A precise method for wilting leaves based on water distribution
By combining near-infrared hyperspectral imaging and deep neural network models, the parameters of the tea leaves were dynamically adjusted, which solved the problem of uneven tea quality caused by differences in moisture distribution in withered leaves, and achieved a stable improvement in the aroma, taste and leaf quality of tea.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HONGHE TWO TEA MOUNTAINS TEA IND CO LTD
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies cannot accurately respond to differences in the distribution of moisture inside withered leaves, resulting in quality defects such as uneven withering, red stems, or insufficient fermentation during the tea-making process. Consequently, the finished tea cannot consistently meet high-quality standards in terms of aroma purity, flavor richness, and leaf uniformity.
By scanning the wilting leaves with a near-infrared hyperspectral imaging system to generate a grayscale map of moisture distribution, and combining this with a deep neural network model to dynamically adjust the wave parameters and provide real-time feedback correction, adaptive control of the wilting leaves is achieved. This includes adjusting the motor speed, the frequency of forward and reverse rotation of the cylinder, and the air volume to ensure uniform moisture distribution.
It significantly improves the uniformity of the greening process, promotes the formation of floral and fruity aroma compounds, suppresses astringency, and enhances the stability and yield of tea aroma, taste, and leaf quality, thus achieving a process upgrade from relying on experience to data-driven processes.
Smart Images
Figure CN122430280A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tea processing technology, specifically to a precise method for processing tea leaves based on the moisture distribution of withered leaves. Background Technology
[0002] In the processing of teas such as oolong and black tea, which require the withering (or rolling) process, the moisture state of the withered leaves is a core factor determining the rolling process parameters and subsequent fermentation quality. Traditional rolling methods rely heavily on the tea master's personal experience, involving manually or mechanically turning and bumping the withered leaves to observe the degree of reddening at the leaf edges and changes in aroma to determine the end point of rolling. However, this experiential judgment ignores the spatial differences in moisture distribution within and between the leaves. In actual production, due to factors such as the time of fresh leaf picking, leaf maturity, and uneven distribution of temperature and humidity in the withering environment, withered leaves often exhibit significant moisture gradients between the leaf edge and the leaf center, the leaf surface and the leaf stem, and the upper and lower layers of leaves. Using a uniform rolling intensity and time can easily lead to excessive localized water loss or moisture retention in the leaves, resulting in uneven withering and quality defects such as "dead greens," "red stems," or insufficient fermentation.
[0003] While existing technologies include devices that detect the overall moisture content of withered leaves to assist in controlling the ripening process, these devices essentially rely on coarse-grained control of overall average parameters and cannot accurately respond to differences in moisture distribution within the leaf's micro-regions. Furthermore, due to the lack of real-time feedback and closed-loop adjustment mechanisms for dynamic changes in moisture distribution, the ripening process cannot adaptively correct itself based on the actual response of the leaves. This results in the finished tea failing to consistently achieve high-quality standards in terms of aroma purity, flavor richness, and leaf uniformity. Summary of the Invention
[0004] To address this issue, the present invention provides a precise method for processing tea based on the moisture distribution of withered leaves, thereby solving the problem in the prior art that it is impossible to adaptively correct according to the actual reaction of the leaves, resulting in the difficulty in consistently achieving high-quality standards in terms of aroma purity, flavor richness, and leaf uniformity of the finished tea.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A precise method for controlling leaf moisture distribution based on wilting includes the following steps: Step 1, Moisture Distribution Collection: Lay the wilted leaves to be harvested flat on the detection platform, scan the wilted leaves using a near-infrared hyperspectral imaging system, and obtain the spectral reflectance of each pixel in the 900-1700nm band. Based on the pre-established partial least squares moisture quantification model, convert the spectral data into moisture content values and generate a grayscale map of the moisture distribution of the wilted leaves with spatial coordinate information, where the grayscale value corresponds to the moisture content. Step 2, Moisture Region Segmentation: The grayscale image of moisture distribution is enhanced and binarized. Based on the overall average moisture content of the wilted leaves, the regions with moisture content higher than 15% of the average content are marked as high-moisture zones, the regions with moisture content lower than 15% of the average content are marked as low-moisture zones, and the regions in between are marked as medium-moisture zones. The area ratio of high-moisture zones, medium-moisture zones, and low-moisture zones, as well as the moisture gradient vector field between high-moisture zones and low-moisture zones, are calculated respectively. Step 3, wave parameter calculation: The area proportion of each moisture region and the average magnitude of the moisture gradient vector field are input into the pre-trained deep neural network model. The model outputs the basic wave intensity coefficient, wave cycle duration and cylinder forward and reverse switching frequency. The basic wave intensity coefficient is positively correlated with the area proportion of the high moisture region, and the wave cycle duration is positively correlated with the average magnitude of the moisture gradient vector field. Step 4, Adaptive wave-forming execution: The wave-forming machine is controlled to set the motor speed according to the basic wave-forming intensity coefficient, set the total time of a single wave-forming cycle according to the wave-forming cycle duration, and alternately change the rotation direction of the wave-forming cylinder according to the forward and reverse rotation switching frequency of the cylinder to dynamically impact and tumble the withering leaves; Step 5, Real-time Feedback Correction: During the wave execution process, Step 1 to Step 2 are repeated every 30 seconds to obtain the current moisture distribution grayscale map and the real-time change curve of the high water zone area ratio. When the real-time value of the high water zone area ratio decreases less than the preset expected decrease threshold compared to the initial value, the basic wave intensity coefficient is increased by one step increment until the real-time change curve of the high water zone area ratio converges to the expected decrease threshold.
