Abnormality tracing method and device, equipment and storage medium

Abstract

The invention relates to an abnormity tracing method, device and equipment and a storage medium, and the method comprises the steps: carrying out the abnormity recognition of the growth state of a target traditional Chinese medicinal material plant, and obtaining the growth abnormity information of the target traditional Chinese medicinal material plant; based on the digital model of the target traditional Chinese medicinal material plant, respectively simulating and predicting the influence degree of each piece of information in the target information on the quality problem of the target traditional Chinese medicinal material plant in the future to obtain an evaluation value corresponding to each piece of information so as to determine the quality risk level of the target traditional Chinese medicinal material plant; and acquiring abnormal traceability data according to the quality risk level by adopting a planting stage traceability strategy. Thus, after the quality risk level is determined through abnormal identification of the plant growth state and simulation prediction of the digital model, the abnormal traceability data can be acquired from the multi-dimensional data in a targeted manner based on the planting stage traceability strategy, screening of massive irrelevant data is avoided, and the traceability speed can be greatly improved.

Description

Technical Field

[0001] This invention relates to the field of data traceability technology, specifically to an anomaly tracing method, apparatus, equipment, and storage medium. Background Technology

[0002] As the foundation of the traditional Chinese medicine industry, the quality of medicinal herbs cultivation directly affects clinical efficacy and industrial development.

[0003] Currently, to ensure the quality of medicinal herbs, data related to abnormal plant growth can be traced during the planting period, and subsequent planting strategies can be adjusted accordingly to intervene in the plants in a timely manner.

[0004] However, there are significant technical problems in traditional plant growth quality control technology: when plant quality problems occur, due to the complexity and unclear correlation of multi-dimensional data involved in the growth process, there is a lack of targeted information screening and positioning mechanisms, resulting in low efficiency in the traceability process, a lot of time costs, and due to the slow data processing speed, it is difficult to meet the actual needs of timely problem investigation and loss reduction, which seriously restricts the timeliness and effectiveness of plant growth quality control.

[0005] Therefore, proposing a method to improve the efficiency of tracing abnormal data in the cultivation of Chinese medicinal herbs is of great practical significance. Summary of the Invention

[0006] This invention provides an anomaly tracing method, apparatus, electronic device, and medium, aiming to improve the efficiency of tracing abnormal data during the cultivation of Chinese medicinal herbs.

[0007] In a first aspect, embodiments of this application provide an anomaly tracing method, the method comprising: Based on plant physiological data related to the growth status of the target Chinese medicinal plant, anomalies in the growth status of the target Chinese medicinal plant are identified to obtain abnormal growth information of the target Chinese medicinal plant. Based on the digital model of the target Chinese medicinal plant, the influence of each piece of information in the target information on the future quality problems of the target Chinese medicinal plant is simulated and predicted to obtain the evaluation value corresponding to each piece of information; wherein, the digital model is used to simulate based on data related to the growth of the target Chinese medicinal plant, and the target information includes the growth abnormality information and multi-dimensional data related to the growth process of the target Chinese medicinal plant; Based on the evaluation values ​​corresponding to each piece of information, the quality risk level of the target Chinese medicinal plant is determined; wherein, the quality risk level is used to characterize the degree of risk that the target Chinese medicinal plant will have quality problems in the future; Using a preset planting stage traceability strategy, based on the quality risk level, abnormal traceability data of the target Chinese medicinal plant is obtained from the multi-dimensional data; wherein, the abnormal traceability data is relevant data that causes growth abnormalities during the growth process of the target Chinese medicinal plant.

[0008] In one implementation, the aforementioned plant physiological data includes multiple physiological indicator data and growth environment data; the step of identifying anomalies in the growth state of the target medicinal plant based on the plant physiological data related to its growth state, and obtaining abnormal growth information of the target medicinal plant, includes: using a long short-term memory network with an attention mechanism of the preset coupling model to extract features from the multiple physiological indicator data and the growth environment data and perform feature splicing and fusion to obtain a feature vector; inputting the feature vector into the backpropagation neural network of the coupling model to obtain the abnormal growth information output by the backpropagation neural network.

[0009] In one implementation, the target information further includes the pest and disease detection results of the target medicinal plant; the pest and disease detection results of the target medicinal plant are obtained by a pest and disease identification model from the medicinal plant image data of the target medicinal plant; before simulating and predicting the impact of each piece of information in the target information on the future quality problems of the target medicinal plant based on the digital model, and obtaining the evaluation value corresponding to each piece of information, the method further includes: acquiring medicinal plant image data from multiple growers; using secure multi-party computation technology, obtaining aggregated data based on the medicinal plant image data from the multiple growers; using differential privacy technology to add noise to the aggregated data to obtain target aggregated data; and processing a preset initial identification model based on the target aggregated data to obtain a pest and disease identification model.

[0010] In one implementation, determining the quality risk level of the target medicinal plant based on the evaluation values ​​corresponding to each piece of information includes: weighting and summing the evaluation values ​​corresponding to each piece of information to obtain the target total evaluation value of the target medicinal plant; and determining the quality risk level corresponding to the target total evaluation value as the quality risk level of the target medicinal plant in the correspondence between the total evaluation value and the quality risk level.

[0011] In one implementation, the method includes: estimating a first growth parameter of the target medicinal plant based on the plant's physiological data and model data of a three-dimensional growth model of the target medicinal plant using an extended Kalman filter state estimator; and adjusting the parameters in the digital model if the difference between the first growth parameter and the second growth parameter of the target medicinal plant estimated by the digital model is greater than a preset parameter threshold.

[0012] In one implementation, the above-mentioned planting stage traceability strategy includes: determining the traceability request type corresponding to the quality risk level in the correspondence between risk level and traceability request type as the target traceability request type; the target traceability request type characterizes the data type being traced; determining the target three-dimensional precision model corresponding to the target traceability request type from multiple three-dimensional precision models; wherein, each three-dimensional precision model has a different precision; acquiring target data that matches the precision of the target three-dimensional precision model and is related to the abnormal growth of the target medicinal plant; simulating the target data in the target three-dimensional precision model to obtain abnormal traceability data.

[0013] In one implementation, the method further includes: during the processing of the target medicinal plant after harvesting, based on a digital model of the target medicinal plant, simulating and predicting the impact of each data point in the processing data on future quality problems of the target medicinal plant, and obtaining an evaluation value corresponding to each data point; wherein the processing data are parameter data related to the processing process of the target medicinal plant; determining the processing quality risk level of the target medicinal plant based on the evaluation values ​​corresponding to each data point; wherein the processing quality risk level is used to characterize the risk degree of quality problems that may occur in the target medicinal plant after processing; and using a preset processing stage traceability strategy, obtaining abnormal traceability processing data of the target medicinal plant from the processing data based on the processing quality risk level.

[0014] Secondly, embodiments of this application provide an anomaly tracing device, the device comprising: The identification module is used to identify abnormalities in the growth status of the target Chinese medicinal plant based on plant physiological data related to the growth status of the target Chinese medicinal plant, and to obtain abnormal growth information of the target Chinese medicinal plant. The evaluation module is used to simulate and predict the impact of each piece of information in the target information on the future quality problems of the target medicinal plant based on the digital model of the target medicinal plant, and obtain the evaluation value corresponding to each piece of information; wherein, the target information includes the growth abnormality information and multi-dimensional data related to the growth process of the target medicinal plant; The determination module is used to determine the quality risk level of the target Chinese medicinal plant based on the evaluation values ​​corresponding to each piece of information; wherein, the quality risk level is used to characterize the degree of risk that the target Chinese medicinal plant will have quality problems in the future; The traceability module is used to acquire abnormal traceability data of the target Chinese medicinal plant from the multi-dimensional data according to the quality risk level, based on a preset planting stage traceability strategy; wherein, the abnormal traceability data is relevant data that causes growth abnormalities during the growth process of the target Chinese medicinal plant.

[0015] Thirdly, embodiments of this application provide an electronic device, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the exception tracing method provided in the first aspect of the embodiments of this application.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the anomaly tracing method provided in the first aspect of embodiments of this application.

[0017] In this embodiment of the application, by identifying anomalies in plant growth status, combining digital model simulation to predict the impact of each target information on future quality problems and obtaining an evaluation value, and determining the quality risk level based on the evaluation value, the abnormal traceability data can be selectively obtained from multi-dimensional data based on the planting stage traceability strategy, avoiding the screening of massive amounts of irrelevant data and greatly improving the traceability speed. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the anomaly tracing method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the four-layer architecture system provided in the embodiments of this application; Figure 3 This is the process framework of the anomaly tracing method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the anomaly tracing device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0022] The anomaly tracing method, apparatus, electronic device, and storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0023] Figure 1 This is a flowchart illustrating the anomaly tracing method provided in an embodiment of this application, as shown below. Figure 1 As shown, the first aspect of this application provides an anomaly tracing method, including the following steps S100-S400: Step S100: Based on the plant physiological data related to the growth status of the target Chinese medicinal plant, anomaly identification is performed on the growth status of the target Chinese medicinal plant to obtain abnormal growth information of the target Chinese medicinal plant.

