Intelligent tea rolling device based on multi-modal information and control method

Abstract

In the tea making process, rolling has an important influence on the quality and yield of tea. In order to solve the problems of uneven rolling, high tea breakage rate and other problems existing in the existing tea rolling process, the present application proposes a kind of tea intelligent rolling device based on multi-modal information, which comprises a mechanical system, a control system and a detection system, wherein the mechanical system comprises a rolling mechanism, a pressure mechanism, a power mechanism and a discharging mechanism. The control system comprises an information acquisition and processing device and an intelligent control device. The detection system comprises a photoelectric switch, a CCD industrial camera and other information acquisition devices. By fusing visual, hyperspectral, electronic nose and other information, the tenderness, color shape, strip rate, cell breakage rate and chemical composition of tea are monitored in real time, and intelligent diagnosis and precise control are carried out according to these information, so as to realize real-time adjustment of rolling parameters, improve the production efficiency and rolling quality of tea.

Description

Technical Field

[0001] This invention relates to the field of tea processing technology, specifically to a smart tea kneading device and control method based on multimodal information. Background Technology

[0002] Rolling is an important step in tea processing. It involves kneading, rolling, or pressing fresh tea leaves to break down the tea's cells and release cell fluid, which is beneficial for chemical reactions and aroma formation. During rolling, the leaves change shape under friction and pressure, forming strips, needles, or other shapes. It is an indispensable process in tea processing.

[0003] During the rolling process, the skill level of the workers or the setting of the rolling machine's process parameters have a significant impact on the production efficiency and quality of the rolling process. Currently, tea rolling machines either start and stop by setting fixed process parameters initially or by relying on human experience to judge the degree of rolling. Existing technology has several problems, such as uneven rolling, high tea breakage rate, inconsistent leaf shape, high subjectivity in judgment, and substandard strip formation rate. Furthermore, the quality of harvested tea leaves varies, including differences in the age and size of leaves. Using the same process parameters to roll different types of tea will lead to either a high breakage rate or a low strip formation rate, ultimately resulting in inconsistent quality of the rolled leaves.

[0004] For example, CN204930234U discloses a self-controlled tea rolling machine, which controls various components through a control system. First, an initial pressing pressure and rolling drum speed are set based on the age and moisture content of the tea leaves. The pressure detected by the pressure sensor is compared with the set pressure in real time, and the signal is fed back to the pressurizing and driving mechanisms to control the pressure and speed, thus achieving closed-loop control of pressure and speed. This avoids the inconsistency of tea quality caused by manually controlling pressure and speed according to the processing technology. Finally, the control system judges the degree of rolling based on the rolling time, and discharges the tea after the standard is met.

[0005] While the above-mentioned scheme can set appropriate initial process parameters based on the initial state of the tea leaves and achieve closed-loop control of pressure during the kneading process, it cannot determine the degree of kneading based on the specific conditions during the kneading process, thereby failing to accurately control other process parameters such as kneading pressure and speed.

[0006] Existing technologies cannot accurately control process parameters such as rolling force and rolling speed in real time based on the specific state of the tea leaves during the rolling process, thus failing to improve tea production efficiency and rolling quality.

[0007] Therefore, in order to solve the problems existing in the current technology, it is urgent to integrate multimodal information, propose an adaptive control method for kneading processing parameters, develop an intelligent tea kneading device, improve the shortcomings of the above-mentioned traditional kneading machine, and improve the kneading quality. Summary of the Invention

[0008] To address the problems and deficiencies of existing technologies, this invention provides an intelligent tea rolling device and control method based on multimodal information. The purpose of this invention is to integrate multimodal information such as vision, hyperspectral imaging, electronic nose, and pressure with a control system to monitor in real time the tenderness of tea leaves before rolling, as well as the color, shape, strip formation rate, cell breakage rate, and chemical composition of tea leaves during rolling. This enables intelligent diagnosis of the rolling degree and precise control of process parameters, thereby improving production efficiency and rolling quality.

[0009] To achieve the above objectives, this invention proposes an intelligent tea kneading device based on multimodal information, comprising a mechanical system, a control system, and a detection system; the mechanical system is used to complete the tea kneading process, the detection system is used to detect the kneading state of the tea, and the control system is used to intelligently adjust the process parameters of tea kneading.

[0010] The mechanical system includes a frame, a kneading mechanism, a pressure mechanism, a power mechanism, and a feeding mechanism. The kneading mechanism is mounted on the frame. The output end of the pressure mechanism is connected to the top of the kneading mechanism, and the pressure mechanism is used to provide the pressure required for kneading the tea leaves. The power mechanism is located on one side of the frame, and the output end of the power mechanism is connected to the side of the kneading mechanism. The power mechanism is used to provide the power required for the operation of the kneading mechanism. The feeding mechanism is connected to the output end at the bottom of the kneading mechanism, and the feeding mechanism is used to control the discharge of the kneaded tea leaves.

[0011] Preferably, the kneading mechanism includes a kneading cover, a transparent kneading barrel, a frame, a secondary operating arm, a kneading disc, and a main operating arm; the kneading disc is mounted on a base; the transparent kneading barrel is mounted on the kneading disc; the kneading cover is mounted on top of the transparent kneading barrel and is connected to the output end of the pressure mechanism; the frame is fitted over the transparent kneading barrel; the main operating arm is located between the power mechanism and the frame, and its two ends are connected to the power mechanism and the frame, respectively; at least two secondary operating arms are also provided between the frame and the base, so that the transparent kneading barrel makes horizontal circular motion within the kneading disc; the surface of the kneading disc is uniformly provided with ribs along its circumference.

[0012] Preferably, the pressure mechanism includes a geared motor, a column, a pressure arm, a pressure rod, and a pressure spring; the geared motor is mounted on the frame; the column is connected to the output end of the geared motor, and the geared motor can drive the lead screw inside the column to rotate; a sliding nut on the pressure arm is mounted on the lead screw, and under the restriction of the column, the sliding nut can only move up and down on the lead screw, thereby driving the pressure arm to move up and down; one end of the pressure arm is sleeved on the column, and the other end of the pressure arm is connected to the pressure rod and the pressure spring; the end of the pressure rod and the pressure spring away from the pressure arm is connected to the kneading cover through a pressure sensor; the up and down movement of the pressure arm transmits pressure to the kneading cover; the pressure sensor forms a closed-loop control of the kneading pressure by detecting the pressure value and the system set value. During the pressurization or depressurization process, the pressure sensor signal determines whether the pressure has reached the system set value, thereby determining whether the geared motor rotates forward or backward, thereby controlling the up and down movement of the pressure arm and realizing pressure feedback control.

