrared spectra, etc.). Subsequently, the collected original signals are processed by adaptive wavelet packet decomposition, and specific wavelet basis functions are used to perform multi-scale frequency band division on non-stationary time-series data. By calculating the energy distribution entropy values of each sub-band, the characteristic frequency bands containing key process information are identified. For the noise-dominated sub-bands with entropy values below the threshold, dynamic soft threshold filtering is used for suppression, eliminating random interference while retaining the edge characteristics of the effective signals. The multi-source data stream after frequency domain filtering enters the spatio-temporal alignment module, and the phase deviation caused by heterogeneous device sampling is eliminated through millisecond-level timestamp matching, and a composite filtering strategy is used to process sensor-specific noise: the median filter eliminates the pulse interference of the temperature sensor, and the anisotropic Gaussian kernel compensates for the motion blur effect of the optical imaging unit in high-speed rotation scenarios, restoring the true morphological characteristics of the leaf texture.

[0020] Next, in the data cleaning stage, a dynamic sliding window mechanism is introduced to scan the data stream with a window length adaptively determined by the process time-series characteristics, and the statistics (mean and standard deviation) within the window are calculated in real time to construct a data quality assessment matrix. For the detected abnormal intervals (such as outliers or data breakpoints caused by transient sensor failures), a repair mechanism is triggered: short-term missing values are compensated by linear interpolation, and persistent anomalies activate redundant sensor data replacement. The heterogeneous data after cleaning undergoes cross-modal standardization. Continuous parameters (such as temperature, pressure, moisture content, etc.) eliminate the dimensionality differences through dynamic range normalization and are mapped to a unified numerical interval; discrete state parameters (such as equipment fault codes, process stage identifiers, etc.) are transformed into standard feature vectors by equal-probability binning encoding. Finally, cross-modal tea processing monitoring information is generated, providing a high-consistency and strongly correlated input basis for subsequent intelligent decision-making models.

[0021] Furthermore, in a preferred embodiment of the present invention, the cross-modal correlation relationships of the monitoring data are analyzed based on the cross-modal tea processing monitoring information, and multi-modal fusion features are constructed to obtain multi-modal tea processing fusion features, specifically including: Obtain the cross-modal tea processing monitoring information, obtain the monitoring data of the processed tea through the cross-modal tea processing monitoring information and extract the monitoring data of the processed tea image, and import the monitoring data of the processed tea image into the HSV color space to obtain the histogram distribution; The hue mean of the local region and the saturation variance are calculated using the histogram distribution and weighted and fused to form the browning degree index. Subsequently, the contour of the largest connected region of the processed tea image is extracted by the morphological algorithm, and the ratio between the contour perimeter and the tea leaf area is calculated as the curl index, and the first feature information is generated in combination with the texture features of the processed tea. The near-infrared spectral data are extracted from the processed tea monitoring data and the first derivative is calculated to generate the near-infrared derivative spectrum. The near-infrared derivative spectrum is input into the pre-trained least squares regression model for the inversion of tea chemical components to generate the spectral domain feature vector of the processed tea, and the second feature information is obtained. The multi-head attention mechanism is introduced, taking the first feature information as the query vector, taking the second feature information as the key vector and the value vector, and calculating the scaled dot attention score for feature fusion to generate the fusion feature information. The processing environment monitoring data and the processing equipment operation data are extracted using the cross-modal tea processing monitoring information, and are input into the BiGRU network for multi-modal feature fusion in combination with the fusion feature information to obtain the multi-modal tea processing fusion feature.

