Jasmine tea scenting process intelligent discrimination method based on machine vision and volatile gas monitoring

Abstract

The invention provides a jasmine tea scenting process intelligent discrimination method based on machine vision and volatile gas monitoring. The jasmine tea scenting process intelligent discrimination method comprises the following steps: periodically collecting image data and volatile gas concentration data in a jasmine tea scenting process; inputting the image data into a trained jasmine flower shape recognition model, and outputting the confidence coefficient of the shape of the jasmine flower when the jasmine flower reaches a tiger claw shape or flowers; according to a flower growing scene and a flower scenting scene, inputting the gas concentration data into a trained gas trend analysis model, and outputting a judgment result of whether gas characteristics accord with a tiger claw-shaped stage or a flower raising stage; when the confidence coefficient aiming at the tiger claw shape reaches a preset threshold value and the gas characteristic judgment result aiming at the tiger claw stage is consistent, generating a prompt signal or a control instruction for preparing scenting; and generating a prompt signal or a control instruction for preparation of flowering when the confidence coefficient of the flowering form reaches a preset threshold value and the judgment result of the gas characteristics of the flowering stage is consistent.

Description

Technical Field

[0001] This invention belongs to the fields of tea processing and artificial intelligence technology, specifically relating to an intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring. Background Technology

[0002] The core quality of jasmine tea is determined by the scenting process. Among them, the judgment of the "tiger claw shape" (the jasmine buds are nurtured until the petals unfold into a tiger claw shape, at which time the aroma is released at its best and it is the key moment to start the scenting process) and the timing of flower opening (the point at which the jasmine flowers need to be separated from the tea before they wither) are the core control points of the scenting process.

[0003] In the traditional scenting process, judging the two key moments mentioned above relies entirely on the tea master's subjective experience and sensory perception: the "tiger claw" shape must be determined by visually observing the jasmine flower's form and smelling its aroma; the starting point for blooming is judged by observing the degree of petal withering and perceiving changes in aroma. This method has significant drawbacks: First, consistency in judgment is poor; different tea masters have varying experience and sensory sensitivity, and even the same tea master's judgment standards may shift when fatigued or when there are fluctuations in ambient temperature and humidity. Second, there is a high risk to quality; if scenting begins too early, the volatile gases released by the jasmine buds contain a lot of grassy odor, resulting in a grassy, ​​unpleasant smell in the tea; if it begins too late, the optimal aroma is lost, reducing the aroma concentration of the tea; if the blooming begins too late, the irritating volatiles released by the decaying jasmine flowers will give the tea a musty, unpleasant smell, severely damaging its quality.

[0004] With the large-scale development of the jasmine tea industry, traditional manual judgment methods can no longer meet the requirements of standardized production for stability and objectivity. There is an urgent need for an intelligent judgment technology that is accurate, stable, and does not rely excessively on human experience to solve the drawbacks of manual judgment and ensure the consistency of jasmine tea quality. Summary of the Invention

[0005] To address the shortcomings and deficiencies of existing technologies that rely on subjective human judgment for key timing in jasmine tea scenting (tiger claw scenting timing, flower initiation timing), resulting in poor consistency, high quality risks, and dependence on highly experienced personnel, this invention provides an intelligent method and system for determining key timing in jasmine tea scenting.

[0006] This invention deploys data acquisition devices in jasmine flower cultivation areas and tea-flower mixing areas to periodically collect image data of jasmine flowers in their open state, as well as gas data including the total concentration of volatile organic compounds, the concentrations of four characteristic volatile gases (benzyl alcohol, linalool, benzyl acetate, and indole), oxygen concentration, and carbon dioxide concentration. The image data acquisition cycle is 3 minutes, and the gas data acquisition cycle is 1 minute. A pre-trained convolutional neural network jasmine flower morphology recognition model, optimized through preprocessing, is constructed to output the confidence level of jasmine flowers reaching a tiger claw shape or the morphology of newly opened flowers. Simultaneously, gas trend analysis models are constructed for both flower cultivation and scenting scenarios, based on the daily baseline of the flower cultivation scenario from 17:00 to 10:00 the next day and 2... The daily baseline for the scenting process from 1:00 AM to 10:00 AM the next day is used to extract key stage gas features after aligning and smoothing the gas data. A dual-condition fusion discrimination logic is adopted. When the confidence level of the tiger claw shape or the flower-forming shape reaches a preset threshold and the gas feature judgment result of the corresponding scene is consistent, the corresponding prompt signal or control command for preparing to scent or prepare to form flowers is generated. The training data corresponding to the key moments of operation are manually confirmed and the model is iteratively optimized. A special discrimination model adapted to a fixed type of jasmine flower and tea base is built for each device. At the same time, the module containing multimodal prompt unit and equipment control unit realizes prompt output and dual-mode control. An emergency pause button is configured to ensure production safety.