[0006] Preferably, in step one, before scanning, the withered leaves of the *Gynostemma pentaphyllum* are subjected to background removal and leaf separation processing. The threshold segmentation method is used to extract individual leaves independently from the group image, and a corresponding grayscale sub-image of water distribution is generated for each independent leaf, and the spatial coordinates of each leaf are recorded.
[0007] Preferably, in step two, when calculating the water gradient vector field, the internal gradient field of each independent leaf is calculated for the grayscale sub-image of water distribution, and the proportion of gradient vectors in all independent leaves with an angle greater than 60 degrees between the gradient direction and the leaf midrib direction is counted. The proportion of the number is defined as the leaf margin-leaf center water difference index.
[0008] Preferably, in step three, the pre-trained deep neural network model includes an input layer, three hidden layers, and an output layer. The nodes of the input layer correspond to feature vectors, which are composed of the area proportions of high-water zones, medium-water zones, and low-water zones, the average magnitude of the moisture gradient vector field, and the moisture difference index between the leaf edge and the leaf center. The nodes of the output layer correspond to the basic wave-washing intensity coefficient, the wave-washing cycle duration, and the frequency of forward and reverse rotation of the cylinder. The training dataset of the deep neural network model consists of multiple sets of historical wave-washing process parameters and their corresponding finished tea leaf uniformity scores, with the highest leaf uniformity score as the objective for supervised learning. When the moisture difference index between the leaf edge and the leaf center is greater than a preset critical threshold, the forward and reverse rotation frequency of the cylinder output by the model is forcibly set to 1.5 times the base frequency to increase the mechanical impact intensity on the leaf edge area and promote the accelerated conduction of moisture from the leaf edge to the leaf center. The base frequency is the standard switching frequency obtained by mapping the average magnitude of the moisture gradient vector field.
[0009] Preferably, in step four, the adaptive wave-forming process further includes synchronously adjusting the airflow parameters inside the wave-forming machine according to the frequency of the cylinder's forward and reverse rotation. Specifically, during forward rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a first airflow rate to promote the dissipation of moisture from the surface of the wilting leaves; during reverse rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a second airflow rate, which is less than the first airflow rate, to reduce excessive water loss during the blade collision process. The ratio of the first airflow rate to the second airflow rate is proportional to the basic wave-forming strength coefficient.
[0010] Preferably, in step five, the real-time feedback correction also includes a protection mechanism for low-moisture areas: during the wave-forming process, the real-time change curve of the area ratio of low-moisture areas is monitored simultaneously. When the real-time value of the area ratio of low-moisture areas increases by more than the preset over-dry threshold relative to the initial value, an intensity reduction command is immediately triggered, reducing the basic wave-forming intensity coefficient by two step increments, and simultaneously reducing the first air volume of the auxiliary fan to 60% of the initial first air volume, until the area ratio of low-moisture areas falls back to within a safe range.
[0011] Preferably, the method also includes a termination determination step: when the proportion of high water zone area of all independent leaves in the current moisture distribution grayscale map obtained by repeating steps one to two three times consecutively is less than 10% of the initial high water zone area proportion, and the moisture difference index between leaf edge and leaf center is less than 0.1, the wave washing is determined to be over, the wave washing machine is controlled to stop running and the wave washing completion signal is output.