[0013] Preferably, the power mechanism is mounted on the frame, and the power mechanism includes a motor and a gearbox; the output end of the motor is connected to the gearbox, and one end of the gearbox is connected to the main rotating arm; the motor can transmit power to the main rotating arm through the gearbox, thereby driving the kneading drum to make circular motion on the kneading disc.

[0014] Preferably, the feeding mechanism is installed at the bottom of the kneading pan, which includes a right-angle hollow geared motor, a tea outlet door, a locking tongue and lock housing, and two baffles. The tea outlet door is located at the tea outlet in the middle of the kneading pan. The output end of the right-angle hollow geared motor is connected to the rotating shaft of the tea outlet door. The locking tongue and lock housing are located at the opening of the tea outlet door and are used to control the opening and closing of the tea outlet door. The two baffles are located below the tea outlet door.

[0015] During feeding, the right-angle hollow geared motor runs, driving the tea outlet gate shaft to rotate. The cam is connected to the shaft by a key and rotates accordingly, causing the locking tongue to separate from and close from the lock case, thereby realizing the opening and closing of the tea outlet gate. Two baffles are installed below the tea outlet gate so that the kneaded tea leaves fall onto the output conveyor belt.

[0016] Preferably, the detection system starts operating ten minutes before the end of the kneading process. The system includes a photoelectric switch, two CCD industrial cameras, an electronic nose acquisition device, and a hyperspectral acquisition device. To achieve real-time image capture during kneading and obtain high-quality photos, thus accelerating subsequent image processing, a CCD industrial camera is selected. CCD industrial cameras offer high compatibility, resulting in images free of noise and of high image quality. Furthermore, a suitable light source can highlight the tea leaf characteristics in the image, improving image quality, reducing post-processing time, and increasing processing efficiency. A strip light source is used to avoid glare caused by direct sunlight on the transparent kneading drum surface; top-down illumination ensures uniform and sufficient light.

[0017] The photoelectric switch is located on the edge of the base in front of the camera. When the detection system is working, the photoelectric switch is activated. After the photoelectric switch is activated, the transparent kneading drum moves to the position of the photoelectric switch for the first time. Then the transparent kneading drum begins to decelerate. When it moves to the position of the photoelectric switch for the third time, the speed drops to 0. The kneading drum stops in front of the No. 1 camera for 5 seconds to prevent the motion blur of the kneading drum from affecting the clarity of the industrial camera and the hyperspectral camera.

[0018] Two CCD industrial cameras are installed. Camera 1, mounted on the base, captures the state of the tea leaves inside the transparent kneading drum during the kneading process. Camera 2, positioned above the conveyor belt, captures the characteristic state of the tea leaves before kneading to determine their tenderness. The main purpose is to capture real-time images of the tea leaves before kneading. Camera 1 begins operation ten minutes before the end of kneading, using continuous shooting. Every three minutes before the end of kneading, when the kneading drum touches the photoelectric sensor and slows down to a stop in front of the camera, five photos are taken to capture the state of the tea leaves inside the transparent kneading drum during the kneading process. Because there is reflection when photographing the transparent kneading drum, a polarizing filter is installed on Camera 1 to eliminate reflection. Camera 2, positioned above the conveyor belt, captures the characteristics of the tea leaves before kneading as they are conveyed to the conveyor belt to determine their tenderness.

[0019] The electronic nose acquisition device includes a plastic tube and an electronic nose sensor. A small hole is provided on the side of the transparent kneading drum. One end of the plastic tube is inserted into the small hole, and the other end of the plastic tube is connected to the electronic nose sensor. The electronic nose sensor is used to detect the concentration of gas emitted by the tea cell breakage caused by kneading in the transparent kneading drum at regular intervals. By establishing the relationship between the sensor display value and the cell breakage rate, the degree of kneading is indirectly judged, thereby controlling the process parameters. The principle of the electronic nose sensor is that the metal oxide sensor comes into full contact with the gas to be measured, causing the conductivity of the sensor to change, thereby converting the gas signal into an electrical signal.

[0020] The hyperspectral acquisition device includes a hyperspectral camera and a hyperspectral image processing and analysis system. The hyperspectral camera is positioned opposite to the primary camera, across the kneading pan, and is used to capture hyperspectral images of the tea leaves on the pan every 3 minutes, 10 minutes before the end of kneading. Since the kneading pan is positioned in front of the camera during operation, the hyperspectral camera is not obstructed and can clearly capture images of the tea leaves scattered on the pan. The hyperspectral image processing and analysis system preprocesses the captured hyperspectral images using various methods and extracts characteristic wavelengths. Because each chemical component has specific spectral characteristics in its absorption and reflection of light, these characteristics can be reflected in the hyperspectral images. By acquiring and analyzing hyperspectral images of the tea surface, the chemical components in the tea can be quantitatively analyzed, thereby indirectly determining the degree of tea kneading.

[0021] Preferably, the control system includes an information acquisition and processing device and an intelligent control device;

[0022] The information acquisition and processing device is used to perform image preprocessing on the acquired tea photos (adjusting contrast, translation, rotation, scaling, noise enhancement, and other data enhancement processing), and to preprocess the acquired electronic nose and hyperspectral data, and transmit them to the intelligent control device to extract the corresponding features.

[0023] The intelligent control device includes a human-computer interaction module, a feature extraction and processing module, a data management module, a rolling parameter recommendation module, and a rolling parameter adjustment module. The human-computer interaction module is used to view rolling process parameters and feedback on the detected degree of tea rolling, as well as collect manually input information such as tea type and origin. The feature extraction and processing module extracts features from relevant indicators of the rolled leaves, converts them into feature vectors, and inputs them into the model for training or prediction. The data management module compares and analyzes the processed tea data collected by the detection system with standard tea parameters in the database and provides feedback, thereby enabling real-time control of the rolling machine's process parameters. The rolling parameter recommendation module, combined with the manually input tea information from the human-computer interaction module, uses the model-predicted tenderness level as a query condition, connects to the database through a database interface provided by a programming language (e.g., SQL), executes a query operation, retrieves matching process parameters from the database, and displays them through the human-computer interaction module, thus recommending initial process parameters. The rolling parameter adjustment module is used to collect and process the data of various indicators, comprehensively evaluate each indicator to calculate the corresponding comprehensive score of tea, and implement the control strategy by writing a program to achieve the purpose of adjusting the process parameters.

