Tea leaf segmented heating fixation system and method based on machine vision

Abstract

The invention discloses a machine vision-based tea leaf segmented heating fixation system and method, and relates to the technical field of tea leaf processing and automatic control. Comprehensive state information such as color, moisture, temperature and three-dimensional stacking thickness of tea leaves is obtained in real time through a multi-mode visual perception module; the central control and processing module carries out fusion processing on the data, dynamically divides a tea processing area into a plurality of micro-areas, and independently carries out self-adaptive precise segmented heating and physical parameter cooperative regulation and control on each micro-area in combination with a machine learning prediction model. And the advanced prediction model can perform continuous learning and self-evolution based on the evaluation result of the final finished product quality, so that the tea fixation quality and uniformity can be remarkably improved, the production efficiency and the automation level are improved, the process adaptability and flexibility to different raw materials and processing requirements are enhanced, and the method is suitable for large-scale popularization and application. And intelligent and traceable process control and continuous optimization oriented to final sensory quality are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of tea processing and automated control, and in particular to a system and method for segmented heating and fixing of tea leaves based on machine vision. Background Art

[0002] Tea leaf withering is a key process in the processing of green tea, oolong tea and other teas. Its purpose is to quickly passivate the activity of enzymes in fresh leaves through high temperature, inhibit the enzymatic oxidation of polyphenols, evaporate part of the water, soften the leaves, and emit a grassy smell, laying the foundation for the formation of excellent color, aroma and taste quality of tea. The traditional tea withering process is highly dependent on manual experience, and has problems such as high labor intensity, poor quality uniformity, and low production efficiency. With the development of technology, automation and intelligence have become important directions for the upgrading of tea processing equipment. Machine vision technology has shown broad application prospects in tea processing due to its non-contact, fast and objective advantages.

[0003] The specification of Chinese invention patent CN108782797B discloses a control method for frying tea leaves in an arm-type tea frying machine and an arm-type tea frying machine. The method takes real-time images of the tea leaves in the frying pan, analyzes the distribution shape of the tea leaves, calculates the rotation matrix if it is not circular, and controls the robotic arm to rotate a specific angle according to the rotation matrix to fry the tea leaves, in order to achieve the purpose of evenly frying the tea leaves and fully withering the green leaves. The invention uses machine vision to obtain the macroscopic spatial distribution information of the tea leaves and adjusts the frying action of the robotic arm accordingly, aiming to improve the uniformity of physical frying.

[0004] The specification of Chinese invention patent CN115736046B discloses an intelligent tea rolling machine based on machine vision and its intelligent control method. The invention is applied to the tea rolling process. Its intelligent control part includes machine vision, temperature and humidity precision control, oxygen precision control, etc. It uses machine vision to identify raw material characteristics, real-time collection of color and texture characteristics, and construction of a convolutional neural network model to determine the degree of rolling, thereby realizing intelligent regulation of the rolling operation. At the same time, the invention emphasizes the multi-dimensional and precise control of rolling process parameters such as temperature, humidity, air, pressure, time, and frequency to improve the rolling effect and product quality.

[0005] The above design achieves a certain degree of automation and intelligent control in the tea frying or rolling process through technologies such as machine vision, but there are still certain limitations. For the core thermal processing link of tea withering, which is extremely sensitive to changes in temperature, time, and material state, the shortcomings of the above-mentioned existing technologies are mainly reflected in: it is difficult to conduct comprehensive, multi-dimensional, accurate real-time perception of the core state of withering (such as internal moisture, surface temperature, actual stacking thickness and color changes), lack of effective means to dynamically and accurately divide micro-areas according to the real-time state of tea and apply differentiated heat control, lack of effective prediction capabilities for the withering process, and especially lack of adaptive learning and closed-loop optimization mechanisms linked to the quality of the final product. In terms of dynamic coupling and global collaborative optimization of multiple links such as heating, material transportation, and stir-frying, the degree of precision and intelligence needs to be improved.

[0006] Therefore, it is urgent to develop an advanced tea withering technology solution, which should be able to perceive the multi-dimensional state of tea leaves during the withering process (covering color, internal moisture, surface temperature, actual stacking thickness, etc.) in real time and accurately; and based on this comprehensive perception data, realize dynamic intelligent zoning of processing areas and adaptive, differentiated and precise control of heat in each zone; at the same time, the solution should also integrate effective predictive models and adaptive learning capabilities, through in-depth analysis of the processing process and associated learning with the quality of the final product, to achieve continuous optimization of the withering process and an overall leap in the level of intelligence, thereby effectively solving the shortcomings of current technology in withering uniformity, quality stability and depth of intelligence, and meeting the urgent needs of the modern tea industry for high-quality, high-efficiency and standardized production. Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies of the prior art and to propose a system and method for segmented heating and fixing of tea leaves based on machine vision to solve the above-mentioned problems.

[0008] The object of the present invention is achieved through the following technical solutions: a system and method for segmented heating and fixing of tea leaves based on machine vision, comprising: The tea conveying and turning module is used to carry, convey and turn the tea leaves to be withered. The tea conveying and turning module includes an adjustable speed conveyor belt and an adjustable speed and variable angle turning mechanism; A multimodal visual perception module is configured above or to the side of the tea conveying and stir-frying module. The multimodal visual perception module includes at least one high-resolution RGB industrial camera, at least one near-infrared spectrometer or hyperspectral camera, at least one infrared thermal imager, and a light source system that provides illumination for the imaging unit. It is used for real-time, non-contact collection of RGB image data, NIR spectral data, and thermal imaging data of tea leaves; An array-type segmented heating execution module, comprising multiple heating units with independently controllable power, energy output, or temperature and air volume. The heating units are arranged in an array and cover the processing area on the tea conveying and stir-frying modules; The central control and processing module is connected to the multimodal visual perception module and the array-type segmented heating execution module respectively. The central control and processing module includes: The multimodal data fusion and processing unit is used to receive the data collected by the multimodal visual perception module and perform image space registration, feature fusion, and abnormal noise filtering to obtain the leaf color characteristics, moisture content distribution characteristics, surface temperature distribution characteristics, and stacking thickness information of the tea leaves; A dynamic partitioning and heat map construction unit is used to dynamically divide the processing area into multiple independently controllable micro-zones based on the feature data obtained by the multimodal data fusion and processing unit, and to generate and update a dynamic heat map reflecting the fixing status of each micro-zone in real time; The zoned heating control unit is used to independently calculate the heating parameters for each micro-zone based on the withering status presented on the dynamic heat map and the preset withering target parameters, and output them to the corresponding heating unit in the array-type segmented heating execution module to achieve differentiated and adaptive heating of tea leaves in each micro-zone; A predictive fixing model and adaptive learning unit, which uses machine learning to combine historical fixing data with real-time data to predict the heat and heating time required to achieve the target fixing state for each micro-zone, optimize the heating strategy parameters online, and provide real-time warnings for abnormal conditions. The collaborative control and stir-fry linkage unit is used to comprehensively consider the heat conduction effect between adjacent micro-zones when outputting the heating parameters of the partitioned heating control unit, and collaboratively optimize the heating instructions of the target micro-zone. According to the uniformity of tea distribution or insufficient stirring detected by the multimodal visual perception module, the conveyor belt speed and the working parameters of the stir-frying mechanism in the tea conveying and stir-frying module are automatically adjusted to ensure that the tea is evenly heated.

[0009] The multimodal visual perception module further includes a 3D vision sensor or a structured light sensor, which is used to obtain the stacking thickness distribution information of tea leaves on the tea conveying and stir-frying module. The stacking thickness distribution information is involved in the dynamic micro-area division and heating decision-making of the central control and processing module.

[0010] The heating units in the array-type segmented heating execution module are one or more of an infrared heating lamp with independently adjustable power, a microwave feed port with independently controlled energy output, or a hot air nozzle with independently controlled temperature and air volume, and the heating range of each heating unit corresponds to at least one micro-zone dynamically divided by the central control and processing module.

[0011] The multimodal data fusion and processing unit in the central control and processing module is further used to extract the average chromaticity value and chromaticity uniformity index of tea leaves from RGB image data, extract the average moisture content and moisture content standard deviation of the tea surface from NIR spectral data, and extract the average temperature and maximum temperature point of each micro-area from thermal imaging data. The indicators participate in the dynamic partitioning and thermal map construction unit to construct the withering thermal map and make partition heating decisions.