[0024] The aforementioned plant physiological data may include morphological characteristic data and physiological index data related to the plant's morphological characteristics. For example, morphological characteristic data may include plant height, crown width, number and area of ​​leaves, leaf area index, and plant structure. Crown width characterizes the width of the plant's canopy, leaf area index reflects the ratio of leaf area to ground surface area, and plant structure includes the number and angle of branches. Physiological index data may include stem microdeformation, sap flow rate, leaf surface temperature, photosynthetic efficiency, respiration rate, water use efficiency, chlorophyll content, etc. In some embodiments, plant physiological data can be collected using implanted physiological sensors.

[0025] The above-mentioned anomaly identification can be understood as timely detection of potential abnormal signs when an abnormal phenomenon has just appeared or has not yet manifested itself.

[0026] The aforementioned abnormal growth information can be information characterizing abnormal plant growth, such as information characterizing continuous cropping obstacles, fertilizer burn, and other abnormal states. This information can also include probability values ​​for the occurrence of various abnormal states.

[0027] In some embodiments, when the yellowing degree of the leaves is detected to be greater than a preset yellowing degree threshold, a local irrigation command is triggered to the irrigation equipment so that the irrigation equipment irrigates the target Chinese medicinal plant.

[0028] In one embodiment, step S100 above may involve obtaining abnormal growth information of the target medicinal plant by comparing each data point in the plant's physiological data with the corresponding preset physiological indicator thresholds, and combining this with a preset reference table. The preset reference table represents the correspondence between the comparison results of each data point with the corresponding preset physiological indicator thresholds and the abnormal state information. The abnormal growth information may include multiple abnormal state information.

[0029] In this embodiment, the type of abnormality can be determined by integrating the coordinated changes of multiple physiological data, thereby improving the accuracy of identification.

[0030] In another embodiment, step S100 above may be to identify anomalies in the growth status of the target Chinese medicinal plant by using a preset coupling model based on plant physiological data related to the growth status of the target Chinese medicinal plant, and obtain abnormal growth information of the target Chinese medicinal plant. The coupling model includes a long short-term memory network with attention mechanism and a backpropagation neural network.

[0031] Specifically, the plant physiological data includes multiple physiological indicators and growth environment data. A long short-term memory network with attention mechanism of a pre-set coupled model is used to extract features and splice and fuse features from the multiple physiological indicators and growth environment data to obtain feature vectors. The feature vectors are then input into the backpropagation neural network of the coupled model to obtain the growth abnormality information output by the backpropagation neural network.

[0032] In this embodiment, the plant growth anomaly identification is performed by a coupled model of a Long Short-Term Memory (LSTM) network with an attention mechanism and a Backpropagation (BP) neural network. This model can give full play to the advantages of each network. LSTM can effectively capture long-term dynamic dependencies in the physiological time series data of plants, the attention mechanism can adaptively focus on key time nodes or feature dimensions that are sensitive to anomalies, and the BP neural network is good at deep feature mapping and pattern classification. The synergistic effect of the three can accurately extract subtle abnormal precursor features in the growth process, significantly improving the accuracy of early anomaly identification and early warning time.

[0033] In some embodiments, before identifying anomalies in the growth status of the target medicinal plant based on plant physiological data related to its growth status using a preset coupling model to obtain information on abnormal growth of the target medicinal plant, the following steps are also included: Obtain the training set.

[0034] The training set includes multiple training samples, each of which contains data on various physiological indicators, growth environment data, and labels for abnormal growth information related to the growth process of Chinese medicinal plants.

[0035] The initial coupled model is trained using training samples to obtain the coupled model.

[0036] During the training process, the following steps are performed for each training sample: A long short-term memory network with an attention mechanism using an initial coupling model is used to extract and fuse features from multiple physiological index data and growth environment data to obtain feature vectors.

[0037] The feature vector is input into the backpropagation neural network to obtain the predicted growth anomaly information output by the backpropagation neural network.

[0038] The loss function value is determined based on the predicted growth anomaly information and growth anomaly information labels corresponding to each training sample.

[0039] When the loss function value reaches the training stopping condition, the initial coupled model is determined to be a coupled model.

[0040] When the loss function value does not reach the training stopping condition, the network parameters of the initial coupled model are adjusted, and the process is returned to execute the steps of using the long short-term memory network with attention mechanism of the initial coupled model to extract features and splice and fuse features from multiple physiological index data and environmental data for each training sample to obtain feature vectors.

[0041] In one embodiment, the training set may include training samples of different Chinese medicinal plants, such as ginseng, angelica, astragalus and other medicinal plants. The multiple physiological indicators may include stem microdeformation, sap flow rate, leaf surface temperature and other physiological indicators of the medicinal plants. The growth environment data may include spectral reflectance data, soil enzyme activity and so on.

[0042] The aforementioned growth environment data refers to data related to the growth environment of the target medicinal plant. This data may include surface spectral data, underground root microenvironment data, air temperature, air humidity, and light intensity. Surface spectral data refers to the reflectance, emission, or transmission spectral information of surface objects (such as vegetation, soil, and water bodies) within different wavelength ranges. Surface spectral data can be used to identify and classify land cover types, monitor vegetation health, and assess soil characteristics. Underground root microenvironment data refers to soil microenvironment information related to plant root growth, including soil temperature, humidity, pH value, nutrient content, and microbial community structure. This data reflects the root growth conditions and health status. In some implementations, surface spectral data can be collected using a surface spectral acquisition module, underground root microenvironment data can be collected using an underground root microenvironment probe, air temperature and humidity can be collected using temperature and humidity sensors respectively, and light intensity can be collected using an ambient light sensor.

[0043] In this embodiment, when the LSTM with attention mechanism extracts features from multiple physiological indicators and growth environment data, it can dynamically weight key features sensitive to anomalies through the attention mechanism. Simultaneously, the LSTM captures long-term dependencies in time-series data, and then integrates complementary information from multiple sources through feature concatenation and fusion. The resulting feature vector is more representative and discriminative. After being input into the BP neural network, the BP network can focus on pattern learning of abnormal states based on high-quality fused features, reducing the interference of redundant information or noise in the original data. This allows for more accurate output of the probability of occurrence of various abnormal states, reducing false positives or false negatives, and improving model accuracy. Furthermore, the feature extraction and concatenation fusion process of the LSTM pre-converts high-dimensional, multi-source original data into low-dimensional, compact feature vectors, significantly reducing the data dimensionality input to the BP neural network. When the BP network processes low-dimensional feature vectors, the computational complexity is significantly reduced, not only accelerating the parameter iteration efficiency during training but also enabling the model to quickly complete feature mapping and probability calculation during the recognition of new data after training, shortening the recognition time per sample and improving the overall recognition speed.

[0044] Step S200: Based on the digital model of the target Chinese medicinal plant, simulate and predict the degree of influence of each piece of information in the target information on the future quality problems of the target Chinese medicinal plant, and obtain the evaluation value corresponding to each piece of information.

[0045] The digital model is used to simulate the growth of the target Chinese medicinal plant based on data related to its growth. The data related to the growth of the target Chinese medicinal plant can include at least one of the following: multiple physiological indicators of the target Chinese medicinal plant, growth environment data, growth abnormality information, etc. In other words, the digital model can not only simulate the growth status of the target Chinese medicinal plant, but also simulate the growth environment around the target Chinese medicinal plant.

[0046] The target information includes information on abnormal growth and multi-dimensional data related to the growth process of the target Chinese medicinal plant.

[0047] The aforementioned multi-dimensional data may include plant physiological data and plant growth management data, etc.

[0048] The plant growth management data mentioned above represents data on human intervention in the plant cultivation process to improve yield and quality. This includes factors such as irrigation frequency, fertilization intervals and amounts, and pruning frequency. This plant growth management data can be manually uploaded to electronic devices.

[0049] The aforementioned digital model refers to a virtual model created in virtual space through digital means that corresponds to a physical entity. It can reflect the state, behavior, and performance of the physical entity in real time and perform simulation, analysis, and optimization based on the data.

[0050] In some implementations, step S200 may include: A three-dimensional growth model of the target Chinese medicinal plant was constructed based on plant physiological data.

[0051] A digital model of the target Chinese medicinal plant was used to simulate the model data, multi-dimensional data, and abnormal growth information of the three-dimensional growth model.

[0052] Using a pre-defined prediction model and based on the simulation results of the digital model, the degree of influence of each piece of information in the predicted target information on the future quality problems of the target Chinese medicinal plant is simulated.

[0053] The aforementioned three-dimensional growth model is a mathematical and computer graphics model used to simulate and visualize the growth process of plants in three-dimensional space. The model data is used to map the growth status of the target medicinal plant.

[0054] In some implementations, the aforementioned construction of a three-dimensional growth model of the target medicinal plant based on plant physiological data can be achieved by combining plant physiological data with a mathematical model of plant growth (such as the Lindenmayer Systems model) to create a three-dimensional virtual plant model, thus obtaining a three-dimensional growth model. When the plant physiological data is updated, the model parameters are dynamically adjusted based on real-time data, allowing the virtual plant to simulate the growth process of a real plant.

[0055] The application of digital models of Chinese medicinal plants can also achieve growth monitoring and anomaly warning mainly through multi-dimensional data collection and AI algorithms. Data such as temperature, humidity, light, and soil nutrients are collected in real time through IoT devices, and combined with environmental dimension curve analysis, two-dimensional / three-dimensional growth models of Chinese medicinal plants are generated.