[0024] The control method for the intelligent tea kneading device based on multimodal information, as described above, includes the following steps:

[0025] The yield rate of tea leaves is a crucial indicator of the rolling process and a significant standard for evaluating tea quality. A higher yield rate typically results in tea leaves with a uniform appearance, good shape, and tightly rolled leaves. Furthermore, the chemical composition content of tea leaves significantly impacts their flavor, aroma, and nutritional value, and can also serve as a standard for judging the degree of rolling. Finally, the cell breakage rate reflects the degree of damage to the tea's cell structure. For certain tea varieties and styles, moderate cell breakage can release more aroma and flavor. However, an excessively high cell breakage rate may lead to increased bitterness or off-flavors. Therefore, during the rolling process, these three indicators—yield rate, cell breakage rate, and chemical composition content—will be evaluated in real-time. By adjusting process parameters in real-time, these three indicators will be brought to the expected levels, improving the final rolling quality. The steps are as follows:

[0026] S1. Process the production data accumulated during the tea rolling process and establish a database of various tea rolling indicators and corresponding rolling process parameters; the data in the database are all derived from the standard parameters and process parameters corresponding to obtaining high-quality rolled tea during the production process.

[0027] S2. Accumulate production image data, extract relevant features of tea leaves, and establish a prediction model for tea leaf rolling indicators; the specific steps are as follows:

[0028] S2.1 Collect a large number of tea images captured by the second camera on the flat conveyor, and perform image preprocessing: image filtering (Gaussian filtering, etc.), image grayscale processing, binarization, morphological processing to improve the shape and structure of the tea leaves to facilitate feature extraction, and then use data augmentation (adjusting contrast, translation, rotation, scaling, adding noise) to expand the dataset and increase the robustness of the model.

[0029] S2.2 Image Feature Extraction: The acquired tea images were color space converted (RGB, HSI, Lab), and nine color indices (R, G, B, H, S, I, L, GR, BY) were extracted. The texture of the fresh tea images was acquired using a gray-level co-occurrence matrix, and features such as entropy, contrast, energy, correlation, and homogeneity were extracted as texture feature indices. Each tea image was assigned a tenderness label. The extracted color and texture features were used as input features, and the tenderness label of the tea images was used as the target variable. The dataset was divided into training and testing sets in an 8:2 ratio. A least squares support vector machine (LS-SVM) model was used to construct a model of the relationship between the color and texture features of tea and the tenderness of tea to predict the tenderness of tea. The model performance was evaluated by accuracy, and the final result was set to three levels: "tender leaves, medium tenderness, and mature leaves".

[0030] The least squares support vector machine model is shown in equation (2) below:

[0031]

[0032] Where, Y = [y 21 ], y 21 For the tenderness of the tea leaves, x = [x 11 ,x 12 ,x 13 ,x 14 ,...],x 11 ,x 12 ,x 13 ,x 14 ,... represent the nine color and texture features of tea leaves, respectively; b is the bias; and σ is the sensitive region of the radial basis kernel function.

[0033] The database contains a large amount of standard tenderness and matching process parameter data. Using the predicted tenderness grade as the query condition, the system connects to the database through a database interface provided by a programming language (such as SQL) and executes the query to retrieve the matching process parameters. Finally, the matching results are output, showing the matched process parameters along with the predicted tenderness for later use or display to users. After confirmation by the staff, these parameters are used for the kneading process.

[0034] S3. Using the CCD industrial camera, hyperspectral camera, and electronic nose device of the detection system, measure the relevant parameters of the tea leaves during the rolling process and construct the corresponding relationship model; the specific steps are as follows:

[0035] S3.1 After image preprocessing of the tea leaf images captured by camera 1 inside the kneading drum, the tea leaf contour features are extracted using an edge detection algorithm (Canny edge detection). Morphological processing is then applied to these contour features, and the tea leaf contour is fitted to an ellipse using the fitting ellipse function in OpenCV. The length, width, and aspect ratio are obtained from the major and minor axes of the ellipse. Furthermore, the tea leaf image is converted into a binary image through contour analysis. Curve fitting is performed on the points on the acquired tea leaf edge contour, and the curvature or bending degree of the fitted curve is calculated to obtain curve measurement indicators. Curve measurement can be used to quantify the degree of bending, the magnitude of curvature, and the overall morphological characteristics of the tea leaves. Generally, tea leaves that are rolled into strips tend to have larger curve measurement values, meaning they are more curved and form a distinct curved shape. This is because during the tea-making process, the tea leaves are rolled into strip-like structures, and strip-like tea leaves typically exhibit a more curved shape. Conversely, tea leaves with a lower degree of strip formation may have smaller curve measurement values, meaning they are relatively straighter and have less curvature.

[0036] Based on the extracted tea leaf dimensions (length, width, aspect ratio, curve measurement, etc.), a suitable model from Support Vector Machines, Decision Trees, and Random Forests is selected to train on the features of a large number of images within the tea-making drum. These models take the tea leaf feature data as input and output classification results indicating whether the tea leaves are in strips or not. Cross-validation is used to evaluate the performance of each model, and the optimal model is selected to construct a tea leaf feature and strip-based discrimination model. Within each of the five image regions captured by camera one, the number of strip-shaped tea leaves and the total number of tea leaves are identified. Object detection algorithms are used to detect and locate the tea leaves in the images. For detected tea leaves, image segmentation techniques are used to separate the tea leaves from the background, allowing for better feature extraction. For each segmented tea leaf region, corresponding features are extracted, and the trained model is used to classify each tea leaf, determining whether it is a strip-shaped tea leaf. Based on the classification results, all regions identified as tea leaves are counted to obtain the total number of tea leaves. Meanwhile, for areas classified as "formed tea leaves," the number of formed tea leaves was counted, and the forming rate was calculated for five images. The average forming rate of the five images was used to evaluate the forming rate of the tea leaves in the kneading drum. A forming rate of 80% was set as the standard forming rate, and 85% was set as the maximum limit. If the forming rate was too high, the amount of broken tea leaves would also increase. The actual forming rate was compared with the standard forming rate. If it was lower than the standard forming rate, the kneading pressure, speed, and kneading time would be appropriately increased. If the forming rate was between 80% and 85%, processing continued according to the established parameters. If the forming rate was higher than 85%, the pressure and speed were reduced.