[0012] The dynamic partitioning and heat map construction unit in the central control and processing module adopts multi-factor threshold segmentation and clustering algorithm when performing dynamic partitioning to realize dynamic division of micro-areas based on parameters such as tea distribution, temperature gradient, moisture content distribution, and stacking thickness. The physical boundaries of the micro-areas are adjusted in real time as the tea leaves move or their state changes.

[0013] The predictive killing model in the central control and processing module and the machine learning-based predictive killing model adopted by the adaptive learning unit are jointly trained through historical killing data and real-time collected data, and perform online parameter optimization during system operation. They can predict the heating power curve and heating time required for each micro-area based on the current multimodal visual features.

[0014] The predictive tea-withering model is based on a deep neural network, and the model parameters are continuously optimized through the process data of each batch of tea withering and the corresponding finished product quality evaluation results.

[0015] The collaborative control and stir-fry linkage unit in the central control and processing module, when collaboratively optimizing the heating instructions of the target micro-area, explicitly sets preventing over-killing or under-killing of the target micro-area or its adjacent micro-areas as one of the optimization goals.

[0016] The system is further equipped with a parameter setting interface and a human-computer interaction terminal, allowing the operator to input parameters such as tea variety, target withering degree, initial micro-zone number reference value and partition adjustment sensitivity, and can display dynamic heat maps, micro-zone status parameters, historical data trends and real-time alarm information in real time. All key process data are automatically archived for tea quality traceability.

[0017] The segmented heating and fixing method of tea leaves based on machine vision comprises the following steps: S1. Use a multimodal visual perception module to collect RGB image data, NIR spectrum data, thermal imaging data, and stacking thickness distribution information of tea leaves in real time; S2, the multimodal data fusion and processing unit of the central control and processing module receives and processes the data collected in step S1, performs image spatial registration, feature fusion, and abnormal noise filtering to obtain the leaf color characteristics, moisture content distribution characteristics, surface temperature distribution characteristics, and stacking thickness information of the tea leaves; S3. The dynamic partitioning and heat map construction unit of the central control and processing module dynamically divides the processing area into multiple micro-zones based on the feature data obtained in step S2, constructs a real-time heat map of the fixing status, and continuously adjusts the micro-zone division according to the analysis results and preset parameters; S4. For each micro-zone, the zoned heating control unit of the central control and processing module independently calculates the heating parameters based on the status of the dynamic heat map and the preset withering target, and instructs the corresponding heating unit in the array-type segmented heating execution module to implement differentiated heating of the tea leaves in that micro-zone. At the same time, the collaborative control and stir-fry linkage unit collaboratively optimizes the heating strategies of adjacent micro-zones. S5. The coordinated control of the central control and processing module and the stir-frying linkage unit continuously monitor and automatically adjust the stir-frying and delivery parameters of the tea conveying and stir-frying module based on the uniformity of tea distribution or insufficient stirring detected by the multimodal visual perception module to ensure uniform distribution and heating of the tea leaves; S6. Repeat steps S2 to S5 until the central control and processing module determines that all micro-area tea leaves have reached the set withering standard, thus completing the withering process.

[0018] The beneficial effects of the present invention are: 1. This invention achieves refined, adaptive heating during the tea withering process through multimodal visual perception (covering color, moisture content, temperature, and precise 3D stacking thickness) and dynamic micro-area precision control. The deep neural network continuously learns based on feedback from the final product quality. The system can ensure the uniformity and thoroughness of the tea withering process to the greatest extent possible, effectively avoiding defects such as burnt edges, red stems, and dull yellowing, and stabilizing and enhancing the aroma, flavor, and color of the tea.

[0019] 2. The system's high degree of automation and intelligence (including automatic data collection, dynamic zoning, intelligent heating decisions, and coordinated stir-frying and conveying) significantly reduces the reliance on the manual experience of skilled technicians in the traditional tea-withering process and reduces the uncertainty of human operation. This not only improves production efficiency and processing throughput, but also makes standardized and large-scale tea production possible, solving the industry's pain point of a shortage of highly skilled tea-withering masters.

[0020] 3. With precise multi-dimensional data perception, especially direct 3D thickness measurement and dynamic partitioning algorithms such as multi-factor threshold segmentation and clustering, and the powerful generalization and adaptive learning capabilities of deep learning models, the present invention can effectively cope with the high uncertainty and volatility of fresh leaf raw materials in terms of variety, age, moisture content, stacking status, etc. The system can "tailor" the best withering solution for tea leaves with different characteristics.

[0021] 4. By introducing a predictive model based on deep neural networks and establishing a continuous learning closed loop with final product quality evaluation as the core feedback, the system goes beyond simple parameter optimization and has the self-evolutionary capabilities of "experience accumulation" and "skill improvement". It can learn from each production, continuously optimize its killing strategy, and independently explore and solidify the optimal process.

[0022] 5. The human-machine interface (HMI) and comprehensive data management functions provide operators with comprehensive real-time process monitoring, parameter adjustment, and historical data analysis capabilities. All key process data and control decisions are automatically recorded and archived, which not only greatly enhances the transparency and controllability of the production process, but also provides a solid data foundation for comprehensive product quality traceability, continuous process improvement, and scientific research. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a system architecture diagram of the present invention; Figure 2 It is a timing diagram of the present invention; Figure 3 Flowchart of the present invention. DETAILED DESCRIPTION

[0024] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0025] It is to be noted that the directions of "left", "right", "up", "down", "front", "back", "inside" and "outside" in the following schemes are all relative directions and are not listed here one by one.

[0026] Example 1: like Figures 1 to 3 As shown, this embodiment discloses a segmented heating and withering system for tea leaves based on machine vision and its working method. The system can perform intelligent and refined closed-loop control of the withering process of tea leaves to ensure the uniformity and high quality of withering. The core of the system lies in its multimodal perception capability, dynamic partition processing logic, and adaptive heating and physical parameter adjustment mechanism.

[0027] In this embodiment, the tea conveying and stir-frying module adopts a stainless steel mesh belt conveyor with a width of 1 meter and a length of 5 meters. Its conveying speed can be accurately adjusted between 0.1 m / min and 1.0 m / min by the central control and processing module according to the withering requirements. Above the conveyor belt, three groups of speed-adjustable and angle-variable stir-frying mechanisms are evenly arranged, such as a multi-axis rotating paddle, whose speed can be adjusted within the range of 10 to 60 rpm, and the angle between the paddle and the conveyor belt plane can be adjusted between 15 and 45 degrees to adapt to the turning requirements of tea leaves with different tenderness and moisture content, ensuring that the tea leaves can be fully and evenly spread and turned during the conveying process to avoid local accumulation or long-term single-sided heating.

[0028] Following closely behind or configured above the key observation points on the conveyor belt is the multimodal visual perception module, which specifically includes: a 5-megapixel RGB industrial camera for capturing the color, gloss and shape of the tea leaves; a near-infrared hyperspectral camera covering the 900-1700 nanometer band for detecting the moisture content information on the surface and inside of the tea leaves; and a 320x240 pixel resolution infrared thermal imager for monitoring the real-time temperature distribution on the surface of the tea leaves. To ensure imaging quality, the module is also equipped with a light source system consisting of multiple groups of high-uniformity LED flat panel lights to provide stable and uniform lighting for RGB and near-infrared imaging. These sensors work together to perform continuous, non-contact scanning of the tea leaves on the conveyor belt, collecting multi-dimensional data in real time.

[0029] The heating system is the key execution part of this embodiment, namely the array-type segmented heating execution module, which is specifically composed of 50 independently controllable heating units arranged in an array of 5 rows and 10 columns above the conveyor belt, covering the main processing area. These heating units can use far-infrared ceramic heating lamps, each lamp has a power of 500 watts, and its radiation intensity can be precisely controlled from 0 to 100% by the central control and processing module by adjusting the input voltage or PWM (pulse width modulation) signal. Alternatively, a microwave feed array can be used, each feed can independently control the microwave output energy. Another optional solution is to use partitioned hot air nozzles, each nozzle can independently adjust the outlet air temperature (for example, between 120°C and 280°C) and air volume. The physical heating area responsible for each heating unit corresponds to the subsequently dynamically divided micro-zone or can be combined to correspond, ensuring the feasibility of refined and differentiated heating of tea leaves.