[0022] It should be noted that, first, the tea leaf image data is separated from the cross-modal monitoring information and converted to the HSV color space to analyze the distribution characteristics of hue (H) and saturation (S). A histogram is constructed by statistically calculating the pixel density in different hue intervals. Based on the histogram distribution, the weighted fusion value of the hue mean and saturation variance in the local leaf area is calculated (the weight coefficient is dynamically adjusted according to the process stage), forming a quantitative index representing the degree of enzymatic browning. At the same time, the morphological dilation-erosion algorithm is used to extract the contour of the largest connected region in the image, and the degree of leaf curling is quantified by the ratio of the contour perimeter to the projected area. Combining the texture contrast and energy values calculated by the gray-level co-occurrence matrix (GLCM), the first feature information describing the change in the appearance morphology of the tea leaves is generated. At the same time, after the extracted near-infrared spectral data is smoothed by Savitzky-Golay filtering, its first derivative spectrum is calculated to highlight the characteristic absorption peaks, and the derivative spectrum is input into a pre-trained least squares regression model (the training data is historical tea sample data) to invert the concentration distributions of key chemical components such as tea polyphenols and moisture, constituting the second feature information reflecting the internal substance transformation of the tea leaves. An association model between the appearance features and chemical components is established through the multi-head attention mechanism: the first feature information is used as the query vector (Query), and the second feature information is used as the key vector (Key) and value vector (Value). The scaled dot product similarity scores of each attention head are calculated, and the highly correlated feature nodes are screened out (such as the negative correlation between high curling degree and low tea polyphenol concentration). After weighted fusion, a fusion feature vector jointly representing the internal and external quality evolution of the tea leaves is generated. Further integrating the processing environment parameters (temperature and humidity gradients, air flow velocity) and the equipment operation status (motor load rate, drum rotation speed time series), and inputting them together with the fusion feature vector into a bidirectional gated recurrent unit (BiGRU). The fusion weight of the cross-modal information is dynamically adjusted through the update gate and reset gate, and finally a multi-modal fusion feature encoding the appearance morphology, chemical components, environmental disturbances, and equipment working conditions is output, providing a highly discriminative state representation for intelligent decision-making.

[0023] Furthermore, in a preferred embodiment of the present invention, the extraction of the processing environment monitoring data and processing equipment operation data by using the cross-modal tea processing monitoring information, and the input of the combined fusion feature information into the BiGRU network for multi-modal feature fusion specifically includes: In the time dimension, the data input into the BiGRU network is converted into a number of time series feature vectors, where each time series feature vector contains the tea leaf texture details, spectral chemical components, equipment operation parameters, and processing environment parameters at the current moment; The forward propagation layer captures the immediate changes in the leaf state caused by the adjustment of the processing parameters through the update gate, and the backward uses the reset gate to screen out the lagged changes in the leaf state caused by the adjustment of the processing parameters, generating time series causal features; In the spatial dimension, the data input into the BiGRU network is spatially convolved using a convolutional layer to obtain a convolutional feature map, and the convolutional feature map is globally averaged pooled in the channel dimension to generate a spatial significance weight vector; Obtain the temporal hidden state of BiGRU, perform matrix multiplication with the spatial significance weight vector, calculate the importance score of each channel in the temporal context, perform channel weighting on the original convolution feature map, and obtain the spatial response enhancement feature; The temporal causal features and the spatial response enhancement features are input into the fully connected layer for feature vector fusion to generate multimodal tea processing fusion features when the target processing equipment is processing tea.

[0024] It should be noted that in the multimodal feature fusion process of tea processing, the cross-modal monitoring information is first expanded into a continuous time series feature vector sequence in the time dimension. Each vector integrates the surface texture details of the tea at the current moment (such as the contrast calculated by the grayscale co-occurrence matrix), the chemical component concentration (such as the content of tea polyphenols) inverted by the near-infrared spectrum, the equipment operation parameters (such as the drum speed, the motor current waveform) and the environmental parameters (such as the temperature and humidity gradient inside the drum). The input is a bidirectional gated recurrent network (BiGRU). Its forward propagation layer dynamically adjusts the retention ratio of historical state information through the update gate, and captures the immediate impact of processing parameter adjustments on the physical and chemical properties of leaves in real time, such as the minute-level response of the change in chlorophyll degradation rate caused by the increase of 5°C in the withering temperature; the backward propagation layer uses the reset gate to screen the key nodes in the historical processing events and explore the hysteresis effect of parameter adjustment, such as the continuous impact of the cumulative effect of the rolling pressure gradient on the activity of polyphenol oxidase in the subsequent fermentation stage, thereby generating hidden state features encoding temporal causal relationships.