[0037] S3.2 Determine the content of tea polyphenols, caffeine, and amino acids in rolled leaves under different rolling degrees; the content of tea polyphenols refers to "Determination of Tea Polyphenols" (GB / T 8313-2002); the content of amino acids refers to "Determination of Total Free Amino Acids in Tea" (GB / T 8314-2013); the content of caffeine refers to "Determination of Tea Caffeine" (GB / T 8312-2002).

[0038] As the rolling process continues, tea leaves rupture and multiply, releasing polyphenols, caffeine, and amino acids. Excessive or insufficient levels of these three chemical components can negatively impact the tea's flavor. Therefore, precise control of their content is crucial during rolling. A hyperspectral camera captures hyperspectral images of tea leaves scattered on the rolling pan at appropriate levels. The acquired hyperspectral images undergo black-and-white correction, outlier removal, and appropriate preprocessing (SG smoothing, multivariate scattering correction, differential processing, etc.). Then, feature wavelength extraction methods such as correlation analysis (CA) and principal component analysis (PCA) are used to extract image wavelengths, simplifying the original wavelengths to dozens of representative wavelengths. Finally, the hyperspectral data undergoes dimensionality reduction, noise reduction, and light scattering elimination. The processed hyperspectral data is transformed into feature vectors, which are then trained using a partial least squares support vector machine model to establish a hyperspectral-based prediction model for the content of the three internal components in tea leaves during the rolling process.

[0039] S3.3 During the kneading process, the tissue cells rupture, volatile substances are released, or the polyphenol oxidase in the tea leaves reacts with the tea polyphenols to form aroma compounds such as alcohols, aldehydes, and esters; by connecting a plastic tube to an electronic nose sensor and inserting it into the small hole on the side of the transparent kneading drum (7), the concentration of aroma formed by the rupture of tea cells inside is detected, and the cell rupture rate of the tea leaves is indirectly judged.

[0040] Two electronic nose sensors, W5S and W2W, were used to detect the gas inside the kneading drum and record the corresponding response values. During the kneading process, the electronic nose device was used to collect aroma concentration data at regular intervals. After each detection, the sensor response value was cleaned by air pump until it returned to the initial value before subsequent experimental data detection and analysis were performed, and the cell breakage rate of tea leaves at the same time was sampled and measured. The electronic nose sensor response value data was cleaned and aligned with the sensor data and the corresponding cell breakage rate data to ensure that the data collected each time and the cell breakage rate data were collected at the same time and under the same conditions. The sensor response value was used as the input and the cell breakage rate was used as the output. The dataset was divided into a training set and a test set in an 8:2 ratio. The linear regression model was trained using the training dataset to fit the relationship between the sensor response value and the cell breakage rate. The relationship model between the sensor response value and the cell breakage rate was established as shown in the following formula (4):

[0041] R=β0+β1×X1+β2×X2+…+β n ×X n (4)

[0042] Where R is the cell disruption rate, X1, X2, ..., X n The response values ​​of each sensor, β1, β2, ..., β nβ is the regression coefficient, representing the effect of each sensor response value on the cell breakage rate, and β0 is the intercept term, which is the cell breakage rate when each sensor response value is 0.

[0043] S4. Start the kneading machine. All systems and devices will work. During the kneading process, the condition of the tea leaves will be monitored in real time to obtain relevant data. A comprehensive evaluation will be conducted, and the process parameters will be adjusted in a timely manner to achieve real-time and precise control of the kneading process.

[0044] After processing, the data obtained by the above detection system are input into the kneading parameter adjustment module of the control system. First, these data are normalized to make them uniform in scale. The Min-Max Normalization method is used to map the value of each index to the range between 0 and 1.

[0045]

[0046] Where X is the normalized strip formation rate, Y is the normalized cell breakage rate, Z1, Z2, and Z3 are the normalized tea polyphenols, amino acids, and caffeine, respectively; x, y, z1, z2, and z3 are the strip formation rate, cell breakage rate, tea polyphenols, amino acids, and caffeine content of tea at a certain moment, respectively; (max)1, (max)2, (max)3, (max)4, and (max)5 represent the expected values ​​of strip formation rate, cell breakage rate, and the three chemical components, respectively; and (min)1, (min)2, (min)3, (min)4, and (min)5 represent the minimum values ​​detected during the historical production process of strip formation rate, cell breakage rate, and the three chemical components, respectively.

[0047] The weights for strip formation rate, chemical composition content, and cell disruption rate are set as shown in the following formula (1):

[0048] S=X×ω X +Y×ω Y +(Z1×ω1+Z2×ω2+Z3×ω3)×ω Z (1)

[0049] Where S is the quality score of the tea, ω X ω Y ω Z The weights are strip formation rate, cell breakage rate, and chemical composition, respectively. ω1, ω2, and ω3 are the weights of tea polyphenols, amino acids, and caffeine in the chemical composition, respectively.

[0050] Based on the comparison between the tea quality score and the target score range, the process parameters of the rolling machine are adjusted accordingly. When the quality score is lower than the target value, the rolling pressure and rolling speed are increased. If the quality score is still lower than the target value in the last test by the detection system, the rolling time will be extended appropriately. When the quality score reaches the expected value, the parameters are not adjusted, and the rolling continues with the current parameters until the end, and the subsequent detection system stops working.

[0051] The beneficial effects of this invention are:

[0052] This invention innovatively incorporates computer vision, hyperspectral imaging, and electronic nose technology into a rolling machine, integrating them with a control system. It recommends initial process parameters based on tea leaf tenderness, establishes closed-loop control of rolling pressure via pressure sensors, uses computer vision to roughly determine the tea leaf strip formation rate during rolling, analyzes changes in the content of chemical components (caffeine, amino acids, tea polyphenols, soluble sugars) during rolling using hyperspectral imaging, and indirectly determines cell breakage rate by detecting gas concentration using an electronic nose sensor. By detecting these factors and adjusting the rolling machine's process parameters accordingly, the desired rolling effect is achieved. This real-time monitoring and control of the rolling process overcomes the subjectivity and uncertainty in judging the degree of rolling, making the rolling process more objective, intelligent, and controllable, thus improving rolling quality and production efficiency. Attached Figure Description

[0053] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0054] Figure 1 This is a three-dimensional structural diagram of the intelligent tea kneading device based on multimodal information of the present invention;

[0055] Figure 2 This is a schematic diagram of the planar structure of the feeding mechanism of the present invention;