[0056] In this embodiment, the digital model serves as a precise virtual mirror of the physical plant, visually reproducing the plant's morphology and growth dynamics through a three-dimensional growth model. Combined with multi-dimensional data, a complete growth correlation network is constructed, while the integration of abnormal growth information provides key anomaly trigger points for prediction. This integration ensures the accuracy of the prediction while also quantifying and visualizing the degree of impact. Simultaneously, the collaborative analysis of multi-dimensional data avoids the one-sidedness of single-factor predictions, accurately locating the key impact paths and strengths of quality issues, and improving the accuracy of quantification.

[0057] Step S300: Determine the quality risk level of the target Chinese medicinal plant based on the evaluation values ​​corresponding to each piece of information.

[0058] Among them, the quality risk level is used to characterize the degree of risk that the target Chinese medicinal plant will have quality problems in the future.

[0059] In some implementations, step S300 above may involve using a preset machine learning algorithm to predict the quality risk level of the target Chinese medicinal plant based on various information and evaluation values.

[0060] The aforementioned machine learning algorithms can be decision trees, random forests, support vector machines, etc.

[0061] For example, by training a random forest model, inputting various information and their evaluation values, the model can output the quality risk level of the plant.

[0062] In this embodiment, machine learning algorithms can automatically learn the complex relationships between various pieces of information, thereby improving the accuracy of risk assessment.

[0063] In other embodiments, step S300 above may involve weighted summation of the evaluation values ​​corresponding to each piece of information to obtain a target total evaluation value, and determining the quality risk level corresponding to the target total evaluation value as the quality risk level of the target Chinese medicinal plant in the correspondence between the total evaluation value and the quality risk level.

[0064] For example, based on preset rules, the total assessment value is correlated with different quality risk levels. For instance, the total assessment value can be divided into different intervals, each corresponding to a specific quality risk level. Assuming the total assessment value ranges from 0 to 100, the correspondence between the total assessment value and the quality risk level can be defined as follows: a total assessment value between 0 and 30 indicates low risk (Level III), 31 to 60 indicates medium risk (Level II), and 61 to 100 indicates high risk (Level I). In this way, the corresponding quality risk level can be quickly determined based on the target total assessment value, thereby providing a basis for quality management decisions.

[0065] In this embodiment, a weighted summation method is used to process the evaluation values ​​corresponding to each piece of information and determine the quality risk level of the plant. This method comprehensively considers multiple factors related to plant growth, fully reflecting the plant's growth status and avoiding the one-sidedness of single-indicator evaluation. Secondly, by comparing the total evaluation value obtained through weighted summation with a preset threshold, three quality risk levels—low, medium, and high—can be quickly and intuitively identified, facilitating management and decision-making. Furthermore, the flexible adjustment of weights allows this method to be optimized for different plant species or growth environments, more accurately reflecting the actual impact of each factor on plant growth.

[0066] In some embodiments, the target information may further include the results of pest and disease detection of the target medicinal plant.

[0067] For example, the quality risk level is graded using three colors: green, yellow, and red. The quality risk assessment is carried out by integrating multi-source data in real time. The main data sources include plant growth abnormality information, pest and disease detection results, and plant physiological data. A weighted scoring algorithm is used to assign weights to different types of data. For example, the weight of pest and disease detection results can be set to 40%, the weight of plant physiological data to 20%, the weight of environmental data to 20%, and the weight of growth abnormality information to 20%. The quality risk level is then divided according to the comprehensive score. A score below 60 is judged as a low-risk green level, a score between 60 and 80 is judged as a medium-risk yellow level, and a score above 80 is judged as a high-risk red level.

[0068] In this embodiment, pests and diseases are one of the important factors affecting plant health. When determining the quality risk level of a plant, incorporating the assessment values ​​of pest and disease detection results into a weighted summation calculation system can more comprehensively reflect the plant's growth status and avoid inaccurate risk assessments due to neglecting pest and disease factors. By quantifying the severity of pests and diseases, the health risk of the plant can be assessed more accurately.

[0069] In some embodiments, the results of pest and disease detection of the target Chinese medicinal plant are obtained by the pest and disease identification model from the image data of the target Chinese medicinal plant.

[0070] The aforementioned pest and disease identification models can be convolutional neural networks (CNN), deep learning models, or machine learning models, etc.

[0071] Prior to step S200, the method may further include the following steps: Obtain image data of medicinal plant plants from multiple growers.

[0072] Using secure multi-party computation technology, aggregated data of medicinal plant images from multiple growers is obtained for the aggregated model.

[0073] Differential privacy technology is used to add noise to the aggregated data to obtain the target aggregated data.

[0074] The initial identification model is processed based on the target aggregated data to obtain the pest and disease identification model.

[0075] In one implementation, the above-mentioned secure multi-party computation technology is used to obtain aggregated data based on the image data of Chinese medicinal plants from multiple growers. This can be achieved by obtaining the calculation results from the node devices corresponding to each grower, and integrating the multiple calculation results through a preset reconstruction algorithm to obtain the aggregated data.

[0076] The above calculation result is obtained by the first node device of the corresponding grower calculating the received sub-secret according to the preset algorithm. The sub-secret is obtained by the second node device of other growers (excluding the first node device) by segmenting the encrypted data.

[0077] In one implementation, the aggregated data is the parameter data of the model. The encrypted data is obtained by encrypting the model parameters of a preset initial model using the second node device. The initial model is trained by the second node device using local medicinal herb plant image data. The initial model is used for pest and disease identification based on the medicinal herb plant image data. The preset algorithm can be an additive calculation. The above-mentioned processing of the preset initial identification model based on the target aggregated data to obtain a pest and disease identification model can be based on updating the parameters of the preset initial identification model to obtain a pest and disease identification model with updated parameters.

[0078] In another implementation, the aggregated data is image data of medicinal plants, and the encrypted data is obtained by encrypting the local medicinal plant image data using the second node device. The above-mentioned processing of the preset initial recognition model based on the target aggregated data to obtain the pest and disease recognition model can be achieved by training the preset initial recognition model based on the target aggregated data to obtain the trained pest and disease recognition model.

[0079] For example, the secure multi-party computation technology specifically employs a secret-sharing algorithm to divide each grower's encrypted data into multiple sub-secrets and distribute them to different participants (node ​​devices). Each participant, without knowing the complete original data, performs calculations on the received sub-secrets and returns the calculation results to the federated learning data gateway. The federated learning data gateway integrates these calculation results through a specific reconstruction algorithm to obtain aggregated data.

[0080] In this embodiment, no single participant can obtain the original data of other growers, thus ensuring data privacy. Differential privacy protection is enhanced by adding interference noise conforming to a Laplace distribution to the aggregated encrypted data. The intensity of the noise can be adjusted according to a preset privacy budget. The privacy budget determines the balance between the degree of data privacy protection and usability. By reasonably setting the privacy budget, data privacy can be effectively protected, while ensuring that the aggregated data is usable for updating the pest and disease identification model. After adding noise, the data is encrypted a second time to generate the final aggregated data.

[0081] Prior to step S200, the method may further include the following steps: By using an extended Kalman filter state estimator, the first growth parameters of the target Chinese medicinal plant are estimated based on plant physiological data and model data from a three-dimensional growth model of the target Chinese medicinal plant.

[0082] If the difference between the first growth parameter and the second growth parameter of the target medicinal plant predicted by the digital model is greater than the preset parameter threshold, the parameters in the digital model are adjusted.

[0083] The extended Kalman filter state estimator described above is used to predict the first growth parameter of the target Chinese medicinal plant based on plant physiological data and model data of the three-dimensional growth model. The first growth parameter can be understood as the predicted actual growth parameter.

[0084] The above preset parameter thresholds can be set according to requirements; the higher the required precision, the lower the preset parameter thresholds.

[0085] For example, the state estimator based on extended Kalman filter continuously receives plant physiological data, environmental data, and three-dimensional growth model data output by a lightweight twin modeling engine from a multimodal sensor array. Then, the nonlinear system is linearized by the extended Kalman filter algorithm. Through prediction and update steps, the deviation between the digital model and the physical plant in growth parameters is calculated in real time, thereby forming a twin error correction mechanism.

[0086] In this embodiment, when the difference between the growth parameters evaluated by the extended Kalman filter state estimator and the plant growth parameters predicted by the digital model is greater than a preset parameter threshold, the parameters in the digital model are adjusted to make them closer to the actual growth state, thereby reducing the difference between the model prediction and the actual measurement and significantly improving the simulation accuracy of the digital model for the plant growth state.

[0087] Step S400: Using a preset planting stage traceability strategy, abnormal traceability data of the target Chinese medicinal plant is obtained from multi-dimensional data based on the quality risk level.

[0088] In one implementation, the planting stage traceability strategy is used to determine the traceability data of the target stage in the growth process of the target Chinese medicinal plant. The target stage is different for different quality risk levels. The target stage is determined from multiple growth stages in the growth process of the target Chinese medicinal plant. The target stage includes at least one growth stage. The higher the quality risk level, the more growth stages the target stage includes.

[0089] The above-mentioned planting stage traceability strategy can characterize the correspondence between the quality risk level and the traceability data of the target stage. The above-mentioned planting stage traceability strategy adopts a preset planting stage and obtains abnormal traceability data of the target Chinese medicinal plant from multi-dimensional data according to the quality risk level. It can be that the traceability data of the target stage corresponding to the quality risk level is determined from the preset planting stage traceability strategy according to the quality risk level, and the traceability data of the target Chinese medicinal plant in the target stage is obtained from multi-dimensional data to obtain abnormal traceability data.