[0030] The core central control and processing module is a high-performance industrial computer that deploys the core algorithm of this invention. The five units under this module work together: The multimodal data fusion and processing unit is responsible for receiving the raw data stream from the vision module. The unit first performs image spatial registration on the RGB, NIR and thermal imaging data to ensure that the data from different sensors can accurately correspond to the same physical location of the tea leaves. It then performs feature fusion and abnormal noise (such as sudden changes in illumination and water vapor interference) filtering. The key is that it extracts the average Lab chromaticity value and chromaticity standard deviation (reflecting uniformity) of the tea leaves from the RGB image, and inverts the average moisture content and moisture content standard deviation of the tea surface based on the intensity of specific moisture absorption peaks (such as around 1450nm) from the NIR spectral data. The average temperature and maximum temperature point of each potential micro-area are extracted from the thermal imaging data. In addition, the stacking thickness information of the tea leaves can be indirectly inferred through texture analysis of the RGB image or combined with the attenuation characteristics of the NIR signal.

[0031] The dynamic partitioning and heat map construction unit receives the multi-dimensional feature data after the above processing. It first dynamically optimizes the processing area into multiple micro-areas of variable shape and size based on the preset initial grid (for example, dividing the processing area into 10x5 initial blocks) and real-time feature data, such as the actual distribution boundary of the tea leaves, moisture content gradient, temperature difference, leaf color difference and stacking thickness. For example, an area with a significantly high moisture content will be divided into a micro-area separately. At the same time, the unit generates and updates a multi-dimensional "withering status heat map" in real time. The map not only includes temperature, but may also use different colors or numerical values ​​to mark the moisture content, leaf color status, etc. of each micro-area, providing an intuitive basis for subsequent control.

[0032] The zoned heating control unit independently calculates the heating power and action time required for each dynamically divided micro-zone based on its current withering status presented on the heat map (such as leaf color L value, moisture content percentage, current temperature) and the preset target withering parameters of the tea variety (such as target L* value range, target moisture content range). For example, for a micro-zone with green leaf color, high moisture content and low temperature, a higher heating instruction will be output.

[0033] The predictive withering model and adaptive learning unit in this embodiment adopt a machine learning model based on the gradient boosting decision tree (GBDT). The model uses a large amount of historical withering data (including the initial state, process parameters, heating strategy and final withering results of different batches of tea) for offline training. In real-time work, it will combine the real-time feature data of each current micro-area to predict the heat and approximate time required for the micro-area to reach the ideal withering state (such as complete inactivation of enzyme activity and reduction of moisture content to a specific value). At the same time, the system will make small online parameter adjustments to the model weights or decision thresholds based on the deviation between the actual withering effect and the prediction. If it is detected that the temperature of a micro-area rises too fast or the moisture content drops abnormally, an early warning will be triggered.

[0034] The collaborative control and stir-fry linkage unit is responsible for overall coordination. When outputting heating instructions, it will take heat conduction into consideration. For example, if the heating power of one micro-zone is very high, the heating power of its adjacent micro-zone (especially the micro-zone downstream in the conveying direction) may be appropriately lowered in advance. More importantly, if the RGB camera detects that the tea leaves are piled too thickly in certain areas or have not been turned over for a long time (for example, by analyzing texture changes or color uniformity), the unit will immediately instruct the tea conveying and stir-frying module to moderately reduce the conveyor belt speed and increase the rotation speed of the stir-frying mechanism in the corresponding area or adjust its paddle angle to promote the dispersion of the tea leaves and even heating.

[0035] In addition to multimodal vision, a physical pressure sensor array can be optionally installed to detect the local stacking thickness and leaf compression in real time, thereby improving the accuracy of thickness information. A radio frequency / capacitive moisture detection module can be used as an alternative or supplement to NIR. The heating unit can adopt an ultrasonic heating array, which uses ultrasonic energy to quickly heat the inside of the tea leaves in a specific area, with a more sensitive response; or a fiber laser heating dot matrix can be used to achieve point-to-point precise heat treatment of micro-areas, which is suitable for high-end tea withering that requires extremely high spatial resolution. The liquid convection cooling / micro-mist cooling module can achieve local cooling and anti-scorching compensation in extremely high temperature abnormalities.

[0036] An electromagnetically driven flexible paddle can be used to assist in stir-frying and evenly distributing materials. This can achieve precise angle and force changes, reduce mechanical wear, and improve stir-frying uniformity. A local micro airflow generator can assist in local material distribution.

[0037] Optional intelligent voice / touch interactive terminal allows operators to adjust process sensitivity, zoning parameters and alarm response levels through natural language or graphical interface, supports remote process operation and maintenance and big data analysis interface, and provides interconnected intelligent management and control for large-scale multi-location factories.

[0038] Working process (S1) Tea import and multimodal real-time perception: Freshly picked tea leaves, such as a batch of Longjing spring tea buds with a moisture content of about 75%, are evenly spread on the stainless steel mesh conveyor belt of the tea conveying and stir-frying module through an automatic loading device. The conveyor belt starts smoothly at a preset initial speed (for example, 0.3 meters per minute) and sends the tea leaves to the withering processing area. Once the tea leaves enter the perception area above the conveyor belt, the multimodal visual perception module is immediately activated. At least one high-resolution RGB industrial camera in the module begins to continuously capture real-time images of the tea leaves, recording their initial bright green color, leaf shape and glossiness; at the same time, at least one near-infrared hyperspectral camera A sensor (or spectrometer) scans the tea leaves on the conveyor belt at a specific frequency, collecting spectral data in the range of 900-1700nm, providing a basis for subsequent analysis of moisture content and changes in some internal chemical components. At the same time, at least one infrared thermal imager continuously records the temperature distribution map of the tea leaves' surface. The raw RGB image data, NIR spectral data, and thermal imaging data acquired by all these sensors, together with the imaging parameters under stable lighting conditions provided by the light source system, and the initial stacking thickness of the tea leaves indirectly assessed through image analysis, are all transmitted to the central control and processing module in real time and contactlessly at a high frame rate.

[0039] (S2) Multi-dimensional feature extraction and deep data fusion: After the data arrives at the central control and processing module, its internal multimodal data fusion and processing unit first performs accurate image space registration on the multi-source heterogeneous data from different sensors to ensure that in subsequent analysis, a pixel point of the RGB image can accurately correspond to the same physical location of the NIR spectrum data and thermal imaging data. Then, the system executes the feature fusion algorithm and abnormal noise (such as ambient light interference and artifacts caused by instantaneous occlusion of water vapor) filtering. The unit will accurately calculate a series of key feature indicators from the registered and purified data: for example, from the RGB image data, it not only extracts the average CIELab chromaticity value of the tea leaves ( L represents brightness, a represents redness and greenness, and b represents yellowness and blueness). The standard deviation of chromaticity within a specific area is also calculated as a quantitative indicator of color uniformity. From NIR spectral data, by analyzing the absorbance or reflectance of specific moisture absorption peaks (such as those near 1450nm and 1180nm), combined with a pre-built calibration model, the average moisture content of the tea surface and the standard deviation of the moisture content in its spatial distribution are accurately inverted. From thermal imaging data, the average surface temperature, the highest temperature point, and the uniformity index of the temperature distribution are extracted for each discernible small area on the conveyor belt. These multi-dimensional features together constitute a comprehensive digital description of the current state of the tea.

[0040] (S3) Dynamic intelligent partitioning and real-time status heat map construction: Based on the above multi-dimensional feature data, the dynamic partitioning and heat map construction unit starts to work. It does not use fixed physical partitions, but logically and dynamically divides the entire processing area into several "micro-zones" of variable shape and size according to the real-time perception of the tea state. The division is based on the spatial continuity and difference of the feature data: for example, a piece of tea that has just entered the heating zone, has a generally high moisture content (such as 70-75%), a bright green leaf color (a value is significantly negative), and a low temperature (such as 30-40℃) will be divided into several larger initial micro-zones as a whole or subdivided; if the tea leaves on the conveyor belt are accidentally piled up and the stacking thickness increases significantly, Or some places may heat up particularly rapidly due to being close to the edge of the heat source. These areas may be divided into smaller micro-areas that require special attention. This unit will generate and continuously update a multi-layer "withering status heat map" in real time. This heat map not only intuitively displays the temperature distribution of each micro-area in pseudo-color, but may also superimpose key information such as moisture content contour lines and leaf color change trend indicators (such as the rate of change of the a value), providing visual decision support and data basis for subsequent precise regulation. The boundaries of the micro-areas will be dynamically adjusted as the tea leaves move on the conveyor belt and evolve in their own state (such as water evaporation and color change) to ensure that the state of the tea leaves in each micro-area is as uniform as possible.