[0056] Figure 3 These are hyperspectral images acquired by the hyperspectral camera of this invention;

[0057] Figure 4 This relates the response value of the electronic nose sensor of the present invention to the cell disruption rate;

[0058] Figure 5 This is a schematic diagram of the process of constructing each model in this invention;

[0059] Figure 6This is a control flowchart of the control method for the intelligent tea kneading device based on multimodal information of the present invention;

[0060] The components shown in the diagram are: base 1, geared motor 2, column 3, pressure arm 4, pressure rod and pressure spring 5, kneading cover 6, transparent kneading drum 7, frame 8, auxiliary operating arm 9, kneading disc 10, rib 11, camera 12, main operating arm 13, motor 14, gearbox 15, baffle 16, electronic nose sensor 17, hyperspectral camera 18, flat conveyor 19, camera 20, right-angle hollow geared motor 21, tea outlet door 22, and locking tongue and lock shell 23. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0062] like Figure 1-3 As shown, this example proposes an intelligent tea rolling device based on multimodal information, including a mechanical system, a control system, and a detection system. The mechanical system is used to complete the tea rolling process, the detection system is used to detect the rolling state of the tea, and the control system is used to intelligently adjust the process parameters of tea rolling. The mechanical system includes a frame, a rolling mechanism, a pressure mechanism, a power mechanism, and a feeding mechanism. The rolling mechanism is mounted on the frame. The output end of the pressure mechanism is connected to the top of the rolling mechanism, and the pressure mechanism is used to provide the pressure required for tea rolling. The power mechanism is located on one side of the frame, and the output end of the power mechanism is connected to the side of the rolling mechanism, and the power mechanism is used to provide the power required for the operation of the rolling mechanism. The feeding mechanism is connected to the output end at the bottom of the rolling mechanism, and the feeding mechanism is used to control the discharge of tea leaves after rolling.

[0063] The kneading mechanism includes a kneading cover 6, a transparent kneading drum 7, a frame 8, auxiliary operating arms 9, a kneading disc 10, and a main operating arm 13; the kneading disc 10 is mounted on the base 1; the transparent kneading drum 7 is mounted on the kneading disc 10; the kneading cover 6 is mounted on the top of the transparent kneading drum 7 and is connected to the output end of the pressure mechanism; the frame 8 is fitted over the transparent kneading drum 7; the main operating arm 13 is located between the power mechanism and the frame 8, and its two ends are connected to the power mechanism and the frame 8 respectively; at least two auxiliary operating arms 9 are also provided between the frame 8 and the base 1; the surface of the kneading disc 10 is uniformly provided with ribs 11 along its circumference.

[0064] The pressure mechanism includes a geared motor 2, a column 3, a pressure arm 4, a pressure rod, and a pressure spring 5; the geared motor 2 is mounted on the frame 8; the column 3 is connected to the output end of the geared motor 2, and the geared motor 2 can drive the lead screw inside the column 3 to rotate; one end of the pressure arm 4 is sleeved on the column 3, and the other end of the pressure arm 4 is connected to the pressure rod and the pressure spring 5; the end of the pressure rod and the pressure spring 5 away from the pressure arm 4 is connected to the kneading cover 6 through a pressure sensor.

[0065] The power mechanism is mounted on the frame and includes a motor 14 and a gearbox 15. The output end of the motor 14 is connected to the gearbox 15, and one end of the gearbox 15 is connected to the main operating arm 13. The motor 14 can transmit power to the main operating arm 13 by reducing speed through the gearbox 15.

[0066] The feeding mechanism is installed at the bottom of the kneading pan 10. The kneading pan 10 includes a right-angle hollow geared motor 21, a tea outlet gate 22, a locking tongue and locking shell 23, and two baffles 16. The tea outlet gate 22 is located at the tea outlet in the middle of the kneading pan 10. The output end of the right-angle hollow geared motor 21 is connected to the rotating shaft of the tea outlet gate 22. The locking tongue and locking shell 23 is located at the opening of the tea outlet gate 22 and is used to control the opening and closing of the tea outlet gate 22. The two baffles 16 are located below the tea outlet gate 22.

[0067] The detection system includes a photoelectric switch, two CCD industrial cameras, an electronic nose acquisition device, and a hyperspectral acquisition device.

[0068] The photoelectric switch is located at the edge of the base 1;

[0069] The CCD industrial camera is provided in two parts, including camera 12 and camera 20. Camera 12 is located on the base 1 and is used to collect the state of the tea leaves in the transparent kneading barrel 7 during the kneading process. Camera 20 is located above the flat conveyor 19 and is used to collect the characteristic state of the tea leaves before kneading in order to determine the tenderness of the tea leaves.

[0070] The electronic nose collection device includes a plastic tube and an electronic nose sensor 17. The transparent kneading drum 7 has a small hole on its side. One end of the plastic tube is inserted into the small hole, and the other end of the plastic tube is connected to the electronic nose sensor 17. The electronic nose sensor 17 is used to detect the concentration of gas emitted by the tea cells due to kneading in the transparent kneading drum 7 at regular intervals. By establishing the relationship between the sensor display value and the cell breakage rate, the degree of kneading can be indirectly judged, thereby controlling the process parameters.

[0071] The hyperspectral acquisition device includes a hyperspectral camera 18 and a hyperspectral image processing and analysis system; the hyperspectral camera 18 is mounted on the base 1 and is used to capture hyperspectral images of tea leaves during the rolling process; the hyperspectral image processing and analysis system is used to preprocess the captured hyperspectral images and extract characteristic wavelengths.

[0072] The control system includes an information acquisition and processing device and an intelligent control device;

[0073] The information acquisition and processing device is used to preprocess the acquired tea photos, as well as the acquired electronic nose and hyperspectral data, and transmit them to the intelligent control device to extract the corresponding features.

[0074] The intelligent control device includes a human-computer interaction module, a feature extraction and processing module, a data management module, a rolling parameter recommendation module, and a rolling parameter adjustment module. The human-computer interaction module is used to view rolling process parameters and feedback information related to the detected degree of tea rolling, as well as to collect manually input information about tea type and origin. The feature extraction and processing module is used to extract features from relevant indicators of the rolled leaves and convert them into feature vectors, which are then input into the model for training or prediction. The data management module is used to compare and analyze the processed tea data collected by the detection system with standard tea parameters in the database and provide feedback, thereby achieving real-time control of the rolling machine's process parameters. The rolling parameter recommendation module, in conjunction with the human-computer interaction module, uses manually input tea-related information and the model-predicted tenderness level as query conditions. It connects to the database through a database interface provided by a programming language, executes a query operation, retrieves matching process parameters from the database, and displays them through the human-computer interaction module, thus recommending initial process parameters. The rolling parameter adjustment module collects the processed indicator data, comprehensively evaluates each indicator to calculate the corresponding comprehensive tea score, and implements control strategies through programming to achieve the purpose of adjusting process parameters.