[0090] For example, multiple growth stages may include a germination stage, a growth stage, a flowering stage, a fruiting stage, and a maturity stage. The germination stage can be the period from seed sowing to seedling emergence. The growth stage can be the period from seedling emergence to the emergence of a certain number of leaves. The flowering stage can be the period when the plant begins to flower. The fruiting stage can be the period when the plant enters the fruiting stage. The maturity stage can be the period when the fruit matures and is ready for harvest. Quality risk levels are divided into low risk, medium risk, and high risk. The traceability data for the target stage corresponding to low risk is the key physiological and environmental data for the current stage. The traceability data for the target stage corresponding to medium risk is all physiological, environmental, and pest / disease data for the current stage and the previous stage. The traceability data for the target stage corresponding to high risk is all data for the current stage and the two stages preceding it. Assuming the current time is July 31, 2025, and the plant's growth status is being monitored, based on the actual growth of the plant, it is determined that the plant is currently in the fruiting stage. During the fruiting stage, the plant is determined to be at a medium risk level. According to the traceability strategy, it is necessary to trace all physiological, environmental and pest data of the current stage (fruiting period) and the previous stage (flowering period).

[0091] In another embodiment, the above method may further include the following steps: In the correspondence between risk level and traceability request type, the traceability request type corresponding to the quality risk level is determined as the target traceability request type.

[0092] The traceability request type represents the data type being traced.

[0093] Determine the target 3D precision model corresponding to the target tracing request type from multiple 3D precision models.

[0094] The accuracy of each 3D precision model varies.

[0095] Acquire target data that matches the accuracy of the target's three-dimensional precision model and is related to abnormal growth of the target Chinese medicinal plant.

[0096] The target data is simulated in a three-dimensional precision model of the target to obtain anomaly tracing data.

[0097] The aforementioned three-dimensional accuracy model can be defined based on three dimensions: temporal resolution, spatial resolution, and parameter dimension, as quantitative indicators.

[0098] Different types of traceability requests may correspond to different levels of precision, and different levels of precision may correspond to different temporal resolutions, spatial resolutions, and parameter dimensions.

[0099] The above method of determining the target three-dimensional accuracy model corresponding to the target traceability request type from multiple three-dimensional accuracy models can be achieved by determining the accuracy level corresponding to the target traceability request type in the correspondence between traceability request type and accuracy level as the target accuracy level, and determining the three-dimensional accuracy model corresponding to the target accuracy level in the correspondence between accuracy level and three-dimensional accuracy model as the target three-dimensional accuracy model.

[0100] The data type of the target data can be the same as the data type being traced as represented by the target tracing request type.

[0101] For example, the accuracy level is automatically matched according to the type of traceability request. Combined with the three-dimensional accuracy model, the corresponding accuracy level is automatically matched. If it is the traceability of the growth process of a single plant, the model with high temporal resolution and single plant spatial resolution is selected first. If it is the traceability of the overall quality of the plot, the model with plot spatial resolution and appropriate temporal resolution is selected. Then, the algorithm retrieves the model data of the corresponding accuracy and generates the latest parameters after calibration according to the twin error correction mechanism. This ensures the accuracy of traceability while improving the system response efficiency and resource utilization.

[0102] In this embodiment, the corresponding traceability request type is determined based on the quality risk level, and a matching three-dimensional precision model is selected to simulate the target data related to abnormal plant growth. This approach significantly improves the targeting and efficiency of traceability, ensuring the accuracy and usability of the data. By dynamically adjusting the data precision, resources are rationally allocated, avoiding resource waste.

[0103] In some embodiments, the above method may further include the following steps: The type of traceability request is determined based on the traceability request initiated by the user.

[0104] The above traceability request carries information that indicates the type of traceability request.

[0105] Determine the target 3D precision model corresponding to the traceability request type from the set of 3D precision models.

[0106] Acquire target data that matches the accuracy of the target's three-dimensional precision model and is related to abnormal growth of the target Chinese medicinal plant.

[0107] The target data is simulated in a three-dimensional precision model of the target to obtain anomaly tracing data.

[0108] In some embodiments, the above method may further include the following steps: During the processing of the target Chinese medicinal plant after harvesting, a digital model based on the target Chinese medicinal plant is used to simulate and predict the impact of each data point in the processing data on the future quality problems of the target Chinese medicinal plant, and to obtain the evaluation value corresponding to each data point; among them, the processing data refers to the parameter data related to the processing process of the target Chinese medicinal plant.

[0109] Based on the assessment values ​​corresponding to each data point, the processing quality risk level of the target Chinese medicinal plant is determined; among which, the processing quality risk level is used to characterize the degree of risk of quality problems occurring in the target Chinese medicinal plant after processing.

[0110] Using a pre-defined processing stage traceability strategy, abnormal processing data of target Chinese medicinal plants are obtained from the processing data based on the processing quality risk level.

[0111] The specific steps in this embodiment can be referred to the above steps for obtaining abnormal source data, and will not be repeated here.

[0112] In this embodiment of the application, by identifying anomalies in plant growth status, combining digital model simulation to predict the impact of each target information on future quality problems and obtaining an evaluation value, and determining the quality risk level based on the evaluation value, the abnormal traceability data can be selectively obtained from multi-dimensional data based on the corresponding planting stage traceability strategy, avoiding the screening of massive amounts of irrelevant data and greatly improving the traceability speed.

[0113] To better understand the above method, this application also provides a specific and complete embodiment, as follows: Reference Figure 2 , Figure 2 This is a schematic diagram of the four-layer architecture system disclosed in an embodiment of this application. The four-layer architecture system is used to implement the above-described method. Figure 3 This is the process framework of the anomaly tracing method disclosed in the embodiments of this application, combined with Figure 2 and Figure 3 The specific complete embodiments will be described below.

[0114] S1. Construct a four-layer architecture system: Construct a four-layer architecture system comprising an intelligent sensing terminal layer (intelligent sensing terminal devices), an edge intelligent processing layer (edge ​​intelligent processing devices), a cloud-based digital twin platform, and an application service layer (application service devices), wherein: The intelligent sensing terminal layer deploys a multimodal sensor array and biodegradable tags. The multimodal sensor array includes an implantable physiological sensor, a surface spectral acquisition module, and an underground root microenvironment probe, which are used to collect plant physiological data, surface spectral data, and underground root microenvironment data in real time. The biodegradable tags use paper-based NFC chip packaging technology and have built-in RFID chips and temperature, humidity, and ambient light sensors. They store basic information such as medicinal herb varieties and planting plot numbers. All data are transmitted to the edge intelligent processing layer in real time through the LoRa wireless communication module. The edge intelligent processing layer deploys a lightweight twin modeling engine and a federated learning data gateway. The lightweight twin modeling engine is based on an improved U-Net convolutional neural network and is used to receive plant physiological data, ground surface spectral data, etc. transmitted from the intelligent sensing terminal layer, providing computational support for subsequent single-plant growth modeling and anomaly early warning. The federated learning data gateway adopts secure multi-party computation technology and differential privacy technology to build a regional data aggregation channel, which can aggregate pest and disease image data of growers in the region. The cloud-based digital twin platform constructs dynamic digital models, develops a smart contract gradient triggering mechanism, and establishes a three-level data encryption mechanism and a blockchain dynamic sharding algorithm, providing underlying support for subsequent digital twin simulation and traceability strategies and data security processing. The application service layer is designed with a three-dimensional semantic traceability interface and a quality credit evaluation system for traceability queries and credit evaluation. S2. Multi-dimensional data acquisition: Physiological index data, surface spectral data, and underground root microenvironment data are collected in real time through a multi-modal sensor array to form a multi-modal physiological feature dataset. Degradable tags are used to automatically degrade and release organic fertilizer during the harvest period. At the same time, basic information about the plot and medicinal materials is written through NFC chips. In addition, pest and disease image data (image data of Chinese medicinal plants) are uploaded by growers. After all data is transmitted to the edge layer through the LoRa module, the edge intelligent processing layer is triggered to perform preliminary cleaning and feature extraction on the data, providing raw data input for edge growth modeling. Plant growth management data can be written through degradable tags.

[0115] The S2 multimodal sensor array includes an implantable physiological sensor, a surface spectral acquisition module, and an underground root microenvironment probe. The implantable physiological sensor is used to monitor plant physiological indicators, the surface spectral acquisition module is used to acquire reflectance data in the 200-2500nm band, and the underground root microenvironment probe is used to monitor rhizosphere soil enzyme activity and microbial community conductivity.