[0012] This invention has the following advantages: By constructing a moisture distribution map, this invention divides the leaves into different moisture regions and dynamically generates personalized withering strategies based on the characteristic parameters of each region, while combining real-time monitoring feedback for closed-loop correction; This invention can significantly improve the uniformity of fermentation of withered leaves during the withering process, avoid processing defects caused by uneven moisture distribution, effectively promote the formation of floral and fruity aroma substances and suppress astringent taste, thereby greatly improving the stability and high yield of tea aroma, taste and leaf quality, and realizing a process upgrade from relying on experience to data-driven, and from overall average to spatially precise. Attached Figure Description
[0013] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).
[0014] Figure 1 A flowchart illustrating a precise method for controlling the moisture distribution of wilted leaves, provided in this application embodiment. Detailed Implementation
[0015] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] Please see Figure 1 A precise method for controlling leaf moisture distribution based on wilting includes the following steps: Step 1, Moisture Distribution Collection: Lay the wilted leaves to be harvested flat on the detection platform, scan the wilted leaves using a near-infrared hyperspectral imaging system, and obtain the spectral reflectance of each pixel in the 900-1700nm band. Based on the pre-established partial least squares moisture quantification model, convert the spectral data into moisture content values and generate a grayscale map of the moisture distribution of the wilted leaves with spatial coordinate information, where the grayscale value corresponds to the moisture content. Step 2, Moisture Region Segmentation: The grayscale image of moisture distribution is enhanced and binarized. Based on the overall average moisture content of the wilted leaves, the regions with moisture content higher than 15% of the average content are marked as high-moisture zones, the regions with moisture content lower than 15% of the average content are marked as low-moisture zones, and the regions in between are marked as medium-moisture zones. The area ratio of high-moisture zones, medium-moisture zones, and low-moisture zones, as well as the moisture gradient vector field between high-moisture zones and low-moisture zones, are calculated respectively. Step 3, wave parameter calculation: The area proportion of each moisture region and the average magnitude of the moisture gradient vector field are input into the pre-trained deep neural network model. The model outputs the basic wave intensity coefficient, wave cycle duration and cylinder forward and reverse switching frequency. The basic wave intensity coefficient is positively correlated with the area proportion of the high moisture region, and the wave cycle duration is positively correlated with the average magnitude of the moisture gradient vector field. Step 4, Adaptive wave-forming execution: The wave-forming machine is controlled to set the motor speed according to the basic wave-forming intensity coefficient, set the total time of a single wave-forming cycle according to the wave-forming cycle duration, and alternately change the rotation direction of the wave-forming cylinder according to the forward and reverse rotation switching frequency of the cylinder to dynamically impact and tumble the withering leaves; Step 5, Real-time Feedback Correction: During the wave execution process, Step 1 to Step 2 are repeated every 30 seconds to obtain the current moisture distribution grayscale map and the real-time change curve of the high water zone area ratio. When the real-time value of the high water zone area ratio decreases less than the preset expected decrease threshold compared to the initial value, the basic wave intensity coefficient is increased by one step increment until the real-time change curve of the high water zone area ratio converges to the expected decrease threshold.
[0017] The above solution is illustrated below through examples: Step 1: Take 200kg of withered oolong tea leaves (moisture content of about 65%), lay them flat on a conveyor belt, and scan them with a hyperspectral camera at a resolution of 2mm / pixel to obtain a grayscale image containing 1.5 million pixels, with each pixel corresponding to the moisture value of a 0.02mm² leaf area.
[0018] Step 2: Moisture Region Segmentation The average moisture content of this batch of withered leaves is 65%. Pixels with a moisture content higher than 74.75% are classified as high-moisture zones (35% of the area), those with a moisture content lower than 55.25% are classified as low-moisture zones (28% of the area), and the rest are classified as medium-moisture zones. The average modulus of the moisture gradient vector field is 0.18 (moisture change rate per millimeter).
[0019] Step 3: Calculation of Wave Green Parameters DNN output: Basic wave intensity coefficient 0.72 (corresponding to motor speed 28rpm), wave cycle duration 14 minutes, forward and reverse switching frequency once every 40 seconds.
[0020] Step 4: Adaptive Wave Execution The wave tube starts at 28 rpm clockwise for 40 seconds, then counterclockwise for 40 seconds, and repeats this cycle for a total duration of 14 minutes.