[0025] In spatial dimension processing, the original multi-modal data extracts spatial local features through a three-dimensional convolutional layer, generating a feature map containing leaf morphology, equipment structure texture, and thermal field distribution. Global average pooling in the channel dimension is performed on this feature map to compress spatial information and generate a weight vector representing the significance of each channel. For example, in the fixation process, the weight of the heat conduction feature channel in the high-temperature area is automatically increased. The temporal hidden state matrix output by the BiGRU is multiplied by the spatial weight vector to calculate the importance score of each feature channel in the temporal context. After the score result is normalized by Sigmoid, channel weighting is performed on the original convolutional feature map to strengthen the spatial response strongly related to the current processing stage. For example, at the end of the drying stage, the feature response intensity of the leaf shrinkage edge is enhanced. Finally, the temporal causal features generated by the BiGRU and the spatially enhanced convolutional features are input into a fully connected network for cross-dimensional fusion, and the information redundancy between modalities is eliminated through a weight sharing mechanism. A residual connection structure is introduced during the fusion process to retain the physical interpretability of the original features. The generated multi-modal fusion features simultaneously encode the spatio-temporal propagation law of process parameter adjustment, the boundary constraint conditions of equipment operating status, and the cross-scale correlation of leaf quality evolution, providing a feature expression with high discrimination and strong generalization ability for intelligent decision-making.

[0026] Further, in a preferred embodiment of the present invention, for the construction of the tea processing quality perception model, the multi-modal tea processing fusion features are input into the trained tea processing quality perception model to perform tea processing quality perception on the current processing batch of tea, which specifically includes: Retrieve historical processing instances of different tea processing qualities based on historical data, and extract the historical tea processing features when each historical processing instance performs tea processing to obtain historical tea processing feature information; Extract the tea processing processes corresponding to each historical processing instance, partition the historical tea processing feature information, and generate several tea processing feature subsets of different historical processing instances. Each tea processing feature subset corresponds to a different tea processing process; Associate the historical tea processing feature information after feature partitioning with the corresponding historical tea processing quality, use the tea processing feature as the first node, the tea processing quality as the second node, and define the tea processing process as the node label of the first node to establish a topological structure diagram; Construct a tea processing quality perception model based on a long short-term memory network and a graph neural network, and input the historical tea processing feature information after feature partitioning into the long short-term memory network to extract multi-scale time features of different processing qualities; Obtain the adjacency matrix through the established topological structure diagram, perform aggregation and update of nodes based on the message passing mechanism and obtain the updated node representations, perform global pooling on the node representations after the whole graph is updated to obtain spatial features, and fuse the multi-scale temporal features and spatial features to generate spatio-temporal fusion features; Input the spatio-temporal fusion features into the fully connected layer for tea processing quality perception, verify and adjust the parameters of the tea processing quality perception results through a preset validation set, and obtain a tea processing quality perception model that meets the expectations after iterative training; Obtain the multi-modal tea processing fusion features, input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch, and obtain tea processing quality perception information.

[0027] It should be noted that first, process instance data of different quality levels are retrieved from the historical database, and multi-dimensional processing features corresponding to each instance are extracted (such as the temperature curve in the fixation stage, the kneading pressure gradient, the humidity fluctuation in the fermentation environment, etc.). Based on the processing procedure stages (withering, fixation, kneading, fermentation, drying), the feature sequence is dynamically partitioned to form feature subsets strongly associated with specific procedures. A topological relationship network is constructed through graph structure modeling technology, where each feature subset serves as a graph node, the node attributes encode feature statistics, and the edge weights between nodes are jointly determined by the process connection relationship and feature mutual information. The historical processing quality (sensory evaluation score, physical and chemical index detection value) is used as the target attribute of the graph node, and the non-linear association between features and quality is learned through a graph embedding algorithm. A tea processing quality perception model is constructed, and the core of the model is jointly composed of a long short-term memory network (LSTM) and a graph neural network (GNN): the LSTM network performs time series modeling on the feature subsets of each procedure to capture the dynamic evolution law of process parameters within the procedure (such as the hourly impact of humidity regulation on polyphenol oxidase activity in the fermentation stage); the GNN, based on the adjacency matrix, transmits topological association information across procedures, and uses the message passing mechanism to aggregate the features of adjacent nodes (such as the kneading procedure node receiving the residual enzyme activity features from the fixation procedure), and generates spatial features encoding the coupling effect between procedures through the node update function. The fusion of spatio-temporal features is achieved through a cascading method: the multi-scale temporal features output by the LSTM (including minute-level fluctuations within the procedure and hourly-level trends across procedures) and the spatial topological features generated by the GNN (reflecting the material and energy transfer relationship between procedures) are concatenated into a joint vector and input into a fully connected network for non-linear mapping. During the training process, a curriculum learning strategy is adopted to optimize the model in stages: initially focusing on capturing the time laws within the procedure, and later strengthening the cross-procedure spatial association modeling, and balancing the contribution ratio of spatio-temporal features through a dynamic weight loss function. Finally, the multi-modal fusion features of the current processing batch (including real-time process parameters, equipment status, and tea characteristics) are input into the trained perception model, and the model analyzes the similarity between the processing state and historical high-quality cases through the spatio-temporal feature decoder, and outputs a quantitative evaluation result covering the immediate quality score (such as the current fixation uniformity index) and the final quality prediction (expected sensory score of the finished tea), providing a decision-making basis for the follow-up.