[0116] S3. Edge-end growth modeling and early warning: Using a lightweight twin modeling engine, a three-dimensional growth model of a single medicinal plant is constructed at the edge node based on the plant physiological data collected in S2. The model uses an attention mechanism LSTM network model coupled with a BP neural network to achieve early identification of 12 abnormal states such as continuous cropping obstacles and fertilizer damage, thus realizing early warning of abnormal growth. When the yellowing degree of leaves is detected to be >30%, a local irrigation command is triggered. The generated growth abnormality report is simultaneously uploaded to the cloud digital twin platform for dynamic digital model updates and traceability strategy triggering. The abnormal growth warning steps in S3 include: S31. Through a multimodal sensor array, implanted physiological sensors are used to collect 12 physiological indicators, such as stem microdeformation, sap flow rate, and leaf surface temperature, of 8 kinds of traditional Chinese medicinal materials, including ginseng, angelica, and astragalus, in real time at the minute level. At the same time, the surface spectral acquisition module and the underground root microenvironment probe acquire environmental data such as spectral reflectance and soil enzyme activity. The collected data is transmitted to the edge intelligent processing layer via the LoRa wireless communication module. After data cleaning and standardization, it is stored in the time series database using InfluxDB according to the timestamp order. S32. Based on multimodal data in a time-series database, an attention-based LSTM network model is constructed. The forget gate, input gate, and output gate structure of the LSTM network are combined with an attention mechanism that can dynamically focus on key data features. The data in the database is divided into training set, validation set, and test set. 24-hour data is used as a sample slice. Each sample contains time-series data of 12 physiological indicators. After being input into the model, the features are extracted by the LSTM layer, weighted by the attention layer, and output by the fully connected layer. After multiple rounds of iterative training and parameter tuning, the model can achieve early identification of 12 abnormal states such as continuous cropping obstacles and fertilizer damage, with an early warning time of >72 hours. S33. Design a physiological and environmental coupled early warning model. Extract 12 physiological indicators of plants and environmental data such as surface spectral reflectance and soil microbial conductivity from a time series database. Form an input vector by feature splicing and fusion. Construct a multi-layer BP neural network architecture. The number of nodes in the input layer is determined according to the dimension of the fused features. Set 2-3 hidden layers. The output layer is the probability value of 12 abnormal states. Use the backpropagation algorithm to adjust the weights and combine the output results of the attention mechanism LSTM network model for joint training. By establishing a medicinal plant growth cycle model, divide it into seedling stage, growth stage, flowering stage and other stages. Analyze the fluctuation range of physiological and environmental indicators under normal conditions in historical data of each stage. Calculate the mean and standard deviation to determine the initial early warning threshold boundary. Use the sliding window method to statistically analyze the mean and variance of the past 7 days of data to dynamically adjust the threshold boundary in real time. The data collected in step S2 forms the basis for growth anomaly early warning. The time series database provides data support for the training of the attention mechanism LSTM network model and the dynamic early warning of the BP neural network coupled model. The attention mechanism LSTM network model realizes anomaly identification. The physiological and environmental coupled early warning model integrates environmental factors to improve the accuracy of early warning. The early warning results are used by the smart contract gradient triggering mechanism to determine the quality risk level and update the state of the digital model.

[0117] S4. Federated Learning Model Update: Through the federated learning data gateway, without leaking the original data, the AI ​​recognition model is periodically updated by aggregating the pest and disease image data uploaded by each grower during the collection process in S2, and the optimized model parameters are sent to the edge intelligent processing layer for accurate identification of pests and diseases in the abnormal growth warning. The S4 federated learning data gateway employs secure multi-party computation (MPC) and differential privacy technologies to aggregate pest and disease image data from growers within a region without disclosing the original data. The updated AI recognition model achieves a 98.7% accuracy rate in identifying pests and diseases. Specifically, the secure MPC technology uses a secret-sharing algorithm to divide each grower's encrypted data into multiple sub-secrets, which are distributed to different participants. Each participant, unaware of the complete original data, performs calculations on the received sub-secrets and returns the results to the federated learning data gateway. The gateway then integrates these results using a specific reconstruction algorithm to obtain the aggregated encrypted data. During this process, no single participant can access the original data of other growers, ensuring data privacy. Differential privacy protection is enhanced by adding Laplace-distributed interference noise to the aggregated encrypted data. The noise intensity is adjusted according to a preset privacy budget, which balances data privacy protection with usability. By setting a reasonable privacy budget, data privacy is effectively protected while ensuring the aggregated data's usability for updating the AI ​​recognition model. After adding noise, the data is encrypted a second time to generate the final aggregated data.

[0118] S5. Digital Twin Simulation and Traceability Strategy: Based on a cloud-based dynamic digital model, it integrates the 3D growth model of a single medicinal plant uploaded from the edge in S3, the multi-dimensional data in S2, and the updated AI recognition model in S4 to achieve full-element simulation of the growth cycle. It also uses a traceability accuracy self-optimization method to calibrate the deviation between the twin and the physical plant in real time. Based on the anomaly warning level generated in S3, different traceability strategies are automatically executed by a smart contract gradient triggering mechanism. It can also execute traceability processes for the planting and processing stages, with the specific processes as follows: Planting stage traceability process: A three-dimensional geological model is generated by using plot images and soil probe data obtained by drones. Combined with the growth anomaly report in S3 and the pest and disease identification results in S4, the compliance of inputs is intelligently verified, and the results are stored on the blockchain. Processing stage traceability process: Based on the digital model to predict the optimal harvest time, during the processing, the temperature field of the drying room, the wear status of the slicer and other parameters are mapped in real time through the equipment digital twin model. The data is uploaded to the blockchain through the blockchain dynamic slicing algorithm to ensure the traceability of the process. In S5, the smart contract gradient triggering mechanism automatically executes different traceability strategies based on the quality risk level. The quality risk level is graded using three colors: green, yellow, and red. The smart contract gradient triggering mechanism performs quality risk assessment by integrating multi-source data in real time. The main data sources include plant growth anomaly reports generated by the lightweight twin modeling engine, pest and disease detection results output by the AI ​​recognition model, and environmental and plant physiological data collected by the multimodal sensor array. A weighted scoring algorithm is used to assign weights to different types of data. For example, the weight of pest and disease detection results can be set to 40%, plant physiological anomaly data to 30%, and environmental data to 30%. The quality risk level is then divided according to the comprehensive score. A score below 60 is judged as a low-risk green level, 60-80 as a medium-risk yellow level, and above 80 as a high-risk red level. When the detection data triggers a red warning, the full-node notarization function of the three-level data encryption mechanism is invoked. The green scenario uses a lightweight node fast consensus.

[0119] The full-element simulation step of the growth cycle in S5 also includes a method for self-optimization of traceability accuracy: S51. The state estimator based on extended Kalman filter continuously receives plant physiological data, environmental data, and single-plant growth model data output by the lightweight twin modeling engine from the multimodal sensor array. Then, it uses the extended Kalman filter algorithm to linearize the nonlinear system. Through prediction and update, it calculates the deviation between the digital model and the physical plant in growth parameters in real time, thereby forming a twin error correction mechanism.

[0120] The tracing strategy in S5 is as follows: S52. Construct and develop a digital twin version management system. During the growth process of medicinal materials, the system automatically saves a complete snapshot of the digital model at the corresponding time based on preset key time nodes or special events triggered by abnormal growth warnings. Each snapshot contains information such as the three-dimensional growth model, environmental parameters, and physiological indicators of that stage. It is stored on the cloud digital twin platform using distributed storage technology. Through the digital twin version management system, the status of medicinal materials at different growth stages can be quickly traced back, making it convenient to compare and analyze growth change trends during the traceability process, and providing historical data support for quality assessment and problem tracing. S53. Define quantitative indicators for three dimensions: temporal resolution, spatial resolution, and parameter dimension, and construct a three-dimensional accuracy model for different varieties of Chinese medicinal materials and traceability scenarios. S54. Design an intelligent adjustment algorithm to automatically match the accuracy level according to the traceability request type. When a user initiates a traceability request through the 3D semantic traceability interface, the intelligent adjustment algorithm first parses the request type, and then automatically matches the corresponding accuracy level based on the request type and the 3D accuracy model. If it is a traceability of the growth process of a single plant, it prioritizes the model with high temporal resolution and single plant spatial resolution. If it is a traceability of the overall quality of the plot, it selects the model with plot spatial resolution and appropriate temporal resolution. Then, the algorithm retrieves the model data of the corresponding accuracy from the twin version management system, and quickly generates the traceability result that meets the user's needs based on the latest parameters after calibration according to the twin error correction mechanism. While ensuring the accuracy of traceability, it improves the system response efficiency and resource utilization.

[0121] The S5 planting stage traceability process includes: Land parcel initialization: A drone equipped with a multispectral camera acquires images of the land parcel, which are then combined with soil probe data to generate a three-dimensional geological model. The smart terminal automatically assigns a digital identity to the land parcel and uploads it to the blockchain to generate the genesis block. Growth monitoring: Implanted sensors collect plant physiological data, edge nodes generate individual plant growth abnormality reports, and federated learning gateways update AI recognition models; Input Management: In the early stages of planting, growers input basic information such as the name, ingredients, and instructions for use of pesticides, fertilizers, and other inputs through the application service layer interface. The system simultaneously scans the input packaging using image recognition technology, automatically extracts key information, and cross-validates it with the input data to ensure accuracy. Based on this information and combined with an agricultural knowledge graph, a digital model containing a kinetic simulation model is constructed for each input. When information such as the time, dosage, and plot of use of an input is recorded, the system automatically retrieves the corresponding digital model. Using real-time environmental data and data on the growth stage of medicinal herbs as input, it dynamically simulates the residual degradation process of the input in the soil and plants. Through a cloud-deployed smart contract for input compliance verification, the system automatically retrieves compliance standards such as applicable scope, maximum dosage, and safe interval stored in the digital model. Combined with data on the current medicinal herb variety, soil type, and plant growth stage, the system compares the dosage, time, and other parameters in the input usage records for compliance. If the standards are met, the system marks the input as compliant and records it on the blockchain. If the input violates the standards, the system issues a real-time warning and prohibits the data from being uploaded to the blockchain, requiring correction and re-verification. The verification results serve as the core basis for the quality credit score in the planting process.