[0007] This invention, through innovative design involving multi-source data fusion, scenario-based modeling, and human-machine collaborative iterative optimization, eliminates reliance on subjective human intervention, significantly improves the accuracy and stability of key timing judgment, effectively avoids quality problems caused by early scenting, late scenting, and late flower picking, reduces dependence on highly experienced tea makers, and adapts to the needs of large-scale and standardized jasmine tea production.

[0008] The specific technical solution adopted by this invention to solve its technical problem is as follows: A smart identification method for jasmine tea scenting process based on machine vision and volatile gas monitoring includes: Image data and volatile gas concentration data are periodically collected during the jasmine scenting process. The volatile gas concentration data includes the total concentration of volatile organic compounds, the concentration of characteristic volatile gases, the oxygen concentration, and the carbon dioxide concentration. The image data is input into a trained jasmine flower morphology recognition model, which outputs the confidence level of the jasmine flower reaching the tiger claw shape or the flowering stage; according to the flower cultivation scenario and the flower scenting scenario, the gas concentration data is input into a trained gas trend analysis model, which outputs the judgment result of whether the gas characteristics meet the tiger claw stage or the flowering stage. When the confidence level for the tiger claw-shaped pattern reaches a preset threshold, and the judgment result for the gas characteristics of the tiger claw-shaped stage is consistent, a prompt signal or control command for preparing to insulate is generated. When the confidence level of the flowering morphology reaches a preset threshold, and the judgment result of the gas characteristics at the flowering stage is consistent, a prompt signal or control command to prepare for flowering is generated.

[0009] Furthermore, the characteristic volatile gases include benzyl alcohol, linalool, benzyl acetate, and indole.

[0010] Furthermore, the periodic data collection is achieved by deploying collection equipment in the jasmine flower cultivation area and the tea-flower mixing area: a high-definition camera is placed at an unobstructed location 15-30cm away from the jasmine or tea flower pile, and the air inlet of the gas collection device is placed 2-3cm high on the side of the pile; the high-definition camera collects image data every 3 minutes, and the gas collection device collects volatile gas concentration data and oxygen and carbon dioxide concentration data every 1 minute.

[0011] Furthermore, the jasmine flower morphology recognition model is a pre-trained convolutional neural network model. During training, the morphological features of tiger claw shape and flower blooming are used as target labels. Before training, the image data is preprocessed by normalization, enhancement, and size unification.

[0012] Furthermore, the gas trend analysis model is a model trained on time series gas data after alignment and smoothing of the daily baselines for the flower growing scenario and the scenting scenario. The daily baseline for the flower growing scenario is from 17:00 to 10:00 the next day, and the daily baseline for the scenting scenario is from 21:00 to 10:00 the next day.

[0013] Furthermore, it also includes a model iteration and optimization step: by manually confirming the operation record of the tiger claw shape and the timing of flowering, the image data and gas data corresponding to them are added to the training dataset, and the jasmine flower morphology recognition model and the gas trend analysis model are trained iteratively; the jasmine flower morphology recognition model and the gas trend analysis model are exclusive discrimination models adapted to fixed types of jasmine flowers and tea leaves.

[0014] Furthermore, the prompt signal or control command is implemented through a judgment result output and control module: the module includes a multimodal prompt unit and a device control unit; the multimodal prompt unit is used to generate a multimodal prompt signal, and the device control unit supports automatic control and manual confirmation dual-mode switching, and is equipped with an emergency pause button.