[0028] Figure 2 It is a flowchart of an intelligent decision-making and auxiliary method for a tea processing process provided by an embodiment of the present invention; As Figure 2 shown, the present invention provides a flowchart of an intelligent decision-making and auxiliary method for a tea processing process, including: S202. Obtain the tea processing quality perception information, and generate a tea quality change perception curve for the current processing batch based on the tea processing quality perception information. The tea quality change perception curve represents the tea processing quality of the current processing batch at different timestamps. S204. Obtain the tea quality requirements for the current processing batch, calculate the deviation between the tea processing quality corresponding to different timestamps and the tea quality requirements in combination with the tea quality change perception curve, and judge the calculated deviation against a preset tolerance range. S206. If it is greater than the preset tolerance range, it means that the current tea processing control parameters do not match the tea quality requirements of the current processing batch. Then obtain the preset processing plan for the current processing batch for processing plan optimization. S208. Retrieve historical processing cases that meet the current tea processing quality requirements and the tea variety being processed from the preset historical processing database, and extract the corresponding historical processing control plans for each historical processing case. S210. Introduce a multi-objective grey wolf optimization algorithm, initialize the population according to the historical processing control plans corresponding to each historical processing case, preset the objective function and set the constraint conditions, calculate the objective function value of each individual in the initial grey wolf population through the objective function, and perform non-dominated sorting to divide all individuals in the population into different front levels. S212. Calculate the crowding distance for each front level to obtain the crowding degree distance, select the individual with the largest crowding degree in each level as the leader wolf, regard the remaining individuals as follower wolves, and calculate the direction vector of the leader wolf in the decision space for position update. Repeat the iteration until the stop condition is met, and then output the optimal solution set. S214. Generate several candidate processing optimization plans based on the output optimal solution set, conduct feasibility screening on each candidate processing optimization plan, and select the optimal processing optimization plan based on the feasibility screening result for tea processing assistance.