[0075] The control process in this embodiment is as follows:

[0076] Step 1: An experiment was conducted at XX Company, using XX tea as the experimental subject. Five professional workers comprehensively evaluated and categorized the tea leaves based on their tenderness, classifying them into tender leaves, medium-tender leaves, and mature leaves. These three types of tea leaves were then separately fed into a rolling machine. Based on production experience, different processing parameters were used for the tea leaves with different tenderness levels. After rolling, the five professional workers evaluated the final quality of the rolled leaves. Based on the evaluation results, the processing parameters for each type of tea were adjusted appropriately until at least three of the five workers evaluated the final rolling result of the three tenderness levels as excellent quality. The corresponding rolling processing parameters were recorded, and the tea leaves of the corresponding tenderness were processed three times using the same processing parameters. If the results still met the above standards, the tenderness and corresponding processing parameters were recorded in the database as initial recommended parameters. In addition, other data in the database, such as the strip formation rate, cell breakage rate, chemical composition, and corresponding processing parameters, were all derived from data recorded during the production process when high-quality rolled tea leaves were obtained. These were used as standard data for statistical analysis of production data to explore the relationship between tea processing parameters and quality indicators.

[0077] Step 2: Manually input relevant information about the tea, such as origin and type. The second camera captures images of the tea on the conveyor. The images are preprocessed by the information acquisition and processing device to facilitate feature extraction. The feature extraction and processing module extracts image features and converts them into feature vectors, which are then input into the least squares support vector machine model to predict the tenderness level.

[0078] Least squares support vector machine model:

[0079]

[0080] Where, Y = [y 21 ], y 21 For the tenderness of the tea leaves, x = [x 11 ,x 12 ,x 13 ,x 14 ,...],x 11 ,x 12 ,x 13 ,x 14 ... represent the color and texture characteristics of tea leaves, respectively.

[0081] Using predicted tenderness as the query condition, the system connects to the database through the kneading parameter recommendation module, matches the corresponding initial process parameters, and displays them in the human-machine interaction module. After the staff confirms the parameters, they use them to perform the kneading operation.

[0082] Step 3: 10 minutes before the rolling machine finishes its work, camera number one starts working, taking five images every 3 minutes. Through the image preprocessing module and feature extraction and processing module, the feature vector is input into the tea strip discrimination model. Based on the discrimination results, the number of tea strips and the total number of tea strips are identified. The strip formation rate of the tea in the five images is calculated and the average value is calculated. The average value is compared with the set standard strip formation rate and the maximum strip formation rate. If it is less than the standard strip formation rate, the rolling pressure and rolling time will be appropriately increased. If the strip formation rate is between 80% and 85%, processing will continue according to the established parameters. Since an excessively high strip formation rate will lead to an increased proportion of broken tea leaves in the final product, if the strip formation rate is higher than 85%, the pressure and speed will be reduced.

[0083] Step 4: 10 minutes before the end of the kneading machine's operation, a hyperspectral camera takes pictures every 3 minutes. The hyperspectral image processing and analysis system performs preprocessing and feature wavelength extraction, converting the hyperspectral data into feature vectors. Using a hyperspectral-based prediction model for the content of three internal components of tea during the kneading process, the content of amino acids, caffeine, and tea polyphenols is predicted.

[0084] Step 5: 10 minutes before the end of the kneading machine's operation, the gas concentration in the kneading drum is detected every 3 minutes using an electronic nose device. Based on the response values ​​of the W5S and W2W sensors, the cell breakage rate is predicted using a model that correlates the sensor response values ​​with the cell breakage rate.

[0085] R=(0.060×X1+0.075×X2-0.493)×100%

[0086] Where X1 is the response value of the W5S sensor, and X2 is the response value of the W2W sensor.

[0087]

[0088] Step 6: Based on the data obtained above regarding the yield, tea polyphenols, caffeine, amino acid content, and cell disruption rate, a comprehensive evaluation is performed. The Min-Max Normalization method is used to normalize these data.

[0089]

[0090] Where X represents the normalized strip formation rate, Y represents the normalized cell breakage rate, Z1, Z2, and Z3 represent the normalized tea polyphenols, amino acids, and caffeine, respectively; x, y, z1, z2, and z3 represent the strip formation rate, cell breakage rate, tea polyphenols, amino acids, and caffeine content of tea leaves at a certain moment; (max)1, (max)2, (max)3, (max)4, and (max)5 represent the expected values ​​of strip formation rate, cell breakage rate, and the three chemical components, respectively; and (min)1, (min)2, (min)3, (min)4, and (min)5 represent the minimum values ​​of strip formation rate, cell breakage rate, and the three chemical components detected during the historical production process, respectively.

[0091] In this embodiment, the expected values ​​for strip formation rate, cell breakage rate, tea polyphenols, amino acids, and caffeine are 85%, 55%, 33.2%, 3.2%, and 4.5%, respectively. In previous production data, the minimum values ​​measured were 60%, 30%, 25%, 2.1%, and 2.5%, respectively.

[0092] Based on the weights of strip formation rate, chemical composition content, and cell breakage rate within the control system

[0093] S=X×0.4+Y×0.3+(Z1×0.4+Z2×0.3+Z3×0.3)×0.3

[0094] Where S is the quality score of the tea, ω X ω Y ω Z The weights are strip formation rate, cell breakage rate, and chemical composition, respectively. ω1, ω2, and ω3 are the weights of tea polyphenols, amino acids, and caffeine in the chemical composition, respectively.

[0095] In this embodiment, the weights of strip formation rate, cell breakage rate, and chemical components (tea polyphenols, amino acids, and caffeine) are 0.4, 0.35, and 0.25, respectively, and the weights of tea polyphenols, amino acids, and caffeine among the chemical components are 0.4, 0.3, and 0.3, respectively.