[0015] And, an intelligent discrimination system for jasmine tea scenting process based on machine vision and volatile gas monitoring, comprising: The data acquisition module is used to periodically acquire image data and volatile gas concentration data during the jasmine scenting process. The volatile gas concentration data includes the total concentration of volatile organic compounds, the concentration of characteristic volatile gases, the oxygen concentration, and the carbon dioxide concentration. The data acquisition module includes high-definition cameras and gas acquisition equipment deployed in the flower cultivation area and the tea-flower mixing area. The data processing module is used to perform preliminary preprocessing on the image data and volatile gas concentration data acquired by the data acquisition module. The preliminary preprocessing includes image format unification and gas data filtering. The model processing module includes a trained jasmine flower morphology recognition model and a gas trend analysis model. The jasmine flower morphology recognition model is used to process the preprocessed image data and output the confidence level of the jasmine flower reaching the tiger claw shape or the flowering stage. The gas trend analysis model is used to process the preprocessed volatile gas concentration data according to the flower cultivation scenario and the flower scenting scenario, and output the judgment result of whether the gas characteristics meet the tiger claw stage or the flowering stage. The fusion discrimination module is used to generate a prompt signal or control command to prepare for scenting when the confidence level of the tiger claw-shaped morphology reaches a preset threshold and the corresponding gas feature judgment result is consistent; and to generate a prompt signal or control command to prepare for flowering when the confidence level of the flowering morphology reaches a preset threshold and the corresponding gas feature judgment result is consistent. The iterative optimization module is used to manually confirm operation records and supplement training data corresponding to tiger claw shape and flowering time, iteratively train jasmine flower morphology recognition model and gas trend analysis model, and supports building exclusive discrimination models for each data acquisition device that are adapted to fixed types of jasmine flowers and tea leaves.

[0016] And an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0017] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0018] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects: It effectively eliminates the reliance on subjective human experience and sensory perception for judging key timing in traditional scenting processes. Through innovative design that integrates multi-source data of images and volatile gases and constructs scenario-specific models, it significantly improves the accuracy and stability of judging the timing of tiger claw-shaped scenting and the timing of flower initiation, avoiding judgment biases caused by personnel differences, environmental fluctuations, and fatigue.

[0019] By accurately capturing the compatibility between the release of jasmine aroma and changes in its form, the quality problems such as residual grassy smell caused by early scenting, loss of aroma caused by late scenting, and decaying odor pollution caused by late-blooming flowers are effectively avoided, providing a reliable guarantee for the consistency of jasmine tea aroma quality.

[0020] It lowers the barrier to entry for highly experienced tea makers, and reduces raw material waste and production risks caused by human error through automated data collection, intelligent judgment and multimodal prompt output. At the same time, it supports switching between automatic control and manual confirmation modes, taking into account both production efficiency and operational flexibility.

[0021] Through a human-machine collaborative model iteration and optimization mechanism and the construction of equipment-specific models, it can be adapted to different varieties of jasmine flowers and tea leaves. As it is used, it continuously accumulates samples and improves the discrimination accuracy, fully meeting the long-term adaptation needs of large-scale and standardized production of jasmine tea. Attached Figure Description

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a flowchart illustrating the intelligent method for determining key timing in the jasmine tea scenting process according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the training accuracy and loss curves of the jasmine flower morphology recognition model in an embodiment of the present invention. Detailed Implementation

[0023] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in detail: It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] This invention discloses an intelligent judgment method for the jasmine tea scenting process based on machine vision and volatile gas monitoring. The invention aims to overcome the shortcomings of traditional jasmine tea scenting, such as poor consistency, high quality risk, and reliance on highly experienced personnel, which are caused by manual judgment of the "tiger claw" shape and the timing of flower initiation. By integrating multi-source data and AI algorithms, the invention achieves accurate judgment of key timing, while supporting continuous iterative optimization of the model to adapt to the needs of large-scale production.

[0026] This method acquires multi-source data by deploying 1080p high-definition cameras (capturing images every 3 minutes) and gas collection devices (collecting data every 1 minute, monitoring TVOC, four characteristic gases, and oxygen and carbon dioxide concentrations) in the flower cultivation and tea-flower mixing areas. An initial dataset covering the bud stage, the "tiger claw" shape, the flowering stage, and the decay stage is constructed. A pre-trained CNN model is used to build an image recognition model (outputting "tiger claw" confidence scores). A gas concentration trend analysis model is built for the flower cultivation and scenting scenarios. Through multi-source data fusion, "ready for scenting" or "ready for flowering" prompts and equipment control commands are generated. Simultaneously, supplementary samples from tea makers' operations are added to continuously iterate and optimize the model. Each device is adapted to a fixed type of jasmine flower and tea base. This invention improves discrimination accuracy, ensures the quality of jasmine tea, reduces reliance on manual labor, and is suitable for large-scale jasmine tea production.