[0029] It should be noted that a dynamic quality evolution curve is constructed based on the tea processing quality perception information obtained through real-time perception. This curve maps the continuous change trends of the key indicators of tea quality (such as the oxidation rate of tea polyphenols, the retention degree of chlorophyll, and the predicted value of sensory evaluation scores) with the time axis as the reference. By comparing the preset quality requirements (such as the moisture content ≤ 6% and the phenolic ammonia ratio of 2.5 - 3.2 in the super-grade tea standard and other thresholds), the multi-dimensional deviation vector between the actual quality indicators and the target values at each process node is calculated. When the composite deviation index at the key processes (such as the end point of fixation and the turning point of fermentation) exceeds the preset tolerance threshold, it is determined that the current process parameters do not match the target quality requirements, and the adaptive optimization mechanism is triggered. The mapping relationship between the process control parameters (temperature gradient, pressure curve, humidity setting value) and the quality compliance rate of high-quality processing examples of the same tea type is retrieved from the historical case library and used as the prior knowledge of the optimization algorithm. A multi-objective grey wolf optimization algorithm is used to construct a parameter optimization framework: in the initialization stage, the control schemes of historical high-quality cases are encoded as grey wolf individuals, and the population diversity is expanded through Latin hypercube sampling; the objective function is defined as a weighted combination of minimizing quality deviation, maximizing energy consumption efficiency, and minimizing equipment loss rate, and the constraint conditions include the physical limits of the equipment (such as the maximum rotation speed of the drum and the maximum power of the heating plate) and the process specifications (such as the minimum duration of fermentation). During the algorithm iteration process, the population is divided into multiple Pareto front levels through non-dominated sorting, and the crowding distance of each level of individuals is calculated to select the most representative leading individuals to guide the population to move towards the optimal solution domain. The position update strategy guided by the leading wolf direction is executed in each round of iteration: the individual with the largest crowding degree in each level is selected as the leading wolf, and the remaining individuals are used as following wolves. The gravitational vector in the decision space is calculated according to their parameter combinations, and the following wolves are driven to explore along the optimization direction of quality - energy consumption - efficiency. An adaptive mutation operator is designed for the physical constraints of the equipment. When the parameter combination exceeds the feasible domain, it is reflected and corrected along the normal vector of the constraint boundary to ensure the engineering feasibility of the solution. After the evolutionary calculation of the preset number of generations, a candidate solution set covering multi-objective balance is output. 3 - 5 optimal solutions are selected through the feasibility verification module, and finally the control parameter set with the highest comprehensive score is selected by the entropy weight - TOPSIS decision method and sent to the execution mechanism of the processing equipment in real time through the edge computing unit, forming a closed-loop control link of "perception - evaluation - optimization - execution" to achieve the adaptive and precise control of the processing process.

[0030] Figure 3 A tea processing assistance system 3 based on intelligent decision-making provided by an embodiment of the present invention includes: a memory 31 and a processor 32. The memory 31 contains a program for the tea processing assistance method based on intelligent decision-making. When the program for the tea processing assistance method based on intelligent decision-making is executed by the processor 32, the following steps are implemented: Monitor the tea processing of the target processing equipment to obtain tea processing monitoring data, and preprocess the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; Analyze the cross-modal correlation relationships of the monitoring data based on the cross-modal tea processing monitoring information, and construct multi-modal fusion features to obtain multi-modal tea processing fusion features; Construct a tea processing quality perception model, and input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perceive the processing quality of the tea in the current processing batch, and obtain tea processing quality perception information; Judge whether the quality of the tea in the current processing batch meets the expectation based on the tea processing quality perception information. If not, optimize the current processing plan to assist in tea processing.

[0031] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the displayed or discussed components can be through some interfaces. The indirect coupling or communication connection of the devices or units can be electrical, mechanical, or other forms.

[0032] The units described above as separate components may or may not be physically separated. The components displayed as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0033] In addition, each functional unit in the embodiments of the present invention can be all integrated in one processing unit, or each unit can be separately used as one unit, or two or more units can be integrated in one unit; the above integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0034] Those of ordinary skill in the art will understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including those of the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0035] Alternatively, if the above integrated unit is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks, or optical discs and other various media that can store program codes.

[0036] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An auxiliary method for tea processing based on intelligent decision-making, characterized in that, Including: Conduct tea processing monitoring on the target processing equipment to obtain tea processing monitoring data, and preprocess the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; Analyze the cross-modal correlation relationships of the monitoring data based on the cross-modal tea processing monitoring information, and construct multi-modal fusion features to obtain multi-modal tea processing fusion features; Construct a tea processing quality perception model, and input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perceive the processing quality of the tea in the current processing batch, and obtain tea processing quality perception information; Judge whether the quality of the tea in the current processing batch meets the expectation based on the tea processing quality perception information. If not, optimize the current processing plan to assist in tea processing.