[0122] The processing stage traceability process in S5 includes: Harvest Confirmation: At the late stage of the growth of Chinese medicinal materials, based on the constructed dynamic digital model, combined with the collected plant physiological data, surface spectral data and growth modeling results, the optimal harvest time is predicted through machine learning algorithms. During actual harvesting, staff use portable near-infrared detectors to conduct rapid on-site detection of key components of the medicinal materials. If the deviation between the detection data and the predicted value of the digital model is >5%, the system automatically triggers a second re-inspection process. The second re-inspection is carried out using high-precision laboratory detection equipment for verification, and the two detection data and analysis process are recorded on the blockchain to ensure the scientificity and accuracy of the harvest time decision. Process control: Establish a digital twin model of the processing equipment. By deploying sensors on equipment such as the drying room and slicing machine, real-time data collection of equipment operation can be achieved. Multiple temperature sensors can be installed at different locations in the drying room. The collected data is transmitted to the cloud through the edge intelligent processing layer. A three-dimensional temperature field distribution is constructed in the digital twin model. Managers can intuitively view whether the temperature in the drying room is uniform and whether there are local overheated or undercooled areas through the interface of the application service layer. Pressure sensors and vibration sensors installed on the slicing machine blades can also monitor the stress and vibration frequency of the blades in real time to determine the wear status of the blades. When the wear of the blades reaches the preset threshold, the system automatically issues a replacement prompt and records the information on the blockchain. During the on-chain processing of processing parameters, a dynamic blockchain sharding algorithm is used. Drying temperature data, which is directly related to quality, is allocated to independent high-security shards for storage, while runtime data related to equipment maintenance is allocated to low-sensitivity shards. At the same time, a sub-chain is generated in the processing stage to link procedures, equipment, and personnel, clearly defining the operators, equipment used, and specific parameters for each processing step. This ensures that process data is tamper-proof and can be shared as needed. Regulatory authorities can quickly obtain complete processing parameters for specific batches of medicinal materials through authorized access, achieving transparent traceability of the processing process.

[0123] S6. Traceability Inquiry and Credit Evaluation: Through the three-dimensional semantic traceability interface, the system calls the three-dimensional growth model of a single medicinal plant in S3, the multi-dimensional detection data in S2, and the traceability process data in S5 stored on the cloud digital twin platform to provide users with full-cycle visualized traceability services. Based on the quality credit evaluation system, the system integrates the abnormal growth warning in S3, the traceability strategy execution results in S5, and data security storage information to generate a dynamic credit score. The score results are fed back to the cloud smart contract for adjusting the accuracy of the traceability strategy. S7. Data Security Processing: Establish a three-level data encryption mechanism based on different data types. Homomorphic encryption technology is used to process core privacy layer data such as the coordinates of farmers' plots and input formulas; Zero-knowledge proof technology is used to verify sensitive business layer data such as parameters of pest and disease identification models and operational data of processing equipment. For publicly displayed data such as the growth cycle of medicinal materials and the summary of test reports, a symmetric encryption algorithm based on timestamps is used. Meanwhile, a blockchain dynamic sharding algorithm is designed to automatically adjust the sharding strategy based on two different traceability scenarios and two different data types, namely physiological data or equipment parameters, at the planting or processing stage. In the processing stage, a sub-chain of processes, equipment, and personnel is generated, and in the circulation stage, dynamic shards of batches, logistics, and warehousing are generated. An improved PBFT consensus algorithm is used to shorten the block confirmation time to less than 200ms, ensuring that the data is tamper-proof and shared on demand throughout the entire cycle.

[0124] Specifically, the method for digital traceability of the entire lifecycle of medicinal herb cultivation quality includes the following steps: S1. Construct a four-layer architecture system comprising an intelligent sensing terminal layer, an edge intelligent processing layer, a cloud-based digital twin platform, and an application service layer: Intelligent sensing terminal layer: Deploy a multimodal sensor array in the medicinal herb planting area, specifically including: Implantable physiological sensors: A stem micro-deformation sensor made using flexible electronics technology with an accuracy of ±0.01mm and a sampling frequency of 1 time / 10 minutes; a liquid flow rate monitoring probe with an accuracy of ±0.1mL / h; and a leaf surface temperature sensor with an accuracy of ±0.5℃. These sensors are fixed to the plant stem or leaves through minimally invasive implantation to monitor 12 physiological indicators in real time.

[0125] Surface spectral acquisition module: A drone equipped with a 200-2500nm band spectrometer and a fixed monitoring station. The drone flies at an altitude of 5-10m and collects data daily from 9:00 to 11:00. The fixed monitoring station is deployed at 50m intervals and collects data once per hour to obtain reflectance data with a resolution of 10nm.

[0126] Underground root microenvironment probe: The embedded sensor can be buried at a depth of 10-30cm. It integrates a soil urease activity sensor, a phosphatase activity sensor, and a microbial community conductivity probe. The soil urease activity sensor has a detection range of 0-100U / g and an accuracy of ±5%; the phosphatase activity sensor has a detection range of 0-50U / g and an accuracy of ±5%; and the microbial community conductivity probe has an accuracy of ±1mS / cm. It automatically collects rhizosphere soil data weekly.

[0127] The biodegradable tag uses paper-based NFC chip encapsulation technology. The chip substrate is a starch-based biodegradable material with a degradation cycle of 90-120 days. It integrates an SHT30 temperature and humidity sensor and an ambient light sensor. The SHT30 temperature and humidity sensor has an accuracy of ±2%RH and ±0.3℃, while the ambient light sensor has an accuracy of ±10 lux. After data collection is completed during the harvest period, the tag detaches from the plant and enters the soil environment. Soil moisture and microorganisms become the degradation triggering factors. Soil microorganisms, such as bacteria and fungi, secrete extracellular enzymes, such as amylase and esterase. These enzymes can specifically act on the molecular chains of starch-based materials, breaking chemical bonds and causing macromolecules to gradually decompose into smaller molecules. Over time, the material structure is gradually destroyed, degrading layer by layer from the surface to the inside and releasing organic fertilizer containing humic acid (≥5%).

[0128] Edge intelligent processing layer: Deploy a lightweight twin modeling engine and federated learning data gateway at the edge nodes of the planting area.

[0129] The lightweight twin modeling engine is based on an improved U-Net convolutional neural network. It introduces a residual module to reduce computational complexity, compresses the number of model parameters to 12MB, and supports the construction of single-tree models in minutes.

[0130] The Federated Learning Data Gateway employs the Secure Multi-Party Computation (MPC) protocol and Differential Privacy (DP) technology. The secure MPC protocol is similar to the Yao Millionaire protocol, and the differential privacy technology parameters are ε=1.0 and δ=1e-5, enabling secure aggregation of grower data within the region and preventing the leakage of raw data.

[0131] Cloud-based digital twin platform: Built on Alibaba Cloud IoT platform, including: Dynamic digital model: Using the Unity3D engine, based on the single plant growth model and environmental data uploaded by the edge intelligent processing layer, the digital model of Angelica growth is rendered in real time. The mapping accuracy of parameters such as plant height and stem diameter reaches 95%, with an error of less than 5%.

[0132] The smart contract gradient triggering mechanism sets three quality risk levels: green, yellow, and red. A weighted scoring algorithm integrates growth anomaly reports generated by a lightweight twin modeling engine, pest and disease detection results output by an AI recognition model, and environmental and physiological data collected by multimodal sensors. For example, pest and disease detection results are weighted at 40%, plant physiological anomaly data at 30%, and environmental data at 30%. A comprehensive score below 60 is considered green, 60-80 at yellow, and above 80 at red. A red alert triggers the full-node notarization function with a three-level data encryption mechanism; green alerts utilize lightweight node rapid consensus.

[0133] A three-tiered data encryption mechanism is established. The core privacy layer, including farmer plot coordinates and input formulas, employs homomorphic encryption. The business-sensitive layer, such as pest and disease identification model parameters and processing equipment operation data, uses zero-knowledge proof technology. The public display layer, including medicinal herb growth cycles and test report summaries, uses a timestamp-based symmetric encryption algorithm. Simultaneously, a dynamic blockchain sharding algorithm is designed, adjusting the sharding strategy based on planting, processing stages, and data types. An improved PBFT consensus algorithm is used to shorten block confirmation time to less than 200ms.

[0134] Lightweight nodes enable rapid consensus: Lightweight nodes do not store complete blockchain data, but only synchronize block headers. When data in a green scenario needs to be verified, lightweight nodes only need to request verification of the data's authenticity from nearby full nodes, without participating in the synchronization and calculation of the complete ledger. Through simplified consensus logic, data on-chain confirmation is quickly completed.