[0027] like Figure 1 As shown, the implementation of the present invention refers to the following steps: 1) Hardware Deployment and Data Acquisition: In the jasmine flower cultivation area and the tea-flower mixing area, a 1080p high-definition camera is installed 15-30cm away from the jasmine flower pile (or tea-flower pile) in an unobstructed location. The air intake head of the gas collection device is placed 2-3cm high on the side of the pile. The 1080p high-definition camera, equipped with a timed photo-taking program, collects image data of the jasmine flowers in open state every 3 minutes. The gas collection device collects data every 1 minute, including the total concentration of volatile organic compounds (TVOC), the concentrations of four characteristic volatile gases (benzyl alcohol, linalool, benzyl acetate, and indole), and the concentrations of oxygen and carbon dioxide.

[0028] 2) Dataset Construction: Three types of core data were collected for key periods / times of jasmine scenting (bud stage, "tiger claw" stage, flowering stage, and decay stage). These were image data obtained from the 1080p high-definition camera in step (1), volatile gas concentration data obtained from the gas collection device, and oxygen and carbon dioxide concentration data. Combined with the period / time information marked by the tea maker, an initial dataset was constructed. Gas data were collected separately for two scenarios: flower cultivation (from bud, tiger claw, full opening to decay) and scenting (from mixing the jasmine tea to flowering stage).

[0029] 3) Discrimination Model Construction and Fusion Discrimination: A pre-trained CNN model is used to construct a jasmine flower morphology recognition model. The model is trained after preprocessing the image data and outputs the confidence score of the "tiger claw" morphology. Gas concentration trend analysis models are constructed for the flower cultivation and scenting scenarios respectively to extract the gas concentration features of the "tiger claw" shape and the flower lifting process. When the image data shows that the jasmine flower has reached the "tiger claw" shape with a confidence score of 80% or higher, and the gas concentration trend in the corresponding scenario matches the "tiger claw" feature, a "preparing for scenting" prompt is generated or an instruction is issued to control the jasmine mixing equipment. When the multi-source data in the scenting scenario all match the flower lifting features, a "preparing for lifting" prompt is generated or an instruction is issued to control the lifting equipment.

[0030] 4) Model Iteration and Optimization: In the flower cultivation stage, when the tea master clicks the "Stop Flower Cultivation" button in the program to confirm that the "tiger claw shape" has been reached, the system automatically records the image data and gas concentration data at this time and marks them as "tiger claw shape" samples to supplement the dataset; when the tea master clicks "Start Flowering" to confirm that the flowering has been reached, the system similarly records and marks the corresponding data to supplement the dataset; based on the supplemented dataset, the image model and gas model are continuously trained iteratively, and each deployed device is adapted to a fixed type of jasmine flower and tea base to build a device-specific discrimination model.

[0031] As a preferred option, the gas acquisition equipment in step (1) includes a volatile gas sensor (using a photoionization detector (PID) to monitor TVOC and characteristic gases such as benzyl alcohol, linalool, benzyl acetate, and indole), an oxygen sensor, and a carbon dioxide sensor. The gas acquisition equipment can use a photoionization detector (PID) to monitor the overall concentration change of volatile organic compounds, and establish a correlation model between the PID comprehensive response signal and the concentrations of the four characteristic gases (benzyl alcohol, linalool, benzyl acetate, and indole) through prior offline calibration using gas chromatography-mass spectrometry (GC-MS); or it can be combined with a dedicated sensor array for the above four characteristic gases to achieve accurate monitoring of the concentration of characteristic gases and avoid cross-interference from a single sensor.

[0032] As a preferred option, the construction of the initial dataset in step (2) includes: conducting a complete scenting cycle under multiple combinations of different jasmine bud maturity, different ambient temperature and humidity, and different tea base types, and collecting corresponding image data and gas concentration data; during each scenting process, the tea maker marks the time nodes of "tiger claw shape" and flower blooming, and extracts image and gas data within a certain time range before and after the node, while collecting data samples of the bud stage and decay stage to form an initial dataset containing thousands of samples, and dividing it into training set, test set and validation set.

[0033] As a preferred option, the image model in step (3) is trained using an iterative training method, with the learning rate set in the range of 0.0001–0.01 and the number of iterations ranging from tens to hundreds. The target labels are "tiger claw" and the morphological features when the flower is in bloom. After training, the recognition accuracy on the test set is not less than 90%.