2. The auxiliary method for tea processing based on intelligent decision-making according to claim 1, characterized in that The conduct of tea processing monitoring on the target processing equipment to obtain tea processing monitoring data, and the preprocessing of the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information specifically includes: When the target processing equipment is performing tea processing, use the installed sensor array to conduct tea processing monitoring on the target processing equipment to obtain tea processing monitoring data. The tea processing monitoring data includes processing environment monitoring data, processing equipment operation data, and processed tea monitoring data; Perform wavelet packet decomposition on the tea processing monitoring data using a preset wavelet basis function to obtain different subbands and calculate the energy entropy of each subband, and perform soft threshold filtering based on the calculated energy entropy to identify the effective feature frequency band; Perform time series alignment processing on the tea processing monitoring data after the filtering process, use median filtering to remove noise and Gaussian smoothing to eliminate motion blur, introduce a sliding window algorithm to slide based on a preset window and calculate the mean and standard deviation of the data within the window, and combine the linear interpolation algorithm to perform data cleaning processing; Perform maximum-minimum normalization on the tea processing monitoring data after the cleaning process, map the continuous data to the [0,1] interval, and standardize the discrete data through binning processing to generate cross-modal tea processing monitoring information.

3. A tea processing assistance method based on intelligent decision-making according to claim 1, characterized in that The analysis of the cross-modal correlation relationships of the monitoring data based on the cross-modal tea processing monitoring information, and the construction of multi-modal fusion features to obtain multi-modal tea processing fusion features specifically includes: Obtain the cross-modal tea processing monitoring information, obtain the processed tea monitoring data through the cross-modal tea processing monitoring information and extract the processed tea image monitoring data, and import the processed tea image monitoring data into the HSV color space to obtain the histogram distribution; Calculate the weighted fusion of the local region hue mean and saturation variance using the histogram distribution to form the browning degree index. Subsequently, extract the contour of the largest connected region of the processed tea image through a morphological algorithm, calculate the ratio between the contour perimeter and the tea area as the curling degree index, and generate the first feature information in combination with the processed tea texture features; By obtaining the monitoring data of processed tea, extracting the near-infrared spectral data, calculating the first derivative to generate the near-infrared derivative spectrum, inputting the near-infrared derivative spectrum into a pre-trained least squares regression model for inverse analysis of tea chemical components to generate the spectral domain feature vector of the processed tea, and obtaining the second feature information; Introduce the multi-head attention mechanism, use the first feature information as the query vector, use the second feature information as the key vector and value vector, calculate the scaled dot attention score for feature fusion to generate the fused feature information; Utilize cross-modal tea processing monitoring information to extract the processing environment monitoring data and the processing equipment operation data, and input them together with the fused feature information into the BiGRU network for multi-modal feature fusion to obtain the multi-modal tea processing fused feature.

4. An auxiliary method for tea processing based on intelligent decision-making according to claim 3, characterized in that, The utilization of cross-modal tea processing monitoring information to extract the processing environment monitoring data and the processing equipment operation data, and input them together with the fused feature information into the BiGRU network for multi-modal feature fusion specifically includes: In the time dimension, convert the data input into the BiGRU network into several time-series feature vectors, where each time-series feature vector contains the tea texture details, spectral chemical components, equipment operation parameters, and processing environment parameters at the current moment; The forward propagation layer captures the immediate changes in the leaf state caused by the adjustment of the processing parameters through the update gate, and the backward pass uses the reset gate to filter the lagged changes in the leaf state caused by the adjustment of the processing parameters to generate the time-series causal features; In the space dimension, perform spatial convolution on the data input into the BiGRU network using the convolutional layer to obtain the convolutional feature map, and perform global average pooling on the convolutional feature map in the channel dimension to generate the spatial saliency weight vector; Obtain the time-series hidden state of the BiGRU, perform matrix multiplication with the spatial saliency weight vector, calculate the importance scores of each channel in the time-series context, and perform channel weighting on the original convolutional feature map to obtain the spatially responsive enhanced feature; Input the time-series causal features and the spatially responsive enhanced features into the fully connected layer for feature vector fusion to generate the multi-modal tea processing fused feature when the target processing equipment processes tea.