[0041] (S4) Differentiated heating and coordinated optimization control: For each dynamically formed micro-zone, the zoned heating control unit takes on the core regulation task. It will calculate the optimal heating parameters independently and in real time based on the specific state parameters of the micro-zone presented on the real-time withering state heat map (for example, micro-zone X: current average temperature 95℃, average moisture content 68%, average a value -15) and the preset ideal withering curve and final target parameters of the batch of tea (for example, the first stage target: leaf temperature reaches 160℃, moisture content drops to 65%; the final target: leaf temperature stabilizes within a specific range, leaf color a value is close to 0, and moisture content is about 60%). In combination with the predictive withering model and the adaptive learning unit's prediction suggestions on the heat and time required for the micro-zone to achieve the next stage target, the optimal heating parameters are calculated independently and in real time. These parameters are then converted into specific control instructions. The array-type segmented heating execution module is provided with several heating units that precisely correspond to the micro-zone (such as the power percentage of the infrared lamp, or the temperature and wind speed settings of the hot air nozzle). For example, for micro-zones where the moisture content is still relatively high and the temperature rise is insufficient, the system will instruct the corresponding heating unit to increase the power or increase the temperature; and for micro-zones that have approached or reached the current stage target, the heating intensity will be reduced or temporarily shut down to achieve true differentiation and adaptive heating. At the same time, the collaborative control and stir-fry linkage unit will intervene to ensure that the heating between micro-zones will not have adverse mutual effects. For example, if a micro-zone is undergoing strong heating, the initial heating instruction of the micro-zone immediately downstream in its conveying direction may be moderately lowered to pre-compensate for possible heat drift and cumulative effects, and prevent the downstream tea from heating up too quickly and prematurely.

[0042] (S5) Intelligent linkage adjustment of physical conveying and stir-frying: During the entire withering process, heating control and physical transportation are not separated. The collaborative control and stir-frying linkage unit continuously monitors the physical state of tea leaves on the conveyor belt in real time through the image data fed back by the multimodal visual perception module (especially RGB images and stacking information analyzed by them). Once the visual analysis algorithm detects that tea leaves in certain areas are obviously accumulated (for example, the brightness of local areas is abnormally low, and the texture features appear as clumps), or a large area of ​​tea leaves has not been effectively turned over for a long time (for example, the temperature difference between the upper and lower surfaces of leaves in an area is too large or the color changes are inconsistent), the unit will take action immediately and instruct the tea leaves to be transported and stirred. Module: Targetedly increase the rotation speed of the stir-frying mechanism above the stacking area or adjust the turning angle and depth of its paddle to quickly break up the tea clumps; if it detects that the overall material is too thick or the tea is prone to uneven distribution during its forward movement, the system may decide to moderately reduce the overall speed of the conveyor belt to give the stir-frying mechanism more time to work, or use more complex stir-frying combination actions to improve the uniformity of the tea spreading. Under certain preset conditions, the system can even issue an alarm to prompt manual intervention to improve the uniformity of the feeding. This series of linkages ensures that the tea can be evenly heated at the microscopic level, creating a good physical prerequisite for the uniformity of the chemical reaction.

[0043] (S6) Closed-loop iterative optimization until completion: The system repeats the precise closed-loop control process of "perception-feature extraction-dynamic partitioning and heat map update-decision-making and regulation (heating and physical handling)-effect feedback" from S2 to S5 at a very high frequency. Every time the tea leaves move a short distance on the conveyor belt, their status will be re-evaluated, the heat map and micro-area division will be dynamically refreshed, and the heating strategy and stir-frying parameters will also be adaptively adjusted accordingly. This is a process of continuous learning and optimization until the central control and processing module comprehensively judges that the tea leaves in the vast majority of micro-areas on the conveyor belt (for example, the threshold is set to 95% of the processing area) have reached the preset withering level. Completion standards - for example, the leaf color completely changes from bright green to uniform dark green (Lab* value enters the target range), the average moisture content drops to the target value (such as 58-62%), there is no scorching on the leaf edges and no reddening in the center, and certain key precursors related to aroma formation (such as amino acids and soluble sugars) shown by NIR specific band analysis have been moderately converted and not excessively lost. At the same time, the leaves are overall soft and elastic, the grassy smell is completely dissipated, and the aroma of cooked chestnuts or tofu pudding is revealed - the system will determine that the entire withering process has been completed. At this time, the tea will leave the system from the discharge port and enter the subsequent rolling, drying and other processes.

[0044] Specific application cases Taking the premium Longjing fresh leaves as an example, the requirements for withering are extremely strict. They must be killed thoroughly and evenly to completely destroy the enzyme activity and prevent red stems and leaves. At the same time, their bright green color and rich tofu pudding and chestnut aroma must be retained to the greatest extent possible to avoid burnt edges, bursting points or dull yellowing. Traditional hand-fried Longjing is extremely dependent on the experience of the master craftsman.

[0045] The system of this embodiment is used to fix the green color of Longjing tea: Fresh leaves (about 75-78% moisture content) are fed into the system.

[0046] (S1-S2) The multimodal vision system immediately captures its tender green color, high moisture content, and room temperature. The RGB camera accurately records the initial color (for example, a high L value, a negative a value with a large absolute value), and the NIR spectrometer focuses on monitoring the moisture absorption peak near 1450nm.

[0047] (S3) The system dynamically divides the fresh leaves into micro-areas based on their initial uniformity (even within the same batch of fresh leaves there may be local differences) and spreading. The heat map shows that the entire system is in a "waiting to be finished" state.

[0048] (S4) In response to the temperature requirements of "high temperature and rapid frying, high temperature first and then low temperature" for the withering of Longjing tea, the zoned heating control unit will issue higher initial heating instructions to all micro-zones. For example, it controls the infrared heating lamps or hot air nozzles to quickly raise the leaf temperature to 200-250℃ (the surface temperature is monitored by a thermal imager and the leaf core temperature is estimated by combining the model). At this time, the prediction model will estimate the time required for rapid water loss in the first stage based on the tenderness and initial moisture content of the fresh leaves.

[0049] As the leaf temperature rises and water evaporates, the NIR data shows that the moisture content drops rapidly, and the RGB data shows that the leaf color begins to turn dark green. The central control module will dynamically lower the heating intensity of the corresponding micro-area based on these changes, and switch to the lower temperature (about 120-150°C) stage of "strip shaping". If a micro-area is slightly thicker due to slow water loss and slow leaf color conversion, the system will maintain the heating intensity of the micro-area or slightly increase it, and at the same time link the stir-frying mechanism to strengthen the stirring of this area.

[0050] (S5) During the entire process, if the vision system detects that leaves are sticking together or forming small clumps, the angle and speed of the stir-frying mechanism will be adaptively adjusted to break them up. The conveying speed will also be fine-tuned according to the overall withering progress and tea condition.

[0051] (S6) Finally, when the system detects that the tea leaves in most micro-areas have reached the target state: the leaf color is uniform dark green (for example, the L value reaches the set threshold, the a value is close to 0 or slightly positive), the moisture content is reduced to about 60-62%, the tea leaves feel soft, the grass smell is dissipated, and a unique fragrance is emitted, the withering process is determined to be complete.

[0052] Traditional tea withering control often relies on a single parameter (such as temperature) or isolated threshold judgments on multiple parameters, and performs weighted fusion and nonlinear mapping of key features from multimodal visual perception (color, moisture content, temperature) and physical morphological features (such as stacking thickness and uniformity) to generate a single scalar value that can more comprehensively, accurately, and dynamically characterize the current comprehensive degree of tea withering in real time.

[0053] The dynamic integrated fixing index (IDIFI) is the smallest perceptible unit (or dynamically formed micro-zone Z) in the processing area. j ) At time t, its dynamic comprehensive completion index I DIFI (Zj,t) is defined as: I DIFI (Z j ,t): micro-area Z j The dynamic comprehensive withering index at time t has a design target value range of [0,1], where 0 represents the initial state of fresh leaves and 1 represents the ideal withering state.