[0034] As a preferred option, in step (3), the gas model uses the baseline of flower cultivation from 17:00 to 10:00 the next day and the baseline of flower scenting from 21:00 to 10:00 the next day as references to perform time alignment and smoothing on the data of TVOC, characteristic volatile gases, oxygen, and carbon dioxide; a time series model is used to fit the trend of gas concentration changes, and a feature vector composed of standardized time and gas concentration is constructed. Stage identification is performed based on a multi-classification discrimination method to extract the gas characteristics of "tiger claw" and flower initiation.

[0035] As a preferred option, the model iteration optimization in steps (4) and (5) involves retraining after accumulating a certain number of new samples and comprehensively optimizing the exclusive model for each device within a fixed period to ensure that the accuracy of the identification of the appropriate jasmine flower and tea base types remains stable above the preset threshold.

[0036] The above solution will be further demonstrated and introduced through more specific implementation methods: (1) Hardware deployment and data acquisition The hardware implementation of this invention is deployed with reference to the following modules: Image acquisition module: In the jasmine flower cultivation area and the camellia mixing area, one high-definition camera is installed at an appropriate distance from the jasmine flower pile (or camellia flower pile) and in an unobstructed position to ensure that the details of the open shape of individual jasmine flowers can be clearly captured. The camera is equipped with a timed photo-taking program, which automatically collects image data once every 3 minutes, and the images are stored in a common format to the data terminal.

[0037] Gas Acquisition Module: The gas acquisition device's inlet pipe head is placed at an appropriate height on the side of the stack in both of the aforementioned areas. The gas acquisition device integrates three types of sensors, including a photoionization detector (PID) for monitoring TVOC and four characteristic gases: benzyl alcohol, linalool, benzyl acetate, and indole; an oxygen sensor; and a carbon dioxide sensor. The device collects gas concentration data once per minute, and the data is transmitted to the data terminal in real time.

[0038] Data processing module: Configured at the field end of each acquisition area, it acts as a small data processor to receive raw data from the image acquisition module and gas acquisition module, complete preliminary preprocessing operations such as image format unification and gas data filtering, and transmit the preprocessed data to the data terminal in real time to support subsequent model calculations and storage.

[0039] Judgment result output and control module: This module communicates bidirectionally with the data terminal and includes a multimodal prompting unit and a device control unit.

[0040] Multimodal prompting unit: Each of the two operating stations is equipped with a display screen and an audible and visual alarm. When the data terminal determines that "the tiger claw shape has been reached, prepare for scenting," the screen displays a red "Prepare for scenting" prompt, simultaneously displaying the image model confidence level and key gas concentration information. The audible and visual alarm provides a low-frequency beep and a flashing red indicator light. When it determines that "the flowering time has been reached, prepare for flowering," the screen displays a yellow "Prepare for flowering" prompt, displaying a comparison of petal expansion data and characteristic gas thresholds. The audible and visual alarm switches to a mid-frequency beep and a constantly lit yellow indicator light to avoid scene confusion.

[0041] Equipment control unit: Connects to the controller of the tea flower mixing and flower raising equipment, supporting dual-mode switching between "automatic control" and "manual confirmation". In automatic mode, after multi-source data meets the discrimination conditions and remains stable for a period of time, the terminal automatically sends a device start command, and the screen displays the operating status. In manual mode, the equipment will only start after confirmation from the tea master. Both modes are equipped with an emergency pause button, which can forcibly interrupt operation. The equipment operating status is fed back to the terminal in real time and stored in the production log for easy traceability and model optimization.

[0042] (2) Dataset Construction This study collects three core data points from key periods / time points throughout the entire jasmine scenting process (including at least the bud stage, the "tiger claw" stage, the initial blooming stage, and the decay stage). These data include image data acquired by high-definition cameras and volatile gas concentration data, as well as oxygen and carbon dioxide concentration data acquired by gas collection equipment. An initial dataset is constructed by combining this data with information from tea makers' annotations for each period / time point. Gas data is collected separately for two scenarios: flower cultivation (from bud to decay) and scenting (from tea mixing to initial blooming). The initial dataset must cover combinations of different jasmine bud maturity levels, environmental temperatures and humidity levels, and different tea base types. Complete scenting cycle data is collected, and image and gas data are extracted within a certain time range before and after the tea maker's annotation points. Samples from the bud and decay stages are also added to form an initial dataset containing thousands of samples, which is then divided into training, testing, and validation sets.