[0021] Step 5: Real-time feedback and correction The initial high-water zone area accounted for 35%. The expected decrease was 2% every 30 seconds. After 2 minutes of execution, the actual decrease was only 1.5%. The system automatically increased the intensity coefficient from 0.72 to 0.77 (speed increased to 30 rpm), and then the decrease rate returned to 2.1%, and the system stabilized.
[0022] In step one, before scanning, the withered leaves of the green leaf are subjected to background removal and leaf separation processing. The threshold segmentation method is used to extract individual leaves independently from the group image, and a corresponding grayscale sub-image of water distribution is generated for each independent leaf, and the spatial coordinates of each leaf are recorded.
[0023] When withered leaves are laid flat, they may overlap or touch at the edges, and group images cannot reflect the moisture differences within individual leaves. By using a threshold segmentation method, each leaf is independently extracted from the background and neighboring leaves, generating its own grayscale sub-image of moisture distribution, and its spatial coordinates are recorded. This allows all subsequent analyses (such as gradient calculation and region segmentation) to be accurate to the leaf level, avoiding mutual interference of moisture signals between leaves and providing a data structure foundation for precise leaf-by-leaf control.
[0024] In step two, when calculating the moisture gradient vector field, the internal gradient field is calculated for each individual leaf's moisture distribution grayscale sub-image. The proportion of gradient vectors whose gradient direction forms an angle greater than 60 degrees with the leaf's midrib is then counted and defined as the leaf margin-center moisture difference index. Moisture in a healthy wilted leaf should be conducted from the leaf margin to the center along the midrib; therefore, the moisture gradient vector should be roughly parallel to the midrib. If a large number of gradient vectors form an angle greater than 60 degrees with the midrib, it indicates an abnormal lateral distribution of moisture (such as localized dryness at the leaf margin or water accumulation in the leaf center). This index directly quantifies the smoothness of moisture conduction within the leaf. A higher index indicates a more severe uneven state of "wet inside, dry outside" or "wet outside, dry inside" in the leaf, requiring subsequent targeted strengthening or weakening of mechanical action on the leaf margin area.
[0025] In step three, the pre-trained deep neural network model includes an input layer, three hidden layers, and an output layer. The nodes in the input layer correspond to feature vectors, which are composed of the area proportions of high-water zones, medium-water zones, and low-water zones, the average magnitude of the moisture gradient vector field, and the moisture difference index between the leaf edge and the leaf center. The nodes in the output layer correspond to the basic wave-washing intensity coefficient, wave-washing cycle duration, and the frequency of forward and reverse rotation of the cylinder. The training dataset of the deep neural network model consists of multiple sets of historical wave-washing process parameters and their corresponding finished tea leaf uniformity scores, with the highest leaf uniformity score as the objective for supervised learning. When the moisture difference index between the leaf edge and the leaf center is greater than a preset critical threshold, the forward and reverse rotation frequency of the cylinder output by the model is forcibly set to 1.5 times the base frequency to increase the mechanical impact intensity on the leaf edge area and promote the accelerated conduction of moisture from the leaf edge to the leaf center. The base frequency is the standard switching frequency obtained by mapping the average magnitude of the moisture gradient vector field.
[0026] This solution incorporates a forced intervention mechanism: when the index exceeds a preset critical threshold, the system no longer relies on the default switching frequency output by the model, but instead forcibly increases the forward and reverse switching frequency to 1.5 times the baseline value. The principle behind this design is that a higher reversal frequency enables the waveguide tube to produce a "rubbing" motion pattern, increasing the number of frictions and collisions between the leaf edge and the tube wall, and between the blades. This effectively breaks down the water conduction resistance of the leaf edge cuticle, accelerates the migration of water from the leaf edge to the leaf center, and rapidly reduces the water difference index.
[0027] Traditional wave-weaving machines typically provide constant airflow, which can easily lead to excessive airflow causing the blades to dry out during forward rotation, or insufficient airflow causing the blades to pile up during reverse rotation, resulting in stuffiness. To address this issue, the following feature is implemented: In step four, the adaptive wave-weaving execution further includes synchronously adjusting the airflow parameters inside the wave-weaving machine according to the frequency of forward and reverse rotation of the cylinder. Specifically, during forward rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a first airflow rate to promote the dissipation of moisture from the surface of the wilted leaves; during reverse rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a second airflow rate, which is less than the first airflow rate, to reduce excessive water loss during blade collision. The ratio of the first airflow rate to the second airflow rate is proportional to the basic wave-weaving intensity coefficient.