[0054] k: represents the index of the key state feature, the set is {C, M, T}, corresponding to color (Color), moisture (Moisture), and temperature (Temperature) respectively.

[0055] w k (t): The dynamic weight coefficient of the k-th feature at time t (or different stages of completion), ∑w k (t) = 1, for example, at the beginning of the fixing period, the temperature is w T and moisture content w M The weight may be higher to ensure rapid heating and dehydration; in the late stage of fixing, the color w C The weights may be increased to precisely control the final sensory qualities, and these weights can be preset or adaptively adjusted by the system according to the tea variety and processing stage.

[0056] : Micro-area Z j The average measurement of the k-th feature at time t.

[0057] : Average color characteristic value (for example, the a* value in the CIEL*a*b* space or the integrated color difference ΔE obtained by analyzing RGB data).

[0058] : Average moisture content (obtained by analyzing NIR spectral data).

[0059] : Average surface temperature (obtained by analyzing infrared thermography data).

[0060] S k,target : The ideal target value of the kth feature (the preset completion standard).

[0061] S k,initial : Initial fresh leaf reference value of the kth feature.

[0062] Φ k (⋅): The normalized state contribution function of the k-th feature, which is a nonlinear function used to map the current value of the feature to its contribution to the overall degree of completion (range [0,1]), For example, for the color Φ C : When the color of tea leaves changes from bright green (such as ≈−20) toward the target dark green (e.g. ≈−2) when Φ C From 0 to 1, it can be designed as an S-shaped function or a Gaussian cumulative function to reflect the nonlinear characteristics of color change and the sensitivity near the target value.

[0063] (This is an example sigmoid function, σ C Control the steepness of the curve and adjust it to S C,initial When it is close to 0, C,target close to 1).

[0064] For the moisture content Φ M : When the moisture content drops from the initial value to the target value, Φ M From 0 to 1.

[0065] For temperature Φ T : When the temperature reaches and maintains the optimal temperature range for killing green, Φ T It tends to be 1, and the value is lower if it is lower or significantly higher than this range.

[0066] Ψ Th ( (Z j ,t)): Stack thickness penalty factor, (Z j ,t) is the micro-zone Z j The average stack thickness at time t (obtained indirectly by the 3D sensor or image texture) is designed so that when the thickness exceeds a certain optimized value, the factor is less than 1, and the value of I DIFI The inhibition was carried out, reflecting the adverse effects of thick stacking on the efficiency and uniformity of fixing.

[0067] Among them Th ideal is the ideal paving thickness, λ Th is the thickness penalty coefficient.

[0068] Ψ U (σ Sk (Z j ,t)): Micro-area internal uniformity factor, σ Sk (Z j ,t) represents the micro-zone Z j Internal key features S k The standard deviation (i.e., unevenness) of the internal state (such as temperature, color) is very uneven, and this factor is less than 1. DIFI Correction shows that even if the average state meets the standard, if the internal variation is too large, the overall quality of the finished product is still not optimal: ; where κ k is the influence weight of the heterogeneity of each feature, norm(σ Sk ) is the normalized standard deviation.

[0069] Through multi-dimensional, real-time and precise perception of the tea state and dynamic micro-area division and differentiated heating based on this, the problem of uneven withering caused by factors such as different ages of fresh leaves, uneven thickness of spread leaves, and inaccurate pot temperature control in the traditional withering process has been effectively solved. The tea leaves in each micro-area can receive nearly "customized" heat treatment to ensure highly consistent withering effects and precise control of key physical and chemical indicators (leaf color, moisture content, temperature). It can better passivate enzyme activity, while reducing excessive damage to chlorophyll and excessive oxidation of tea polyphenols, thereby better preserving the color, aroma and taste of the tea. For example, in the case of Longjing tea, it can effectively avoid the production of red stems, burnt edges, and bursting points. The finished tea is emerald green, fragrant, and mellow, and the quality stability between batches is also greatly enhanced. , automated and continuous closed-loop control replaces heavy and experience-dependent manual operations, reduces labor intensity, and reduces the impact of human factors on product quality. The predictive model and adaptive learning capabilities enable the system to cope with fluctuations in different raw materials, improving the overall operating efficiency and intelligence level of the production line. Precise zone heating avoids continuous overheating of areas that have met the withering requirements, thereby saving energy. At the same time, due to the improvement and stabilization of withering quality, the loss of raw materials caused by withering failure or substandard quality is reduced. The system records various sensor data, control parameters and decision-making processes throughout the withering process, providing valuable data support for subsequent process analysis, quality traceability and further optimization of the withering model.

[0070] To sum up, this embodiment, by constructing a core system that integrates multimodal perception, dynamic partitioning, adaptive heating, intelligent prediction and collaborative control, and supplemented by specific method processes, can significantly improve the uniformity, quality and intelligence level of tea withering, and provide strong technical support for the standardized and large-scale production of tea.

[0071] Example 2: like Figures 1 to 3 As shown, the tea withering system disclosed in this embodiment inherits the core architecture described in Example 1, including a tea conveying and stir-frying module, a multimodal visual perception module, an array-type segmented heating execution module, and a central control and processing module and its internal five core functional units. On this basis, this embodiment introduces more advanced sensing means, more sophisticated algorithms, and more complete human-computer interaction and management functions, so that the system's understanding and regulation of the tea withering process has reached a new level, and is particularly good at handling complex situations with uneven raw material states.

[0072] Compared with Example 1, this embodiment has made a key upgrade in the multimodal visual perception module. Based on the original RGB camera, near-infrared hyperspectral camera and infrared thermal imager, this module further integrates a high-precision 3D vision sensor, such as a line scanning 3D profile sensor or a structured light sensor based on the laser triangulation principle. The 3D sensor is precisely calibrated and triggered synchronously with other visual sensors. It can obtain the precise three-dimensional profile of the tea leaves on the conveyor belt in real time and non-contact, thereby directly calculating the stacking thickness distribution information of the tea leaves at various positions with an accuracy of up to millimeter level. Compared with the indirect inference of thickness information in Example 1, this provides more direct and reliable physical dimension data. The acquired stacking thickness data stream is time-stamped together with the RGB, NIR and thermal imaging data and sent to the central control and processing module.

[0073] Within the central control and processing module, its multimodal data fusion and processing unit can now directly integrate this precise stacking thickness information as another key input feature dimension alongside leaf color, moisture content, and surface temperature. The functionality of the dynamic partitioning and heat map construction unit has been significantly enhanced. It no longer relies solely on general feature gradient analysis for dynamic partitioning, but instead explicitly employs multi-factor threshold segmentation and advanced clustering algorithms. For example, the system may employ an adaptive iterative threshold segmentation method. First, based on multiple parameters such as the tea's real-time temperature gradient, moisture content change rate, the degree of deviation of the leaf color Lab* value from the target, and stacking thickness directly measured by the 3D sensor, it preliminarily identifies the boundaries of regions with significant state differences. Subsequently, it may apply methods such as DBSCAN (Density-Based Spatial Clustering of Applications with Clustering algorithms such as K-Means or improved K-Means cluster pixels or small blocks with similar feature vectors (composed of temperature, moisture content, chromaticity, thickness, etc.) and spatial proximity into dynamic "micro-regions". This algorithm can more intelligently identify and outline micro-regions with irregular shapes but uniform states. The physical boundaries of these micro-regions will adjust smoothly and in real time as the tea leaves move on the conveyor belt or their internal state evolves (such as thickness reduction due to water evaporation), ensuring that the partitions always accurately reflect the actual state of the tea leaves.

[0074] In addition, the collaborative control and stir-fry linkage unit has clearer and more precise goals when performing collaborative optimization of heating instructions. When adjusting the heating instructions of the target micro-area, the unit will explicitly prevent the target micro-area or its adjacent (especially upstream or downstream in the conveying direction) micro-areas from local over-killing (such as burnt edges, bursting points) or under-killing (such as red stems, green leaves) as one of its core optimization goals. This means that the system not only considers the physical phenomenon of heat conduction, but also combines the thermal sensitivity of tea in this state, as well as the risk thresholds of over-killing or under-killing (these can be read from the process parameter library of the tea variety) to perform forward-looking, prevention-oriented power fine-tuning.