[0135] The blockchain dynamic sharding algorithm is based on the Ethereum 2.0 framework. It dynamically adjusts the sharding strategy according to the data type and scenario. In the processing stage, it generates a process-equipment-personnel subchain with a shard size of no more than 2MB. In the circulation stage, it generates a batch-logistics-warehousing shard with a shard size of no more than 5MB. It adopts an improved PBFT consensus, reduces one round of communication, and the confirmation time is no more than 200ms.

[0136] Application Service Layer: Develop web and mobile application services. The 3D semantic traceability interface, based on WebGL technology, allows users to input batch numbers and dynamically view a full-cycle twin animation of Angelica sinensis from planting to processing, displaying multi-dimensional testing data for each stage. The quality credit evaluation system integrates abnormal growth warnings, traceability strategy execution results, and data security storage information. It uses the analytic hierarchy process (AHP) to determine indicator weights and generates dynamic credit scores in real time.

[0137] S2. Multi-dimensional data acquisition and processing: The multi-dimensional data includes physiological data, spectral data, root system data, and label data, and the data collection process for each data is as follows: Physiological data: The implanted sensor collects data such as stem micro-deformation and sap flow rate at 10-minute intervals, transmits the data via the LoRaWAN protocol at a frequency of 868MHz and a transmission rate of 250bps, uploads it to the edge node, and stores it in the InfluxDB time series database for a retention period of 3 years.

[0138] Spectral data: Data in the 200-2500nm band collected by UAVs and fixed monitoring stations were filtered and denoised using Savitzky-Golay filtering. Seven characteristic bands, such as 450nm, 650nm, and 800nm ​​reflectance, were extracted to generate spectral feature vectors.

[0139] Root data: The underground probe automatically collects rhizosphere soil enzyme activity and electrical conductivity data weekly, which are transmitted to the edge node via RS485 interface and fused with soil pH and nutrient data to generate a root microenvironment report. Soil pH and nutrient data are from third-party testing.

[0140] Tag data: The biodegradable tags synchronize temperature and humidity data to the handheld terminal daily via the NFC interface. The handheld terminal is an Android system with an integrated NFC read / write module. One day before harvesting, the full-cycle data is uploaded to the edge node via Bluetooth BLE5.0.

[0141] S3, Edge Growth Modeling and Early Warning: The specific steps for constructing a 3D growth model of a single medicinal plant at edge nodes using a lightweight twin modeling engine are as follows: Data preprocessing: The data from the implanted sensor were normalized to a range of 0-1. Vegetation indices such as NDVI and PRI were extracted from the spectral data. The root data were normalized and used as the model input.

[0142] Model Construction: The improved U-Net network consists of 5 downsampling layers and 5 upsampling layers. Each layer uses a 3×3 convolutional kernel and the activation function is LeakyReLU. The output layer generates a single-tree point cloud model through bilinear interpolation, with a point cloud density of no less than 1000 points / m².

[0143] The abnormal growth early warning process is as follows: S31. Collect 12 physiological indicators of 8 kinds of medicinal materials such as ginseng and angelica, and construct a time series dataset containing 100,000 samples, labeled with 12 abnormal states such as continuous cropping obstacles and fertilizer damage. S32. Develop an attention-based LSTM network model to enhance sensitivity to key physiological indicators through the attention mechanism, enabling early identification of 12 abnormal states such as continuous cropping obstacles, fertilizer damage, and drought. The input layer consists of a 12-dimensional feature vector, the hidden layer has 128 neurons, the attention mechanism weight matrix has a dimension of 12×128, the early warning time is no less than 72 hours, and the accuracy is no less than 92%. S33. Design a BP neural network coupling model. The BP neural network has an 18-dimensional input layer, including 12 physiological indicators and 6 environmental indicators. It has 2 hidden layers, each with 64 neurons. The output layer is the abnormal probability value. The dynamic threshold is automatically adjusted by ±15% according to the growth cycle, such as the seedling stage, flowering stage, and fruiting stage.

[0144] When the leaf yellowing degree exceeds 30%, the edge node sends a command to the irrigation system via the MQTT protocol, with QoS set to 1, to control the solenoid valve to open local drip irrigation at a flow rate of 5 L / h for 30 minutes.

[0145] S4. Federated Learning Model Updates and Pest Control: The federated learning process is as follows: Data preprocessing: The pest and disease images of the grower edge nodes are processed. The image resolution is no less than 512×512. The images are grayscaled and standardized. A 64-dimensional feature vector is extracted using the YOLOv5s model, which includes 7 common pest and disease categories.

[0146] Secure aggregation: Each node shares feature vectors through the secure multi-party computation (MPC) protocol. The server uses the FedAvg algorithm to aggregate model parameters. The global model update cycle is 7 days. Differential privacy technology adds Laplacian noise with a scale parameter of 0.5.

[0147] Model Deployment: The updated model has 7.3MB of parameters and is distributed to each edge node. The accuracy of identifying diseases such as root rot of Angelica sinensis and rust of ginseng has been improved to 98.7%, the false alarm rate is no more than 1.2%, and the inference time is no more than 200ms / image.

[0148] The federated learning data gateway aggregates pest and disease image data from growers within the region. Each grower's edge node performs local preprocessing of the pest and disease images, and then transmits the feature vectors to the federated learning server without disclosing the original images using secure multi-party computation technology. The server periodically updates the AI ​​recognition model using the FedAvg algorithm. The updated model is then protected against noise using differential privacy technology before being redistributed to each edge node. Testing showed that the updated model improved the accuracy of identifying common Chinese medicinal herb pests and diseases to 98.7%, while reducing the false positive rate to 1.2%.

[0149] S5, Digital Twin Simulation and Traceability Strategy Execution: A dynamic digital model is used to simulate all elements of the growth cycle, including predicting the impact of environmental factors such as light, temperature, moisture, and nutrients on the growth of medicinal herbs. Self-optimization methods for traceability accuracy include: S51. Establish a twin error correction mechanism and design a state estimator based on extended Kalman filter to calibrate the deviation of growth parameters between the digital model and the physical plant in real time, such as plant height and stem diameter deviation <5%. The extended Kalman filter (EKF) state estimator calibrates the twin parameters in real time. The state transition matrix F is 12×12-dimensional, the observation matrix H is 9×12-dimensional, and the covariance matrix is ​​updated once per minute. After calibration, the plant height error is less than 2cm and the stem diameter error is less than 0.5mm.

[0150] S52. Develop a twin version management system that automatically saves daily twin model snapshots and supports backtracking queries of historical growth status: The twin version management system automatically saves model snapshots at 00:00 every day in .glb format, with a file size not exceeding 500KB, and supports backtracking of historical status for the most recent 30 days.

[0151] S53. Define a three-dimensional accuracy model and design an intelligent adjustment algorithm to automatically match the accuracy level according to the type of traceability request: The three-dimensional accuracy model is defined as time resolution, spatial resolution, and parameter dimensions. Time resolution includes minute-level, hour-level, and day-level; spatial resolution includes single plant, row, and plot; and parameter dimensions include physiological, environmental, and spectral. The intelligent adjustment algorithm automatically switches model parameters according to the type of traceability request, such as minute-level single plant accuracy for regulatory queries and day-level plot accuracy for enterprise self-inspections.

[0152] S54. In the above steps, the intelligent adjustment algorithm automatically matches the accuracy level according to the traceability request type. When a user initiates a traceability request through the 3D semantic traceability interface, the intelligent adjustment algorithm first parses the request type, and then automatically matches the corresponding accuracy level based on the request type and the 3D accuracy model. If it is a single plant growth process traceability, a model with high temporal resolution and single plant spatial resolution is selected first. If it is a plot overall quality traceability, a model with plot spatial resolution and appropriate temporal resolution is selected. Then, the algorithm retrieves the model data of the corresponding accuracy from the twin version management system, and quickly generates a traceability result that meets the user's needs based on the latest parameters after calibration according to the twin error correction mechanism. This ensures the accuracy of traceability while improving system response efficiency and resource utilization.

[0153] The rules for implementing the tracing strategy are as follows: The planting stage traceability process is as follows: drones equipped with multispectral cameras acquire images of the plots, and combine them with soil probe data to generate a three-dimensional geological model. Smart terminals automatically assign digital identities to the plots and upload them to the blockchain to generate genesis blocks. Edge nodes generate real-time reports of abnormal growth of individual plants and update the pest and disease identification model through federated learning. Digital models of inputs are established to simulate the residual degradation curves of pesticides and fertilizers, and smart contracts automatically verify whether the dosage and interval of inputs are compliant.

[0154] Processing stage traceability process: Based on the digital model, the optimal harvesting time is predicted, and the content of effective ingredients is quickly detected on-site by a portable near-infrared detector. When the deviation between the detection data and the twin prediction value is >5%, a second re-inspection process is triggered. A digital twin model of the processing equipment is established, and the processing parameters are uploaded to the blockchain in real time. The blockchain dynamic sharding algorithm is used to generate sub-chains related to processes, equipment and personnel to ensure that the process data is tamper-proof and can be shared as needed.

[0155] S6. Traceability Inquiry and Credit Evaluation: The system provides interactive query functionality through a 3D semantic traceability interface: after the user enters the batch number of the medicinal materials, the system dynamically renders a twin animation of the entire lifecycle from planting to processing, displaying multi-dimensional testing data at key nodes, such as the heavy metal content of the soil in the planting area and the pesticide residue test results in the processing stage. The quality credit evaluation system uses the analytic hierarchy process (AHP) to determine the weights of indicators, including 40% for planting compliance, 30% for testing pass rate, 20% for traceability response time, and 10% for user feedback. It generates and updates credit scores in real time, and the scores are linked to blockchain addresses to form an immutable credit profile.