[0028] This invention dynamically adjusts the airflow based on the cylinder's rotation direction: a larger airflow (first airflow) is used during forward rotation to promote surface moisture evaporation by utilizing the opportunity of blade dispersion; a smaller airflow (second airflow) is used during reverse rotation to prevent the blades from becoming brittle and losing excessive water due to excessive airflow during collision and compression. Simultaneously, the ratio of the first airflow to the second airflow is directly proportional to the basic wave strength coefficient—the greater the wave strength, the stronger the moisture dissipation capacity required, therefore the forward airflow is relatively larger than the reverse airflow. This coordinated airflow-rotation control significantly improves the flexibility and precision of the wave system.
[0029] In step five, the real-time feedback correction also includes a protection mechanism for low-moisture areas: during the wave-forming process, the real-time change curve of the area ratio of low-moisture areas is monitored simultaneously. When the real-time value of the area ratio of low-moisture areas increases by more than the preset over-dry threshold relative to the initial value, an intensity reduction command is immediately triggered, reducing the basic wave-forming intensity coefficient by two step increments, and simultaneously reducing the first air volume of the auxiliary fan to 60% of the initial first air volume, until the area ratio of low-moisture areas falls back to within a safe range.
[0030] During the slicker blade process, if the proportion of the low-water zone area increases abnormally (exceeding the preset over-drying threshold), it indicates that the current process parameters have caused irreversible excessive water loss to some blades. At this point, the system immediately triggers emergency intervention: reducing the slicker blade intensity coefficient by two steps (larger than the normal correction step, demonstrating rapid correction), and simultaneously reducing the first airflow of the auxiliary fan to 60% of its initial value. This dual intensity reduction measure can quickly alleviate the mechanical impact and airflow sweeping on the over-drying blades, preventing "dead blades." Once the low-water zone area returns to a safe range, the system can resume normal adjustment.
[0031] It also includes a termination determination step: when the proportion of high water zone area of all independent leaves in the current moisture distribution grayscale map obtained by repeating steps one to two for three consecutive times is less than 10% of the initial high water zone area proportion, and the moisture difference index between leaf edge and leaf center is less than 0.1, the wave-washing is determined to be over, the wave-washing machine is controlled to stop running and the wave-washing completion signal is output.
[0032] The termination criteria include two simultaneous conditions: first, the area ratio of high-water zones in all individual leaves is less than 10% of the initial value, indicating that the high-water core within the leaf has essentially disappeared and the water distribution is becoming more uniform; second, the leaf margin-to-center water difference index is less than 0.1, indicating that the water gradient direction within each leaf is basically consistent with the midrib, and the abnormal water conduction has been sufficiently corrected. Simultaneously, three consecutive tests (30-second intervals) must meet these conditions to avoid misjudgments caused by single-shot noise or random fluctuations. Once the conditions are met, the system automatically shuts down and outputs a signal. This termination mechanism replaces the traditional method relying on manual experience or fixed-time judgments, achieving precise and reliable endpoint control based on the actual state of the leaves.
[0033] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A precise method for controlling leaf moisture distribution based on wilting, characterized in that, Includes the following steps: Step 1, Moisture Distribution Collection: Lay the wilted leaves to be harvested flat on the detection platform, scan the wilted leaves using a near-infrared hyperspectral imaging system, and obtain the spectral reflectance of each pixel in the 900-1700nm band. Based on the pre-established partial least squares moisture quantification model, convert the spectral data into moisture content values and generate a grayscale map of the moisture distribution of the wilted leaves with spatial coordinate information, where the grayscale value corresponds to the moisture content. Step 2, Moisture Region Segmentation: The grayscale image of moisture distribution is enhanced and binarized. Based on the overall average moisture content of the wilted leaves, the regions with moisture content higher than 15% of the average content are marked as high-moisture zones, the regions with moisture content lower than 15% of the average content are marked as low-moisture zones, and the regions in between are marked as medium-moisture zones. The area ratio of high-moisture zones, medium-moisture zones, and low-moisture zones, as well as the moisture gradient vector field between high-moisture zones and low-moisture zones, are calculated respectively. Step 3, wave parameter calculation: The area proportion of each moisture region and the average magnitude of the moisture gradient vector field are input into the pre-trained deep neural network model. The model outputs the basic wave intensity coefficient, wave cycle duration and cylinder forward and reverse switching frequency. The basic wave intensity coefficient is positively correlated with the area proportion of the high moisture region, and the wave cycle duration is positively correlated with the average magnitude of the moisture gradient vector field. Step 4, Adaptive wave-forming execution: The wave-forming machine is controlled to set the motor speed according to the basic wave-forming intensity coefficient, set the total time of a single wave-forming cycle according to the wave-forming cycle duration, and alternately change the rotation direction of the wave-forming cylinder according to the forward and reverse rotation switching frequency of the cylinder to dynamically impact and tumble the withering leaves; Step 5, Real-time Feedback Correction: During the wave execution process, Step 1 to Step 2 are repeated every 30 seconds to obtain the current moisture distribution grayscale map and the real-time change curve of the high water zone area ratio. When the real-time value of the high water zone area ratio decreases less than the preset expected decrease threshold compared to the initial value, the basic wave intensity coefficient is increased by one step increment until the real-time change curve of the high water zone area ratio converges to the expected decrease threshold.