[0075] Finally, the system of this embodiment is further equipped with a comprehensive parameter setting interface and human-machine interface terminal (HMI), which is generally manifested as a 15-inch industrial-grade touch screen installed on the control cabinet. The user-friendly interface allows operators to use this terminal to make detailed settings before production. For example, the operator can select the tea variety currently being processed (such as "West Lake Longjing" or "Biluochun Grade 1"), and the system will automatically load the corresponding recommended withering process parameter curve; set the target withering degree (such as "70% withering" or "withering thoroughly and fully dried"); enter the reference value for the initial number of micro-zones and the sensitivity of the dynamic adjustment of the partitions (for example, a high sensitivity will result in refined partitions even with small state differences, while the opposite will result in relatively rough partitions). During operation, the HMI displays real-time dynamic thermal maps (including multi-layer visualization of temperature, moisture content, leaf color, and other information), status parameters of key micro-zones (such as average temperature, maximum temperature, average moisture content, leaf color Lab* value, etc.), historical data trend charts (such as the change curve of average leaf temperature over a period of time), and any real-time alarm information (such as a heating unit failure, severe tea accumulation, etc.). All of this critical process data, including sensor readings, control instructions, micro-zone evolution history, alarm records, etc., will be automatically archived and stored, and can be associated with batch information, providing detailed data support for subsequent tea quality traceability, process analysis, and system performance optimization.

[0076] Working process The workflow of this embodiment is more sophisticated and efficient due to the introduction of new technical features based on the first embodiment: (S1) Tea import and enhanced multimodal real-time perception: After the tea leaves enter the conveyor belt, in addition to the RGB, NIR, and thermal imaging sensors, the newly added 3D vision sensor is activated simultaneously to accurately scan the three-dimensional contours of the tea leaves' surface and output a high-resolution stacking thickness distribution map in real time. The data of all four modes are fused and timestamped and sent together to the central control and processing module.

[0077] (S2) Multidimensional feature extraction including accurate thickness information: While performing the operations of image registration, feature fusion, and noise filtering described in Example 1, the multimodal data fusion and processing unit of the central control and processing module incorporates the directly measured stacking thickness data as an independent, high-weight basic feature into subsequent analysis. At this time, the system's description of the tea state is expanded from a two-dimensional plane to a three-dimensional space.

[0078] (S3) Refined dynamic partitioning and heat map construction based on multi-factor algorithms: The dynamic partitioning and heat map construction unit starts the multi-factor threshold segmentation and clustering algorithm, which comprehensively considers leaf color characteristics, moisture content distribution, surface temperature distribution and the now very accurate stacking thickness information. For example, if the RGB data of an area shows that the leaf color is normal, the NIR data shows that the moisture content is moderate, the thermal imaging shows that the temperature is uniform, but the 3D sensor shows that its stacking thickness is far above the average, the algorithm will determine that this is a potential "stuffy green" risk area that requires special attention, and will independently divide it into a micro-area. Even if the other two-dimensional features are not much different, the boundary of the micro-area will be adjusted in real time and smoothly at the millisecond level according to the rolling, spreading or thickness collapse of the tea leaves due to water loss during transportation. The heat map will therefore become more three-dimensional and rich in information.

[0079] (S4) Collaborative optimization of zoned heating based on prevention objectives: When calculating the heating parameters for each micro-zone, the zoned heating control unit will make full use of the precise thickness information. For thicker micro-zones, not only may the initial heating power be higher, but the heating strategy may also include a longer duration or a specific pulse heating mode to ensure that the heat can penetrate. At the same time, the collaborative control and stir-fry linkage unit will actively prevent over-killing or under-killing when optimizing the heating strategy of adjacent micro-zones. For example, if a thick-layer micro-zone requires strong heating, the system will use a model to predict its impact on the thermal radiation and hot air drift of the downstream thin-layer micro-zone, and reduce the heating power of the downstream micro-zone in advance, or start airflow isolation between them (if supported by the heating module) to ensure that the downstream tea will not be affected by the strong fire of the "neighbor".

[0080] (S5) Smarter physical transport and stir-frying linkage: Thanks to the precise thickness data, the collaborative control and stir-fry linkage unit can judge the physical state of the tea more accurately. When tea accumulation is found, not only will the stir-fry mechanism be linked, but the depth of the stir-fry arm, the turning frequency or the instantaneous speed change of the conveyor belt may be adjusted more finely according to the severity (thickness value) and range of the accumulation.

[0081] (S6) Closed-loop iterative optimization and traceable completion: The entire process is still a closed-loop iteration from S2 to S5, but due to the rich information of each link and the optimization of the algorithm, the system's control is more precise and efficient. The operator can monitor the entire process in real time through the HMI. When necessary, certain parameters (such as the speed of the overall withering rhythm) can be fine-tuned according to preset permissions. After the withering is completed, all sensor data, control parameters, micro-area evolution maps, alarm records, etc. related to the batch of tea are automatically archived in the database, and detailed process reports can be generated, providing a solid basis for quality management and traceability.

[0082] Specific application cases: Suppose a batch of fresh leaves is mixed with a lot of coarse old leaves or leaf stems due to picking or initial sorting, or young leaves are mixed with mature leaves. This unevenness is a huge challenge to the traditional withering process, which can easily lead to over-burning of young leaves and incomplete killing of old leaves.

[0083] The system of this embodiment: 3D vision sensors can accurately identify areas of localized thickening or morphological abnormalities caused by leaf stalks or thick, old leaves. Multi-factor threshold segmentation and clustering algorithms quickly and precisely separate these areas from the surrounding normal young leaf areas, forming independent micro-regions. For leaf stalks or thick, old leaf micro-regions, since their tissue structure is denser and their moisture content may be different from that of young leaves, the system will match specific heating strategies (which may require longer osmotic heating or microwave-assisted heating in a specific band (if supported by the heating unit)). For delicate leaf micro-regions, a relatively gentle but rapid heating method is adopted. The collaborative control unit will pay special attention to thermal crosstalk between adjacent micro-regions with very different characteristics, strictly implementing the optimization goal of preventing over-killing (young leaves) or under-killing (old stalks). Operators can observe the formation, heating status, and system response strategies of these special micro-regions in real time through the HMI, and adjust the partition sensitivity as needed to ensure the best processing effect.

[0084] Through precise 3D stack thickness sensing and a refined dynamic zoning algorithm based on multiple factors, the system can better identify and independently process tea leaves of varying thickness, age, and dryness, significantly improving the uniformity and overall quality of tea leaves during withering. This is particularly advantageous when processing low-grade or mixed raw materials. The collaborative control unit is optimized to prevent over- and under-withering, making the system more proactive and protective in its control. This effectively reduces common withering defects such as burnt edges, cracked spots, red stems, and green leaves, resulting in a higher yield. The introduction of an HMI allows operators to easily set and adjust process parameters based on different tea varieties and quality requirements, comprehensively monitor the withering process in real time, and receive timely alarm information. Automatic archiving of key data provides powerful data support for production management, quality traceability, and continuous process improvement, enhancing the standardization and informatization of tea processing. More precise micro-zoning and targeted heating strategies, particularly understanding the actual three-dimensional distribution of tea leaves, ensure that heat is more effectively applied to the target tea leaves, reducing ineffective heating and energy waste.

[0085] In summary, Example 2, through enhancements in perception, partitioning algorithms, collaborative control objectives, and system management, enables the machine vision-based segmented heating and withering system for tea to demonstrate more outstanding performance and practical value in processing complex raw materials, improving product uniformity and quality, and improving production management levels.

[0086] Example 3: like Figures 1 to 3 As shown, the tea withering system described in this embodiment inherits the sophisticated hardware configuration (including the tea conveying and stir-frying module, the multimodal visual perception module containing a 3D visual sensor, and the array-type segmented heating execution module) and refined control logic (such as dynamic partitioning based on a multi-factor algorithm, collaborative control aimed at preventing defects, and complete human-computer interaction and data management) described in Examples 1 and 2. At the same time, the predictive withering model and adaptive learning unit within its core central control and processing module have undergone a qualitative leap. This unit no longer uses only conventional machine learning models, but deploys a more complex and powerful predictive withering model based on a deep neural network (DNN).