[0156] S7. Data security processing: A three-tiered data encryption mechanism is established to encrypt core privacy layer data, business-sensitive layer data, and publicly displayed layer data. Specific application scenarios for this three-tiered data encryption mechanism are as follows: Core privacy layer data, such as grower identity information and formula process data, are encrypted using homomorphic encryption technology, supporting data analysis in encrypted form; Business-sensitive data, such as pest and disease detection results and input usage records, can be verified using zero-knowledge proof technology without disclosing specific content when verifying data compliance. Publicly displayed data, such as photos of medicinal herb growth and logistics information, can use the AES-256 symmetric encryption algorithm based on timestamps, with the encryption key automatically updated every 24 hours.

[0157] The blockchain dynamic sharding algorithm is implemented as follows: In the processing stage, independent sub-chains are generated according to the process type, such as drying, slicing, and packaging. Each sub-chain contains data such as the corresponding equipment ID, operator employee number, and process parameters. In the circulation stage, dynamic shards are generated according to the batch number, logistics order number, and warehousing node. An improved PBFT consensus algorithm is adopted to shorten the block confirmation time to 180ms. At the same time, the system throughput is improved through a dynamic adjustment mechanism of shard weight.

[0158] Figure 4 This is a schematic diagram of the anomaly tracing device provided in the embodiments of this application, as shown below. Figure 4 As shown, a second aspect of this application provides an anomaly tracing device 10, the device 10 comprising: The identification module 101 is used to identify abnormalities in the growth status of the target Chinese medicinal plant based on plant physiological data related to the growth status of the target Chinese medicinal plant, and to obtain abnormal growth information of the target Chinese medicinal plant. The evaluation module 102 is used to simulate and predict the impact of each piece of information in the target information on the future quality problems of the target medicinal plant based on the digital model of the target medicinal plant, and obtain the evaluation value corresponding to each piece of information; wherein, the digital model is used to simulate based on data related to the growth of the target medicinal plant, and the target information includes growth abnormality information and multi-dimensional data related to the growth process of the target medicinal plant. The determination module 103 is used to determine the quality risk level of the target Chinese medicinal plant based on the evaluation values ​​corresponding to each piece of information; wherein, the quality risk level is used to characterize the degree of risk of the target Chinese medicinal plant having quality problems in the future; The traceability module 104 is used to obtain abnormal traceability data of the target Chinese medicinal plant from multi-dimensional data according to the traceability strategy of the planting stage corresponding to the quality risk level; wherein, the abnormal traceability data is the relevant data that causes growth abnormalities in the growth process of the target Chinese medicinal plant.

[0159] The anomaly tracing device 10 provided in the second aspect of this application can implement the various processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0160] Please see Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. A third aspect of this application provides an electronic device 500, including a processor 510 and a memory 520. The memory 520 stores machine-executable instructions that can be executed by the processor 510. The processor 510 can execute the machine-executable instructions to implement the above-mentioned anomaly tracing method.

[0161] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to implement the above-described anomaly tracing method.

[0162] In one embodiment of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the exception tracing method according to the above embodiments.

[0163] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0164] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0165] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0166] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0167] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0168] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0169] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0170] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. An anomaly tracing method, characterized in that, The method includes: Based on plant physiological data related to the growth status of the target Chinese medicinal plant, anomalies in the growth status of the target Chinese medicinal plant are identified to obtain abnormal growth information of the target Chinese medicinal plant. Based on the digital model of the target Chinese medicinal plant, the influence of each piece of information in the target information on the future quality problems of the target Chinese medicinal plant is simulated and predicted to obtain the evaluation value corresponding to each piece of information; wherein, the digital model is used to simulate based on data related to the growth of the target Chinese medicinal plant, and the target information includes the growth abnormality information and multi-dimensional data related to the growth process of the target Chinese medicinal plant; Based on the evaluation values ​​corresponding to each piece of information, the quality risk level of the target Chinese medicinal plant is determined; wherein, the quality risk level is used to characterize the degree of risk that the target Chinese medicinal plant will have quality problems in the future; Using a preset planting stage traceability strategy, based on the quality risk level, abnormal traceability data of the target Chinese medicinal plant is obtained from the multi-dimensional data; wherein, the abnormal traceability data is relevant data that causes growth abnormalities during the growth process of the target Chinese medicinal plant.

2. The method according to claim 1, characterized in that, The plant physiological data includes multiple physiological indicators and growth environment data; The method involves identifying anomalies in the growth status of the target medicinal plant based on plant physiological data related to its growth state, thereby obtaining abnormal growth information of the target medicinal plant, including: The long short-term memory network with attention mechanism of the preset coupling model is used to extract features and splice and fuse features from the multiple physiological index data and the growth environment data to obtain feature vectors; The feature vector is input into the backpropagation neural network of the coupled model to obtain the growth anomaly information output by the backpropagation neural network.

3. The method according to claim 1, characterized in that, The target information also includes the results of pest and disease detection of the target Chinese medicinal plant; the results of pest and disease detection of the target Chinese medicinal plant are obtained by the pest and disease identification model from the image data of the target Chinese medicinal plant. Before simulating the impact of each piece of information in the target information on the future quality problems of the target medicinal plant based on the digital model, and obtaining the evaluation value corresponding to each piece of information, the method further includes: Acquire image data of medicinal herb plants from multiple growers; Using secure multi-party computation technology, aggregated data is obtained based on the image data of medicinal plants from the multiple growers. Differential privacy technology is used to add noise to the aggregated data to obtain the target aggregated data; Based on the target aggregated data, the preset initial identification model is processed to obtain the pest and disease identification model.

4. The method according to claim 1, characterized in that, The step of determining the quality risk level of the target medicinal plant based on the evaluation values ​​corresponding to each piece of information includes: The weighted summation of the evaluation values ​​corresponding to each piece of information yields the total target evaluation value of the target Chinese medicinal plant. In the correspondence between the total assessment value and the quality risk level, the quality risk level corresponding to the target total assessment value is determined as the quality risk level of the target Chinese medicinal plant.

5. The method according to claim 1, characterized in that, The method includes: The first growth parameter of the target Chinese medicinal plant is estimated by using the extended Kalman filter state estimator based on the plant physiological data and the model data of the three-dimensional growth model of the target Chinese medicinal plant. If the difference between the first growth parameter and the second growth parameter of the target medicinal plant predicted by the digital model is greater than a preset parameter threshold, the parameters in the digital model are adjusted.

6. The method according to claim 1, characterized in that, The planting stage traceability strategy includes: In the correspondence between risk levels and traceability request types, the traceability request type corresponding to the quality risk level is determined as the target traceability request type; the target traceability request type represents the data type being traced. The target 3D precision model corresponding to the target tracing request type is determined from multiple 3D precision models; wherein the precision of each 3D precision model is different. Obtain target data that matches the accuracy of the target three-dimensional accuracy model and is related to the abnormal growth of the target Chinese medicinal plant. The target data is simulated in the target three-dimensional precision model to obtain anomaly tracing data.

7. The method according to claim 1, characterized in that, The method further includes: During the processing of the target Chinese medicinal plant after harvesting, based on the digital model of the target Chinese medicinal plant, the degree of influence of each data in the processing data on the future quality problems of the target Chinese medicinal plant is simulated and predicted, and the evaluation value corresponding to each data is obtained; wherein, the processing data refers to the parameter data related to the processing process of the target Chinese medicinal plant. Based on the evaluation values ​​corresponding to the data, the processing quality risk level of the target Chinese medicinal plant is determined; wherein, the processing quality risk level is used to characterize the degree of risk of quality problems occurring in the target Chinese medicinal plant after processing; Using a preset processing stage traceability strategy, abnormal traceability processing data of the target Chinese medicinal plant is obtained from the processing data based on the processing quality risk level.

8. An anomaly tracing device, characterized in that, The device includes: The identification module is used to identify abnormalities in the growth status of the target Chinese medicinal plant based on plant physiological data related to the growth status of the target Chinese medicinal plant, and to obtain abnormal growth information of the target Chinese medicinal plant. An evaluation module is used to simulate and predict the impact of each piece of information in the target information on the future quality problems of the target medicinal plant based on a digital model of the target medicinal plant, and to obtain the evaluation value corresponding to each piece of information; wherein, the digital model is used to simulate based on data related to the growth of the target medicinal plant, and the target information includes the growth abnormality information and multi-dimensional data related to the growth process of the target medicinal plant; The determination module is used to determine the quality risk level of the target Chinese medicinal plant based on the evaluation values ​​corresponding to each piece of information; wherein, the quality risk level is used to characterize the degree of risk that the target Chinese medicinal plant will have quality problems in the future; The traceability module is used to acquire abnormal traceability data of the target Chinese medicinal plant from the multi-dimensional data according to the quality risk level, based on a preset planting stage traceability strategy; wherein, the abnormal traceability data is relevant data that causes growth abnormalities during the growth process of the target Chinese medicinal plant.

9. An electronic device, characterized in that, The device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the exception tracing method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the anomaly tracing method as described in any one of claims 1-7.

Images