5. A tea processing assistance method based on intelligent decision-making according to claim 1, characterized in that The construction of the tea processing quality perception model, inputting the multi-modal tea processing fused feature into the trained tea processing quality perception model to perform the processing quality perception of the tea in the current processing batch specifically includes: Retrieve historical processing instances of different tea processing qualities based on historical data, extract the historical tea processing features of each historical processing instance when processing tea according to the obtained historical processing instances, and obtain the historical tea processing feature information; Extract the tea processing processes corresponding to each historical processing instance, perform feature partitioning on the historical tea processing feature information to generate several tea processing feature subsets of different historical processing instances, and each tea processing feature subset corresponds to a different tea processing process; Associate the historical tea processing feature information after feature partitioning with the corresponding historical tea processing quality, use the tea processing feature as the first node, the tea processing quality as the second node, and define the tea processing process as the node label of the first node to establish a topological structure diagram; Construct a tea processing quality perception model based on long short-term memory network and graph neural network, and input the historical tea processing feature information after feature partitioning into the long short-term memory network to extract multi-scale time features of different processing qualities; Obtain the adjacency matrix through the established topological structure diagram, perform node aggregation and update based on the message passing mechanism to obtain the updated node representation, perform global pooling on the node representation after the whole graph is updated to obtain spatial features, and fuse the multi-scale time features and spatial features to generate spatio-temporal fusion features; Input the spatio-temporal fusion features into the fully connected layer for tea processing quality perception, verify the tea processing quality perception results and adjust the parameters through a preset validation set, and obtain a tea processing quality perception model that meets the expectations after iterative training; Obtain multi-modal tea processing fusion features, input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch, and obtain tea processing quality perception information.

6. The auxiliary method for tea processing based on intelligent decision-making according to claim 1, characterized in that Judge whether the quality of the tea in the current processing batch meets the expectations based on the tea processing quality perception information. If not, optimize the current processing plan for tea processing assistance, specifically including: Obtain the tea processing quality perception information, and generate a tea quality change perception curve for the current processing batch based on the tea processing quality perception information. The tea quality change perception curve represents the tea processing quality of the tea in the current processing batch at different timestamps; Obtain the tea quality requirements for the current processing batch, calculate the deviation between the tea processing quality corresponding to different timestamps and the tea quality requirements in combination with the tea quality change perception curve, and judge the calculated deviation with a preset tolerance range; If it is greater than the preset tolerance range, it means that the current tea processing control parameters do not match the tea quality requirements of the current processing batch, then obtain the preset processing plan for the current processing batch for processing plan optimization; Retrieve historical processing cases that meet the tea quality requirements of the current processed tea and the tea variety of the processed tea in the preset historical processing database, and extract the corresponding historical processing control plans for each historical processing case; Introduce a multi-objective grey wolf optimization algorithm, initialize the population according to the historical processing control plans corresponding to each historical processing case, preset the objective function and set the constraint conditions, calculate the objective function values of each individual in the initial grey wolf population through the objective function, and perform non-dominated sorting to divide all individuals in the population into different front levels; Perform congestion calculation on each front level to obtain the crowding distance, select the individual with the largest crowding degree in each level as the leading wolf, regard the remaining individuals as the following wolves, and calculate the direction vector of the leading wolf in the decision space for position update, and output the optimal solution set after repeating the iteration until the stop condition is met; Generate several candidate processing optimization plans based on the output optimal solution set, perform feasibility screening on each candidate processing optimization plan, and select the optimal processing optimization plan based on the feasibility screening results for tea processing assistance.

7. An intelligent decision-making-based tea processing assistance system, characterized in that, The system includes: a memory and a processor. The memory contains a program for an intelligent decision-making-based tea processing assistance method. When the program for the intelligent decision-making-based tea processing assistance method is executed by the processor, the following steps are implemented: Perform tea processing monitoring on the target processing equipment to obtain tea processing monitoring data, and preprocess the obtained tea processing monitoring data to obtain cross-modal tea processing monitoring information; Analyze the cross-modal correlation relationships of the monitoring data based on the cross-modal tea processing monitoring information, and construct multi-modal fusion features to obtain multi-modal tea processing fusion features; Construct a tea processing quality perception model, input the multi-modal tea processing fusion features into the trained tea processing quality perception model to perform tea processing quality perception on the tea of the current processing batch, and obtain tea processing quality perception information; Judge whether the quality of the tea of the current processing batch meets the expectation based on the tea processing quality perception information. If not, optimize the current processing plan for tea processing assistance.

Images