[0087] In this embodiment, the core of the predictive tea-withering model and the adaptive learning unit is a deep learning architecture. For example, a hybrid neural network model can be used: a convolutional neural network (CNN) is used to efficiently extract features from the image data of each micro-area obtained from the multimodal visual perception module (RGB images, thermal images, and thickness maps and moisture distribution maps generated by 3D sensors and NIR data), and automatically learn key visual representations of tea leaves at different withering stages, such as spatial texture, color distribution, and thermal field morphology; these deep visual features are then combined with other numerical sensor data of the micro-area (such as average moisture content, average temperature, historical heating power, etc.) and time The hybrid DNN model combines temporal sequence information (such as the state of the micro-zone on the conveyor belt at the previous moment and the treatment it received) with temporal sequence information and inputs them into variants of recurrent neural networks (RNNs), such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs). LSTMs or GRUs can effectively capture the dynamic characteristics of the tea state evolution over time and the temporal dependencies during the heating process. Ultimately, the output layer of the hybrid DNN model is designed to directly predict the entire heating power variation curve (rather than a single power value) required for each dynamic micro-zone to reach the ideal withering state (preset by the operator through the HMI or learned by the system from historical optimal batches) and accurately estimate the total heating time.

[0088] The training of this DNN model is an ongoing process. The initial model is trained jointly by massive historical tea-withering data and real-time multimodal visual feature data collected synchronously with these data. The historical data not only includes all sensor readings and system control parameters (heating unit power, conveyor belt speed, stir-frying frequency, etc.) of tea of ​​different varieties, different tenderness, and different moisture content in the previous tea-withering process, but also crucially includes the corresponding finished product quality evaluation results of these batches of tea. These quality evaluation results may come from professional sensory evaluation (such as appearance, color, aroma, taste, and leaf bottom scores), or from laboratory quantitative analysis of key chemical components (such as tea polyphenols, amino acids, caffeine, chlorophyll content, and the concentration of key aroma substances such as linalool and phenylacetaldehyde).

[0089] During the actual operation of the system, the DNN model not only outputs prediction results, but also continuously performs online parameter optimization. This means that the model will make small-scale dynamic adjustments to certain hyperparameters of the current heating strategy (such as the overall scaling factor of the heating curve and time fine-tuning of key turning points) based on the real-time feedback data of the current batch of tea (such as the leaf temperature rise rate, water evaporation efficiency, and the deviation of the leaf color change speed from the prediction) to more accurately match the actual response characteristics of the current tea.

[0090] More importantly, its continuous learning mechanism is based on feedback from finished product quality. After each batch of tea is withered and undergoes subsequent rolling, drying and other processes to become dry tea, its quality will be evaluated. This finished product quality evaluation result, which contains a specific score or grade, will be fed back to the predictive withering model and adaptive learning unit together with the complete withering process data of the batch (all sensor records, control instruction sequences, micro-area evolution history, etc.) as a new high-quality training sample. The system will regularly (for example, after the end of daily production or weekly) use this newly added data with clear quality labels to perform incremental training or model fine-tuning on the core DNN model. In this way, the model can learn which complex processing parameter combinations have a stronger positive correlation with high-quality finished tea, and which subtle process deviations may lead to quality degradation, thereby continuously improving its prediction accuracy and the effectiveness of its control strategy, enabling the system to autonomously and continuously evolve towards the goal of producing higher-quality tea.

[0091] Working process The workflow of this embodiment is based on the refined operations of Example 2 and deeply integrates the prediction and learning capabilities of the DNN model: The initial steps of tea leaf import, multimodal real-time perception (including 3D thickness information), multidimensional feature extraction, and dynamic partitioning and heat map construction based on a multi-factor algorithm are similar to those in Example 2, providing high-quality, high-dimensional input for the DNN model.

[0092] The key difference is reflected in the (S4) zoned differentiated heating and coordinated optimization control stage. At this time, for each dynamically generated micro-zone, the predictive withering model and the deep neural network model in the adaptive learning unit will comprehensively analyze all its current feature vectors (color, temperature, water, shape, thickness, and historical state sequence). Based on its powerful nonlinear fitting and pattern recognition capabilities, the DNN model will directly output an optimal, time-varying heating power curve for the micro-zone and the total time expected to reach the ideal withering state. This curve may not be a simple constant power or linear change, but is based on the physiological and chemical changes of the tea at this stage (such as rapid heating and dehydration in the early stage, stabilization in the middle stage, etc.). After receiving this "guidance curve", the zoned heating control unit will drive the corresponding heating unit through high-frequency closed-loop control to reproduce the dynamic process of heating power or energy output as accurately as possible. At the same time, the estimated duration output by the DNN model also provides an important reference for the overall production scheduling and the connection between downstream processes. During the heating execution process, if the actual state (such as leaf temperature) deviates from the "ideal trajectory" implied by the DNN model, the online parameter optimization mechanism will fine-tune the execution of the current heating curve, for example, slightly advancing or delaying a certain stage of the curve, or slightly increasing or decreasing the power peak.

[0093] Similarly, (S5) the intelligent linkage adjustment between physical transportation and stir-frying will also be indirectly affected by the output of the DNN model. For example, if the DNN model predicts that the tea leaves in a certain micro-area need to maintain a higher leaf temperature for a longer period of time due to their special state, the collaborative control unit may accordingly instruct the stir-frying mechanism to adopt a gentler stir-frying mode that focuses more on uniform heating rather than rapid water loss, and adjust the transportation speed to ensure that the micro-area has sufficient processing time.

[0094] In the entire cycle of (S6) closed-loop iterative optimization until the tea withering is completed, in addition to the real-time adjustment described in Example 2, the system of this embodiment does not simply end after each batch of production tasks is completed. The operator or the automated quality inspection system will input the quality data of the finished product batch into the system (through the HMI or data interface). The predictive tea withering model and the adaptive learning unit will automatically add these new "process-result" data pairs to the training database and start retraining or parameter fine-tuning of the DNN model at an appropriate time (such as when the system is idle). This means that if the system produces a batch of high-scoring tea according to the current model, the relevant weights of the model will be strengthened; conversely, if a batch of tea is of poor quality, the model will analyze the differences between the processed data of the batch and the high-scoring batches and adjust its internal parameters to avoid repeating decisions that may lead to quality degradation in similar situations in the future.

[0095] Specific application cases: Suppose a tea company wants to produce a limited edition high-end green tea with a specific floral aroma or unique sweet aftertaste. Its quality standards are far beyond the norm, and the company is extremely sensitive to subtle differences in the raw materials (such as fresh leaves from different mountains and picked at different times).

[0096] The system of this embodiment: In the initial stage, the system may be based on an existing general green tea withering model or a small amount of trial production data of this high-end tea. The operator will record the source information and characteristics of each small batch of raw materials in detail through the HMI (as described in Example 2), and conduct a very detailed sensory evaluation and physical and chemical analysis on each batch of finished products. These finished product quality evaluation results containing rich details (for example, the content of a certain specific component in the aroma, the freshness score of the taste, the brightness level of the soup color, etc.) are input into the system.

[0097] After several production cycles, the DNN model, by learning from these "small-batch, high-precision feedback" data, began to gradually grasp the complex mapping relationship between the fresh leaf characteristics of this high-end tea and the specific quality targets. It may find that to stimulate that unique floral aroma, it is necessary to use a special heating curve that first quickly heats up, then slightly cools down, and then slowly heats up at a specific stage of withering (defined by a specific leaf temperature and moisture content window), while combining it with gentle stir-frying at a specific frequency. This sophisticated combination of process parameters may be difficult to accurately summarize and stably reproduce with traditional experience.

[0098] When companies try to introduce a new tea tree variety with unknown characteristics for trial production, the system can also demonstrate strong adaptability. The initial batches of trial production may have average results, but as long as they insist on conducting detailed quality evaluations of the finished products and feeding back to the system, the DNN model can quickly learn the processing characteristics of the new variety, and gradually explore the withering process parameters suitable for it to achieve its optimal quality potential, greatly shortening the time for new product research and development and process finalization. For example, after several rounds of learning, the system may find that the new variety is more sensitive to high temperatures, and it is necessary to lower the withering peak temperature overall, but extend the medium-temperature treatment time to obtain the best balance of aroma and taste.

[0099] Deep neural networks can capture and learn deeper and more complex nonlinear relationships than traditional machine learning models, thereby making extremely accurate predictions of the heating power curve and duration required for each micro-area of ​​tea to reach the ideal withering state. This makes the heating process no longer a simple "insulation" or "heating", but a true "empowerment on demand" and "empowerment according to rhythm".

[0100] By introducing the actual finished product quality evaluation results as key feedback into model training, the system can go beyond the level of merely pursuing the compliance of intermediate process parameters and truly optimize and iterate itself with the goal of improving the quality of the final product. The system will become more and more "understand" how to make good tea.