[0096] The detection system runs every three minutes for ten minutes before the end of the kneading process. It compares the tea quality score with the target score (0.8 in this example), and uses the difference between the target and actual tea quality scores as the deviation. The process parameters of the kneading machine are adjusted accordingly, with the magnitude of the deviation serving as the basis for the adjustment. A conditional statement is added: if the quality score is lower than the target value, the kneading pressure and speed are increased. If the quality score is still lower than the target value in the last detection, the kneading pressure and speed are adjusted according to the deviation, and the kneading time is extended. This extension is only executed during the last operation of the detection system; afterwards, the system continues to operate at the previous time intervals. A conditional statement is added to the control loop: if the quality score is equal to or greater than the expected value, the loop is exited, the control process is terminated, parameters are not adjusted, and kneading continues with the current parameters until the end, after which the subsequent detection system stops operating.

[0097] Where e represents the deviation, P′, v′, and T′ represent the adjusted rolling pressure, rolling speed, and rolling time parameters, respectively, S0 represents the target score for tea quality, P0, v0, and T0 represent the initial rolling pressure, rolling speed, and rolling time parameters, k1 and k2 represent the adjustment coefficients for rolling pressure and rolling speed, and the unit of time T is min.

[0098] Furthermore, since an excessively high strip yield leads to an increase in tea dust and an excessively high cell breakage rate can increase bitterness or off-flavors in the tea, limiting conditions are introduced in this process. When the limiting conditions are exceeded, the kneading operation is stopped. This achieves a comprehensive evaluation of multiple indicators and controls the process parameters. The above control process is programmed into the corresponding program by a programmable logic controller (PLC).

[0099] Table 1: Detection Factors and Corresponding Methods

[0100]

[0101] Table 2: Comparison of Kneading Equipment

[0102]

[0103]

[0104] Table 3 Comparison of Kneading Quality

[0105]

[0106] As shown in Tables 1-3, this embodiment detects various factors and corresponding methods during the rolling process, achieving precise identification of tenderness, strip formation rate, chemical composition (caffeine, amino acids, tea polyphenols) content, and cell breakage rate during the rolling process, as well as intelligent control of various process parameters. It also demonstrates the advantages of this embodiment in terms of final tea quality and the innovation of control methods compared to traditional rolling machines.

[0107] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of the present invention. All equivalent changes or modifications made to the structure, features and principles described in the claims of the present invention should be included in the scope of the claims of the present invention.

[0108] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A tea intelligent kneading device based on multimodal information, characterized in that, It includes a mechanical system, a control system, and a detection system; the control system is used to intelligently adjust the process parameters of tea rolling. The mechanical system includes a frame, a kneading mechanism, a pressure mechanism, a power mechanism, and a feeding mechanism. The kneading mechanism is mounted on the frame. The output end of the pressure mechanism is connected to the top of the kneading mechanism, and the pressure mechanism provides the pressure required for kneading the tea leaves. The power mechanism is located on one side of the frame, and its output end is connected to the side of the kneading mechanism. The power mechanism provides the power required for the operation of the kneading mechanism. The feeding mechanism is connected to the output end at the bottom of the kneading mechanism, and the feeding mechanism controls the discharge of the kneaded tea leaves. The kneading mechanism includes a kneading cover (6), a transparent kneading barrel (7), a frame (8), a secondary operating arm (9), a kneading disc (10), and a main operating arm (13); the kneading disc (10) is mounted on the base (1); the transparent kneading barrel (7) is mounted on the kneading disc (10); the kneading cover (6) is mounted on the top of the transparent kneading barrel (7), and the kneading cover (6) is connected to the output end of the pressure mechanism; the frame (8) is fitted over the transparent kneading barrel (7); the main operating arm (13) is located between the power mechanism and the frame (8), and the two ends of the main operating arm (13) are connected to the power mechanism and the frame (8) respectively; at least two secondary operating arms (9) are also provided between the frame (8) and the base (1); the surface of the kneading disc (10) is uniformly provided with ribs (11) along its circumference. The detection system includes a photoelectric switch, two CCD industrial cameras, an electronic nose acquisition device, and a hyperspectral acquisition device. The photoelectric switch is disposed on the edge of the base (1); The CCD industrial camera is provided in two parts, including camera 1 (12) and camera 2 (20); camera 1 (12) is located on the base (1) and is used to collect the state of tea leaves in the transparent kneading barrel (7) during the kneading process; camera 2 (20) is located above the flat conveyor (19); The electronic nose collection device includes a plastic tube and an electronic nose sensor (17). The transparent kneading drum (7) has a small hole on its side. One end of the plastic tube is inserted into the small hole, and the other end of the plastic tube is connected to the electronic nose sensor (17). The electronic nose sensor (17) is used to detect the concentration of gas emitted by the tea cells in the transparent kneading drum (7) due to kneading at regular intervals. By establishing the relationship between the sensor display value and the cell breakage rate, the degree of kneading can be indirectly judged to control the process parameters. The hyperspectral acquisition device includes a hyperspectral camera (18) and a hyperspectral image processing and analysis system; the hyperspectral camera (18) is mounted on the base (1) and is used to capture hyperspectral images of tea leaves during the rolling process; The hyperspectral image processing and analysis system is used for preprocessing and extracting characteristic wavelengths from the captured hyperspectral images.

2. The intelligent tea kneading device based on multimodal information according to claim 1, characterized in that, The pressure mechanism includes a geared motor (2), a column (3), a pressure arm (4), a pressure rod, and a pressure spring (5); the geared motor (2) is mounted on the frame (8); the column (3) is connected to the output end of the geared motor (2), and the geared motor (2) can drive the lead screw inside the column (3) to rotate; one end of the pressure arm (4) is sleeved on the column (3), and the other end of the pressure arm (4) is connected to the pressure rod and the pressure spring (5); the end of the pressure rod and the pressure spring (5) away from the pressure arm (4) is connected to the kneading cover (6) through a pressure sensor.

3. The intelligent tea kneading device based on multimodal information according to claim 1, characterized in that, The power mechanism is mounted on the frame and includes a motor (14) and a gearbox (15). The output end of the motor (14) is connected to the gearbox (15), and one end of the gearbox (15) is connected to the main operating arm (13). The motor (14) can transmit power to the main operating arm (13) by reducing speed through the gearbox (15).

4. The intelligent tea kneading device based on multimodal information according to claim 1, characterized in that, The feeding mechanism is installed at the bottom of the kneading pan (10). The kneading pan (10) includes a right-angle hollow geared motor (21), a tea outlet gate (22), a locking tongue and locking shell (23), and two baffles (16). The tea outlet gate (22) is located at the tea outlet in the middle of the kneading pan (10). The output end of the right-angle hollow geared motor (21) is connected to the rotating shaft of the tea outlet gate (22). The locking tongue and locking shell (23) is located at the opening of the tea outlet gate (22) and is used to control the opening and closing of the tea outlet gate (22). The two baffles (16) are located below the tea outlet gate (22).