2. The precise method for determining leaf moisture distribution based on wilting according to claim 1, characterized in that, In step one, before scanning, the withered leaves of the green leaf are subjected to background removal and leaf separation processing. The threshold segmentation method is used to extract individual leaves independently from the group image, and a corresponding grayscale sub-image of water distribution is generated for each independent leaf, and the spatial coordinates of each leaf are recorded.
3. The precise method for determining leaf moisture distribution based on wilting, as described in claim 2, is characterized in that... In step two, when calculating the water gradient vector field, the internal gradient field of each independent leaf is calculated for the grayscale sub-image of water distribution, and the proportion of gradient vectors with an angle greater than 60 degrees between the gradient direction and the leaf midrib direction in all independent leaves is counted. The proportion of the number is defined as the leaf margin-leaf center water difference index.
4. The precise method for determining leaf moisture distribution based on wilting, as described in claim 3, is characterized in that... In step three, the pre-trained deep neural network model includes an input layer, three hidden layers, and an output layer. The nodes in the input layer correspond to feature vectors, which are composed of the area proportions of high-water zones, medium-water zones, and low-water zones, the average magnitude of the moisture gradient vector field, and the moisture difference index between the leaf edge and the leaf center. The nodes in the output layer correspond to the basic wave-washing intensity coefficient, wave-washing cycle duration, and the frequency of forward and reverse rotation of the cylinder. The training dataset of the deep neural network model consists of multiple sets of historical wave-washing process parameters and their corresponding finished tea leaf uniformity scores, with the highest leaf uniformity score as the objective for supervised learning. When the moisture difference index between the leaf edge and the leaf center is greater than a preset critical threshold, the forward and reverse rotation frequency of the cylinder output by the model is forcibly set to 1.5 times the base frequency to increase the mechanical impact intensity on the leaf edge area and promote the accelerated conduction of moisture from the leaf edge to the leaf center. The base frequency is the standard switching frequency obtained by mapping the average magnitude of the moisture gradient vector field.
5. The precise method for determining leaf moisture distribution based on wilting according to claim 1, characterized in that, In step four, the adaptive wave-forming process also includes synchronously adjusting the air volume parameters inside the wave-forming machine according to the frequency of the cylinder's forward and reverse rotation. Specifically, during forward rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a first air volume to promote the dissipation of moisture from the surface of the wilting leaves. During reverse rotation of the cylinder, the auxiliary fan is controlled to deliver air into the cylinder at a second air volume, which is less than the first air volume, to reduce excessive water loss during the blade collision process. The ratio of the first air volume to the second air volume is proportional to the basic wave-forming strength coefficient.
6. The precise method for determining leaf moisture distribution based on wilting according to claim 1, characterized in that, In step five, the real-time feedback correction also includes a protection mechanism for low-moisture areas: during the wave-forming process, the real-time change curve of the area ratio of low-moisture areas is monitored simultaneously. When the real-time value of the area ratio of low-moisture areas increases by more than the preset over-dry threshold relative to the initial value, an intensity reduction command is immediately triggered, reducing the basic wave-forming intensity coefficient by two step increments, and simultaneously reducing the first air volume of the auxiliary fan to 60% of the initial first air volume, until the area ratio of low-moisture areas falls back to within a safe range.
7. The precise method for determining leaf moisture distribution based on wilting according to claim 1, characterized in that, It also includes step six, the termination judgment step: when the proportion of high water zone area of all independent leaves in the current moisture distribution gray map obtained by repeating steps one to two for three consecutive times is less than 10% of the initial high water zone area proportion, and the moisture difference index between leaf edge and leaf center is less than 0.1, the wave washing is judged to be over, the wave washing machine is controlled to stop running and output a wave washing completion signal.