[0043] (3) Discriminant model construction and fusion discrimination, specifically including: Image recognition model: A jasmine flower morphology recognition model was constructed using a pre-trained CNN model. The training set image data was preprocessed and then iteratively trained. The training parameters were set as follows: learning rate 0.0001–0.01, iteration rounds tens to hundreds of rounds, with “tiger claw shape” and the morphological features at the time of flowering as the target labels. After training, the model output the confidence score of the “tiger claw shape” morphology judgment, and the recognition accuracy on the test set was no less than 90%.

[0044] Gas concentration trend analysis model: Gas models were constructed for the flower growing and flower scenting scenarios respectively. The daily baselines for flower growing (17:00 to 10:00 the next day) and flower scenting (21:00 to 10:00 the next day) were used as references. The data of TVOC, characteristic volatile gases, oxygen, and carbon dioxide were time-aligned and smoothed. The gas concentration change trend was fitted by a time series model, and a feature vector composed of standardized time and gas concentration was constructed. The gas concentration features of "tiger claw" pattern and flower start were extracted based on a multi-classification discrimination method.

[0045] Multi-source data fusion and discrimination: When the image data shows that the jasmine flowers have reached the "tiger claw" shape with a confidence level of 80% or above, and the gas concentration trend of the corresponding flower cultivation scene matches the "tiger claw" characteristic, the data terminal generates a "preparing for scenting" prompt or issues an instruction to control the camellia mixing equipment to work; when the image and gas multi-source data in the scenting scene both meet the flower-starting characteristics, a "preparing for flower starting" prompt is generated or an instruction is issued to control the flower-starting equipment to work.

[0046] (4) Model Iterative Optimization Manual interaction to supplement samples: The data terminal is equipped with "Stop growing flowers" and "Start flowering" operation buttons. When the tea master confirms that the tea has reached the "tiger claw" shape or the flowering stage has started, based on the actual production situation, he clicks the corresponding button. The system automatically records the image data and gas concentration data at the time of the click and marks them as samples of the corresponding stage, supplementing the dataset.

[0047] Model iterative training: After a certain number of new samples have been accumulated in the supplementary dataset, the new samples are divided and merged with the original training set and test set to retrain the image model and gas model, thereby improving the discrimination accuracy.

[0048] Equipment-specific model construction: Each deployed device is adapted to a fixed variety of jasmine flowers and tea leaves, and the device-specific discrimination model is trained only with samples corresponding to that variety; within a fixed period, the device-specific model is comprehensively optimized based on new samples accumulated by the device to ensure that the discrimination accuracy is stable above the preset threshold.

[0049] Based on the above design, a specific test instance is implemented as follows: 1. Equipment Deployment In the jasmine flower cultivation area, a 1080p high-definition camera is installed 15-30cm away from the jasmine flower pile in an unobstructed position; the air intake head of the gas collection device is placed 2-3cm high on the side of the pile. The gas collection device integrates a PID sensor, an oxygen sensor and a carbon dioxide sensor. The device is connected to a data terminal (which has data storage and model running functions and can be connected to a prompter and device controller) via a data cable.

[0050] In the tea-flower mixing area, a second high-definition camera and gas collection device were installed according to the same specifications and location requirements as described above to ensure that the opening status of the jasmine flowers and the gas concentration in the tea-flower pile could be monitored.

[0051] 2. Dataset Initialization and Model Training Starting from multiple combinations of different jasmine bud maturity, different environmental temperature and humidity, and different tea base types, data collection was conducted for each complete scenting cycle. Each scenting process was marked by the tea master with the "tiger claw" shape and the time node when the flowers started to bloom. Image and gas data were extracted within a certain time range before and after the node. Samples were also collected during the bud stage and the decay stage to form an initial dataset containing thousands of samples, which was then divided into training set, test set and validation set according to the proportion.

[0052] Training the pre-trained CNN image model: set the learning rate to 0.0001–0.01, iterate for dozens to hundreds of rounds, and after training, the accuracy of "tiger claw" recognition on the test set is no less than 90%; Training the gas concentration trend analysis model: use 17:00 to 10:00 the next day as the daily baseline for flower cultivation and 21:00 to 10:00 the next day as the daily baseline for flower scenting, fit the trend through a time series model, and extract the gas features of key stages.