5. The intelligent tea kneading device based on multimodal information according to claim 1, characterized in that, The control system includes an information acquisition and processing device and an intelligent control device; The information acquisition and processing device is used to preprocess the acquired tea photos, as well as to preprocess the data acquired by the electronic nose acquisition device and the hyperspectral acquisition device, and transmit them to the intelligent control device to extract the corresponding features. The intelligent control device includes a human-computer interaction module, a feature extraction and processing module, a data management module, a rolling parameter recommendation module, and a rolling parameter adjustment module. The human-computer interaction module is used to view rolling process parameters and feedback information related to the detected degree of tea rolling, as well as to collect manually input information about tea type and origin. The feature extraction and processing module is used to extract features from relevant indicators of the rolled leaves, convert them into feature vectors, and input them into the model for training or prediction. The data management module is used to compare and analyze the processed tea data collected by the detection system with standard tea parameters in the database and provide feedback, thereby achieving real-time control of the rolling device's process parameters. The rolling parameter recommendation module, in conjunction with the human-computer interaction module, uses manually input tea-related information and the model-predicted tenderness level as query conditions. It connects to the database through a database interface provided by a programming language, executes a query operation, retrieves matching process parameters from the database, and displays them through the human-computer interaction module, thus recommending initial process parameters. The rolling parameter adjustment module collects the processed indicator data, comprehensively evaluates each indicator to calculate the corresponding comprehensive tea score, and implements control strategies through programming to achieve the purpose of adjusting process parameters.

6. The control method for the intelligent tea kneading device based on multimodal information according to any one of claims 1-5, characterized in that, Includes the following steps: S1. Process the production data accumulated during the tea rolling process and establish a database of various tea rolling indicators and corresponding rolling process parameters; the data in the database are all derived from the standard parameters and process parameters corresponding to obtaining high-quality rolled tea during the production process. S2. Accumulate production image data, extract relevant features of tea leaves, and establish a prediction model for relevant indicators of tea rolling; S3. Using the CCD industrial camera, hyperspectral camera, and electronic nose acquisition device of the detection system, measure the relevant parameters of tea leaves during the rolling process and construct the corresponding relationship model. S4. Start the kneading machine. All systems and devices will work. During the kneading process, the condition of the tea leaves will be monitored in real time to obtain relevant data. A comprehensive evaluation will be conducted, and the process parameters will be adjusted in a timely manner to achieve real-time and precise control of the kneading process. After processing, the data obtained by the above detection system are input into the kneading parameter adjustment module of the control system. First, these data are normalized to make them uniform in scale. The Min-Max Normalization method is used to map the value of each index to the range between 0 and 1. Where X is the normalized strip formation rate and Y is the normalized cell disruption rate. , , These represent the normalized values ​​of tea polyphenols, amino acids, and caffeine, respectively, x, y, , , These are the percentage of tea leaves that have formed into strips, the percentage of cells broken, the content of tea polyphenols, the content of amino acids, and the content of caffeine at a certain point in time. , , , , These represent the expected values ​​of strip formation rate, cell disruption rate, and three chemical components, respectively. , , , , These represent the minimum values ​​detected during the historical production process for strip yield, cell breakage rate, and the three chemical components, respectively. The weights for strip formation rate, chemical composition content, and cell disruption rate are set as shown in the following formula (1): S= （1） Where S represents the quality score of the tea. , These are the weights of strip formation rate, cell breakage rate, and chemical composition, respectively. The weights of tea polyphenols, amino acids, and caffeine in the chemical composition are respectively. Based on the comparison between the tea quality score and the target score range, the process parameters of the rolling machine are adjusted accordingly. When the quality score is lower than the target value, the rolling pressure and rolling speed are increased. If the quality score is still lower than the target value in the last test by the detection system, the rolling time will be extended appropriately. When the quality score reaches the expected value, the parameters are not adjusted, and the rolling continues with the current parameters until the end, and the subsequent detection system stops working.

7. The control method for the intelligent tea kneading device based on multimodal information according to claim 6, characterized in that, In step S3, the specific steps for constructing the relational model are as follows: S3.1 After image preprocessing of the tea leaf image inside the transparent kneading drum acquired by camera No. 1, the tea leaf contour features are extracted using an edge detection algorithm. Morphological processing is then performed on the contour features, and the tea leaf contour is fitted to an ellipse using the fitting ellipse function in OpenCV. The length, width, and aspect ratio are obtained from the major and minor axes of the ellipse. In addition, the tea leaf image is converted into a binary image through contour analysis. The points on the obtained tea leaf edge contour are then curve-fitted, and the curvature or bending degree is calculated from the fitted curve to obtain curve measurement indicators. Based on the extracted length, width, aspect ratio, and curve metric of the tea leaves, a suitable model is selected from support vector machine, decision tree, and random forest models to train the features of a large number of images inside the kneading drum. These models will take the tea leaf feature data as input and output the corresponding classification results of whether the tea leaves are in strips or not. The performance of each model is evaluated using cross-validation, and the optimal model is selected to construct a tea leaf feature and tea leaf strip discrimination model. S3.2 Determine the content of tea polyphenols, caffeine, and amino acids in tea leaves rolled to different degrees; S3.3 During the rolling process, the tissue cells rupture, releasing volatile substances, or the polyphenol oxidase in the tea reacts chemically with tea polyphenols to form aroma compounds such as alcohols, aldehydes, and esters; by connecting a plastic tube to an electronic nose sensor and inserting it into a small hole on the side of the transparent rolling drum (7), the concentration of aroma formed due to the rupture of tea cells can be detected, indirectly determining the cell breakage rate of the tea: Using the electronic nose sensor response value as input and cell breakage rate as output, the dataset is divided into a training set and a test set in an 8:2 ratio. The linear regression model is trained using the training dataset to fit the relationship between the sensor response value and the cell breakage rate, and a relationship model between the sensor response value and the cell breakage rate is established, as shown in the following equation (4): （4） Where R represents the cell disruption rate. , For the response values ​​of each sensor, , , , The regression coefficients represent the impact of each sensor response value on the cell disruption rate. is the intercept term, and is the cell breakage rate when the response value of each sensor is 0.

Images