[0101] DNN's powerful fitting ability and continuous learning mechanism enable the system to quickly learn new patterns from limited samples, adapt to the characteristics of fresh leaves from different origins, different seasons, different picking standards, and even completely new tea varieties, and quickly find or approximate their optimal withering process parameters, accelerating product innovation and responsiveness to market changes. Through deep learning of large amounts of high-quality batch data, the system may discover some subtle process parameter combinations that traditional experience has not clearly defined or are difficult to pass on stably, and solidify this "implicit knowledge" into the model, realizing the digital inheritance and large-scale replication of top withering techniques, and overall improving the quality ceiling of the tea industry. Although the DNN model itself is sometimes called a "black box", by analyzing the decisions made by the model when processing different inputs (such as the selection of a specific heating curve), combined with a large amount of process data and quality results, it can in turn help researchers gain a deeper understanding of the complex chemical changes of tea during the withering process and their relationship with the final sensory quality, promoting the development of tea science.

[0102] In summary, Example 3 introduces an advanced deep neural network prediction model and a continuous learning feedback mechanism based on the quality of the finished product, so that the system can not only complete the current production tasks accurately and efficiently, but also continuously learn and evolve to approach or even surpass the level of top manual skills, providing a powerful technical engine for the quality improvement, personalized customization and continuous innovation of high-end tea.

[0103] The above description is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be regarded as excluding other embodiments. Instead, it can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or technology or knowledge in related fields. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention and should be within the scope of protection of the claims attached to the present invention.

Claims

1. The tea segmented heating and fixing system based on machine vision is characterized by: include: The tea conveying and stir-frying module is used to carry, convey and stir the tea leaves to be withered. The tea conveying and stir-frying module includes an adjustable speed conveyor belt and an adjustable speed and variable angle stir-frying mechanism; A multimodal visual perception module is configured above or to the side of the tea conveying and stir-frying module. The multimodal visual perception module includes at least one high-resolution RGB industrial camera, at least one near-infrared spectrometer or hyperspectral camera, at least one infrared thermal imager, and a light source system that provides illumination for the imaging unit. It is used for real-time, non-contact collection of RGB image data, NIR spectrum data, and thermal imaging data of tea leaves; An array-type segmented heating execution module, comprising a plurality of heating units capable of independently controlling power, energy output, or temperature and air volume, the heating units being arranged in an array and covering the processing area on the tea conveying and stir-frying module; A central control and processing module is connected to the multimodal visual perception module and the array-type segmented heating execution module, and includes: A multimodal data fusion and processing unit is used to receive the data collected by the multimodal visual perception module and perform image space registration, feature fusion and abnormal noise filtering to obtain leaf color characteristics, moisture content distribution characteristics, surface temperature distribution characteristics and stacking thickness information of the tea leaves; A dynamic partitioning and heat map construction unit is used to dynamically divide the processing area into a plurality of independently controllable micro-areas based on the feature data obtained by the multimodal data fusion and processing unit, and to generate and update a dynamic heat map reflecting the fixing status of each micro-area in real time; a zoned heating control unit for independently calculating heating parameters for each micro-zone according to the withering status presented on the dynamic heat map and preset withering target parameters, and outputting the parameters to the corresponding heating unit in the arrayed segmented heating execution module, so as to achieve differentiated and adaptive heating of the tea leaves in each micro-zone; A predictive fixing model and adaptive learning unit, which uses machine learning to combine historical fixing data with real-time data to predict the heat and heating time required to achieve the target fixing state for each micro-zone, optimize the heating strategy parameters online, and provide real-time warnings for abnormal conditions. The collaborative control and stir-fry linkage unit is used to comprehensively consider the heat conduction effect between adjacent micro-areas when outputting the heating parameters of the partitioned heating control unit, collaboratively optimize the heating instructions of the target micro-area, and automatically adjust the conveyor belt speed and the working parameters of the stir-frying mechanism in the tea conveying and stir-frying module according to the uniformity of tea distribution or insufficient stirring detected by the multimodal visual perception module to ensure that the tea is evenly heated.

2. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The multimodal visual perception module further includes a 3D vision sensor or a structured light sensor for obtaining the stacking thickness distribution information of the tea leaves on the tea conveying and stir-frying module. The stacking thickness distribution information participates in the dynamic micro-area division and heating decision of the central control and processing module.

3. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The heating units in the array-type segmented heating execution module are one or more of an infrared heating lamp with independently adjustable power, a microwave feed port with independently controlled energy output, or a hot air nozzle with independently controlled temperature and air volume, and the heating range of each heating unit corresponds to at least one micro-zone dynamically divided by the central control and processing module.

4. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The multimodal data fusion and processing unit in the central control and processing module is further used to extract the average chromaticity value and chromaticity uniformity index of the tea leaves from the RGB image data, extract the average moisture content and moisture content standard deviation of the tea surface from the NIR spectrum data, and extract the average temperature and maximum temperature point of each micro-area from the thermal imaging data. The indicators are used by the dynamic partitioning and thermal map construction unit to construct the withering thermal map and make partition heating decisions.

5. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The dynamic partitioning and heat map construction unit in the central control and processing module adopts a multi-factor threshold segmentation and clustering algorithm when performing dynamic partitioning to achieve dynamic division of micro-areas based on parameters such as tea distribution, temperature gradient, moisture content distribution, and stacking thickness. The physical boundaries of the micro-areas are adjusted in real time as the tea leaves move or their state changes.

6. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The predictive killing model in the central control and processing module and the predictive killing model based on machine learning adopted by the adaptive learning unit are jointly trained through historical killing data and real-time collected data, and online parameter optimization is performed during the system operation. It can predict the heating power curve and heating time required for each micro-area based on the current multimodal visual features.

7. The machine vision-based segmented heating and fixing system for tea leaves according to claim 6, characterized in that: The predictive tea withering model is a model based on a deep neural network, and the model parameters are continuously optimized through the process data of each batch of tea withering and the corresponding finished product quality evaluation results.

8. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The collaborative control and stir-fry linkage unit in the central control and processing module prevents over-killing or under-killing of the target micro-area or its adjacent micro-areas when collaboratively optimizing the heating instructions of the target micro-area.

9. The machine vision-based segmented heating and fixing system for tea leaves according to claim 1, characterized in that: The system is further provided with a parameter setting interface and a human-computer interaction terminal, allowing the operator to input parameters such as tea variety, target withering degree, initial micro-zone number reference value and partition adjustment sensitivity, and can display dynamic heat maps, micro-zone status parameters, historical data trends and real-time alarm information in real time. All key process data are automatically archived for tea quality traceability.

10. A segmented heating and fixing method for tea leaves based on machine vision, characterized in that: The following steps are involved: S1. Use a multimodal visual perception module to collect RGB image data, NIR spectrum data, thermal imaging data, and stacking thickness distribution information of tea leaves in real time; S2, the multimodal data fusion and processing unit of the central control and processing module receives and processes the data collected in step S1, performs image spatial registration, feature fusion, and abnormal noise filtering to obtain the leaf color characteristics, moisture content distribution characteristics, surface temperature distribution characteristics, and stacking thickness information of the tea leaves; S3. The dynamic partitioning and heat map construction unit of the central control and processing module dynamically divides the processing area into multiple micro-zones based on the feature data obtained in step S2, constructs a real-time heat map of the fixing status, and continuously adjusts the micro-zone division according to the analysis results and preset parameters; S4. For each micro-zone, the zoned heating control unit of the central control and processing module independently calculates the heating parameters based on the state of the dynamic thermal map and the preset withering target, and instructs the corresponding heating unit in the array-type segmented heating execution module to implement differentiated heating of the tea leaves in the micro-zone; at the same time, the collaborative control and stir-frying linkage unit collaboratively optimizes the heating strategies of adjacent micro-zones; S5. The coordinated control and stir-frying linkage unit of the central control and processing module continuously monitors and automatically adjusts the stir-frying and conveying parameters of the tea conveying and stir-frying module according to the uniformity of tea distribution or insufficient stirring detected by the multimodal visual perception module to ensure uniform distribution and heating of the tea leaves; S6. Repeat steps S2 to S5 until the central control and processing module determines that all micro-area tea leaves have reached the set withering standard, thus completing the withering process.

Images