[0053] 3. Practical Applications and Model Optimization During the scenting process of a certain batch of jasmine flowers, the system collects images of the flower-growing area every 3 minutes and gas data every 1 minute. When the image model outputs a "tiger claw" shape with a confidence level of 82%, and the gas data (characteristic gases such as benzyl alcohol and linalool, as well as the concentrations of oxygen and carbon dioxide) match the "tiger claw" shape, the system triggers a "prepare for scenting" prompt. After the tea master confirms, he clicks "stop flower growing," and the system records the data at that moment and adds it as a "tiger claw" shape sample.

[0054] In this embodiment of the invention, a pre-trained convolutional neural network (CNN) is used to train image samples of jasmine flowers at different stages. The training parameters are set to a learning rate of 0.0003 and approximately 735 iterations (16 rounds). The training is completed in a single CPU environment.

[0055] Training curves Figure 2 As shown.

[0056] Accuracy curve ( Figure 2 (Above): In the initial stage, the model accuracy rose rapidly from about 20% to over 90%, and then stabilized. After about 100 iterations, the training accuracy remained above 95%, eventually reaching 98.83%. This indicates that the model can effectively learn key morphological features such as "tiger claw shape" and "flowering time", and the recognition accuracy meets the requirements for determining the timing of the indentation.

[0057] Loss curve ( Figure 2 Below: Initially, the loss value is relatively high (about 1.6), but it decreases rapidly as the iteration progresses. After 100 iterations, it converges to below 0.2 and remains stable thereafter. The trend of the validation set loss is consistent with that of the training set loss, and no obvious overfitting is observed.

[0058] The results show that the image recognition model constructed in this invention has high stability and accuracy in recognizing the "tiger claw" shape. When combined with the gas monitoring model, it can achieve reliable intelligent identification of critical timing for sealing.

[0059] After a certain number of supplementary samples are accumulated, the image model and gas model are retrained, and the model's discrimination accuracy is improved. For the jasmine and tea varieties that the device is compatible with, the dedicated model is optimized with the accumulated new samples every fixed period to ensure that the accuracy of the "preparing for scenting" and "preparing for flowering" prompts remains stable above the preset threshold during long-term use.

[0060] Compared with the prior art, the beneficial effects of the present invention include: Improve the accuracy and stability of discrimination: By fusing multi-source data from machine vision and volatile gas monitoring, combined with AI algorithm discrimination, we can get rid of the reliance on human subjectivity and avoid judgment bias caused by human and time factors. The discrimination accuracy of "tiger claw shape" and flowering stage is stable at over 90%.

[0061] To ensure the quality of jasmine tea: accurately capture the best time for scenting and opening the flowers, avoid the problems of early scenting (residual grassy smell), late scenting (loss of aroma), and late opening of the flowers (contamination of decaying aroma), and effectively improve the aroma quality of jasmine tea.

[0062] Reduce reliance on manual labor and production risks: The system generates objective judgment prompts in real time, reducing reliance on highly experienced tea masters and minimizing raw material waste and quality risks caused by human error.

[0063] Highly adaptable and continuously optimized: It supports the construction of exclusive models based on jasmine flowers and tea base varieties to meet the scenting requirements of different production areas and varieties; through manual interaction to supplement samples, the model accuracy gradually improves with the use time, making it suitable for long-term large-scale production.

[0064] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0065] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0066] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0068] This invention is not limited to the preferred embodiment described above. Anyone inspired by this invention can derive other forms of intelligent identification methods for jasmine tea scenting processes based on machine vision and volatile gas monitoring. All equivalent variations and modifications made within the scope of this invention should be included in the scope of this invention.

Claims

1. A method for intelligently identifying the jasmine tea scenting process based on machine vision and volatile gas monitoring, characterized in that, include: Image data and volatile gas concentration data are periodically collected during the jasmine scenting process. The volatile gas concentration data includes the total concentration of volatile organic compounds, the concentration of characteristic volatile gases, the oxygen concentration, and the carbon dioxide concentration. The image data is input into a trained jasmine flower morphology recognition model, which outputs the confidence level of the jasmine flower reaching the tiger claw shape or the flowering stage; according to the flower cultivation scenario and the flower scenting scenario, the gas concentration data is input into a trained gas trend analysis model, which outputs the judgment result of whether the gas characteristics meet the tiger claw stage or the flowering stage. When the confidence level for the tiger claw-shaped pattern reaches a preset threshold, and the judgment result for the gas characteristics of the tiger claw-shaped stage is consistent, a prompt signal or control command for preparing to insulate is generated. When the confidence level of the flowering morphology reaches a preset threshold, and the judgment result of the gas characteristics at the flowering stage is consistent, a prompt signal or control command to prepare for flowering is generated.

2. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: The characteristic volatile gases include benzyl alcohol, linalool, benzyl acetate, and indole.

3. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: The periodic data collection is achieved by deploying collection equipment in the jasmine flower cultivation area and the tea-flower mixing area: a high-definition camera is placed 15-30cm away from the jasmine flower pile or tea flower pile in an unobstructed position, and the air inlet of the gas collection device is placed 2-3cm high on the side of the pile; the high-definition camera collects image data every 3 minutes, and the gas collection device collects volatile gas concentration data and oxygen and carbon dioxide concentration data every 1 minute.

4. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: The jasmine flower morphology recognition model is a pre-trained convolutional neural network model. During training, the morphological features of tiger claw shape and flower blooming are used as target labels. Before training, the image data is preprocessed by normalization, enhancement and size unification.

5. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: The gas trend analysis model is a model trained on time series gas data after alignment and smoothing of the daily baselines for the flower growing scenario and the scenting scenario. The daily baseline for the flower growing scenario is from 17:00 to 10:00 the next day, and the daily baseline for the scenting scenario is from 21:00 to 10:00 the next day.

6. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: It also includes model iteration and optimization steps: manually confirming and recording the image data and gas data corresponding to the tiger claw shape and the timing of flowering, and supplementing them into the training dataset, and iteratively training the jasmine flower morphology recognition model and the gas trend analysis model; the jasmine flower morphology recognition model and the gas trend analysis model are exclusive discrimination models adapted to fixed types of jasmine flowers and tea leaves.

7. The intelligent discrimination method for jasmine tea scenting process based on machine vision and volatile gas monitoring according to claim 1, characterized in that: The prompt signal or control command is implemented through the discrimination result output and control module: the module includes a multimodal prompt unit and a device control unit; the multimodal prompt unit is used to generate multimodal prompt signals, and the device control unit supports automatic control and manual confirmation dual-mode switching, and is equipped with an emergency pause button.

8. An intelligent discrimination system for jasmine tea scenting process based on machine vision and volatile gas monitoring, characterized in that, include: The data acquisition module is used to periodically acquire image data and volatile gas concentration data during the jasmine scenting process. The volatile gas concentration data includes the total concentration of volatile organic compounds, the concentration of characteristic volatile gases, the oxygen concentration, and the carbon dioxide concentration. The data acquisition module includes high-definition cameras and gas acquisition equipment deployed in the flower cultivation area and the tea-flower mixing area. The data processing module is used to perform preliminary preprocessing on the image data and volatile gas concentration data acquired by the data acquisition module. The preliminary preprocessing includes image format unification and gas data filtering. The model processing module includes a trained jasmine flower morphology recognition model and a gas trend analysis model. The jasmine flower morphology recognition model is used to process the preprocessed image data and output the confidence level of the jasmine flower reaching the tiger claw shape or the flowering stage. The gas trend analysis model is used to process the preprocessed volatile gas concentration data according to the flower cultivation scenario and the flower scenting scenario, and output the judgment result of whether the gas characteristics meet the tiger claw stage or the flowering stage. The fusion discrimination module is used to generate a prompt signal or control command to prepare for scenting when the confidence level of the tiger claw-shaped morphology reaches a preset threshold and the corresponding gas feature judgment result is consistent; and to generate a prompt signal or control command to prepare for flowering when the confidence level of the flowering morphology reaches a preset threshold and the corresponding gas feature judgment result is consistent. The iterative optimization module is used to manually confirm operation records and supplement training data corresponding to tiger claw shape and flowering time, iteratively train jasmine flower morphology recognition model and gas trend analysis model, and supports building exclusive discrimination models for each data acquisition device that are adapted to fixed types of jasmine flowers and tea leaves.

9. A computer device, characterized in that, It includes a processor and a non-transitory computer-readable storage medium storing a computer program, which, when executed by the processor, implements the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method of any one of claims 1-7.

Images